Automatic Optimum Phase Selection In Cardiac CT Imaging. Master Thesis Fahmi Noor

Transcription

1 Automatic Optimum Phase Selection In Cardiac CT Imaging Master Thesis Fahmi Noor Place Siemens Medical Germany CTE Physical Application Siemensstrasse 1, Forchheim Supervisor Dr. Herbert Bruder Referee Prof. Dr. Dieter Höpfel 2nd Referee Dr. Rainer Stotzka Date

2 Abstract In cardiac CT imaging, an optimal phase determination for reconstruction is a prerequisite for good image quality. The aim of this work is to develop an alternative gating strategy in optimizing cardiac image reconstruction by automatically defining the optimum gating phase of the heart indepently from patient-patient and cycle-cycle variability. Automatic optimum phase detection introduced by Phillips Research laboratories is reproduced for this purpose. The method successfully shows motion pattern of the heart, and delivers the systole-diastole phase directly from a motion map. However in coronary CT angiography the domain of Cardiac CT imaging, the method is not accurate enough. Motion of the arteries are overshadowed by the chamber motion, thus the real motion pattern of arteries are not presented well. Defining subset voxel containing mainly the coronary arteries will expand the functionality of the algorithm. Instead of the whole axial slices, the motion map is derived from the subset voxels. Additional histogram HU weighting function is also used in order to focus the motion calculation on pixels containing contrast media. The algorithm is validated with three patient data from University of Tübingen measured with a 16 slices scanner. The right coronary artery is analyzed and the cycle-depent local phases are determined. Comparing image reconstructed with the proposed algorithm and images reconstructed with conventional method, shows the improvement in the image quality and demonstrates the benefit of the automatic method. The coronary artery is shown with high contrast and in good continuity with reduced motion artifact. Nevertheless, its limitation to eliminate residual step artifact is one important topic, which needs to be investigated in the further work. Keyword: Cardiac CT Imaging, gating strategies, Motion Map, optimum phase selection

3 Acknowledgements In the Name of Allah, the Most Gracious, the Most Merciful. First and foremost, I would like to thank Dr. Herbert Bruder for the chance he gave to me for doing my master thesis under his supervising. And also for his support in my work, for the time he spent with me, for all discussion and help he provided me, and also for a lot of fresh idea he came with in solving the problems. I learn so much from him and I thank him a lot for that. In particular I would like to thank Mr. Thomas Flohr, who allowed me to do my master thesis in his Department at Siemens Medical. I would also like to thank all the CTE-PA employees for their friliness and their support. In spite of my lack of German language, they always show a lot of patience and take time to talk to me, and answer my numerous questions. Thanks to them, the adaptation to a working culture was not too difficult. In this chance, I would also like to thank Prof. Dieter Höpfel who supported my work and help me finishing my thesis, evaluate and examine my work, and also for his visit to the factory even it takes a whole day to get to the factory from his place. Last, I would like here to mention the most important people in my life: my Papa and Mama, my princess Loly and all of my brothers, people who always support me no matter what, for always pray for me and for always giving me happiness. I hope this report is useful for everyone who would like to read it.

4

5

6 Chapter 1. Introduction 1 Introduction 1.1 Motivation Cardiac disease and in particular coronary artery disease (CAD) are still the leading causes of death in Europe and the US. In 1998 about one in five deaths in Europe and the United States was related to cardiac disease. Approximately, 600,000 CAD-related deaths were reported every year in Europe, nearly 500,000 reported in USA. More than 2.5 million such investigations are performed every year in Europe and the USA. These data show high need and importance of reliable non-invasive imaging for early and preventive diagnosis of CAD and other cardiac disease [1]. The goal diagnosis with non-invasive cardiac CT imaging is to identify the location and degree of stenosis in CAD that may be hemodynamically relevant in terms of perfusion and viability of myocardium, cardiac function, and coronary blood flow [1], [14]. Beside the CAD, other cardiac structures such as the cardiac chambers, valves, myocardium, pericardium, are important targets for diagnostic of cardiac disease that may cause reduced or even lack of cardiac functionality. Since its introduction in 1972, X-Ray computed tomography has become a robust and frequently used non-invasive imaging modality for vascular diagnosis [1]. However at that time image of the heart and coronary is very limited in spatial and temporal resolution. It is difficult to reconstruct images of the heart because of its continuous motion. With standard image reconstruction methods, the motion causes artifacts in the image, which is of limited use in diagnostic imaging. Nowadays, Cardiac CT imaging has become more feasible with the availability of multi-row scanners [1], [11], [13]. Multiple detectors provide reliable way to obtain non-invasive coronary angiograms. In combination with ultra-fast rotating gantries, and latest UFC detectors images at good temporal and high spatial resolution can be reconstructed. Siemens itself has produced such a modality for this purpose. With the SOMATOM Sensation 16, and the latest Sensation Cardiac 64, dedicated scan and reconstruction technique have been newly developed that allow virtually motion-free 3D and also 4D imaging of the heart, especially the coronary vessel tree that can be visualized at high level detail. 1.2 Background The domain of cardiac CT is coronary angiography. It needs stable phase of heart with minimum motion in order to reconstruct the coronary arteries sharply delineated. Basically ECG information is used to synchronize the reconstruction with the motion-state of the heart and the data acquisition. The problems come when considering the electrocardiogram cycle-to-cycle variability that causes the optimal phase to vary strongly for each cycle. Even 1

7 Chapter 1. Introduction normal people with regular heartbeat also have differentiation between cycles in their cardiac signal. And it is getting worse for patients with an arrhythmia case whose heartbeat is irregular and changes rapidly. There is also a consideration about patient-to-patient variability in defining the stable phase. Deping on how strong patient s cardiac muscle, heart rate, and other personal characteristics may result in a different stable phase for different patients. Thus, the stable phase may vary strongly from patient to patient [2]. Furthermore, each anatomy of interest, e.g. every branch of the coronary arteries, may be reconstructed better at different cardiac phase, for example coronary artery, which lies in the surface of the cardiac chamber. Dedicated cardiac CT acquisition is needed due to the presence of extensive motion. Currently clinicians have to choose the reconstruction phase, usually between 50%-65% of RR-peaks of the ECG signal [1], [2]. With this manual determination of the stable cardiac phase, several high-resolution data sets must be reconstructed, which is inefficient and a time-consuming task. Sometimes several images reconstructions with incremented cardiac phase have to be made. Therefore, an alternative strategy in determining the optimum reconstruction phase is needed, indepently from patient-patient and cycle-cycle variability and that also can be applied to small structures of the heart, especially the coronary arteries. Several techniques have already been introduced to improve the image quality in cardiac CT. Including the knowledge-based methods involving well-known physiological information of the heart by using model-based reconstruction approaches. In contrast to using the ECG waveform data, it may also be possible to extract motion information from the acquired projection data [3] (known as Kymogram). Other modalities, e.g. phonocardiogram and Doppler ultrasound may be taken as an additional source to obtain more information about the patient-specific heart motion [2]. 1.3 Aims of the Work The aim of this thesis is to develop an alternative strategy in optimizing the reconstruction technique in cardiac CT by automatically defines the stable phase of the heart indepently from patient-patient and cycle-cycle variability. A simple and efficient image-based technique is introduced which is able to deliver patient-specific stable cardiac phases in an automatic way. Here also introduced a technique that uniquely analysed specific structure of the heart coronary arteries into the closer look with the same method. In summary the step of the work is as follow: 1. Reproduce the automatic patient-specific optimum phase selection using the motion map introduced by Phillips Research Laboratories. 2. Expand the functionality of the algorithm for specific voxel-subset of the heart, in this case right coronary artery - RCA. 3. Validate with patient data 4. Analysis and summary 2

8 Chapter 1. Introduction 1.4 Thesis Overview This report consists of 7 chapters with three additional appices information. The next chapter, chapter 2 discusses the basic anatomy of the heart, and also explains about the ECG signal, its derivation and basic interpretation of the signal form. The heart phase is also discussed to get the basic understanding about the motion pattern of the heart. Chapter 3 presents an overview of CT data acquisition and image reconstruction, setup of a CT system, and the basic scan method used nowadays. The specific cardiac CT Imaging technique is discussed in chapter 4, including the explanation of triggering and gating strategies as basic workflow in developing the algorithm in this work. Chapter 5 discusses about the presented method, and explains about the basic idea, and the algorithm used. The experiment and result data is shown in chapter 6. The last, Chapter 7 provides the conclusion of the work, the discussion and an outlook to future work in this field. 3

9 Chapter 2. Anatomy of Heart 2 Anatomy of Heart The Heart is the most important organ in the human body and the motor of the blood circulation system. A brief introduction anatomy of the heart and the coronary arteries will be given below, including ECG-signal derivation and heart phase. 2.1 Heart anatomy 1 The heart is located between the lungs, behind and slightly to the left of the breastbone [4] (see fig. 2.1). As part of the circulation system, the heart constantly pumps blood throughout the body. The heart has four separate compartments or chambers (two atriums and two ventricles). The atriums receive and collect the blood coming to the heart. The atriums Fig 2.1 Location of Heart then deliver blood to the ventricles, which pump blood away from the heart through powerful, rhythmic contractions. The right parts of the heart contain always the de-oxygenated blood from the whole of the body, and Left parts always have fresh blood contain oxygen from the lungs. The functions of the different heart parts are here described more detail. Right atrium: It receives de-oxygenated blood from the body through the superior vena cava (head and upper body) and inferior vena cava (legs and lower torso). The tricuspid valve, which separates the right atrium from the right ventricle, opens to allow the de-oxygenated blood collected in the right atrium to flow into the right ventricle. Right ventricle: It receives de-oxygenated blood as the right atrium contracts. The pulmonary valve leading into the pulmonary artery is closed, allowing the ventricle to fill with blood. Once the ventricles are full, they contract. As the right ventricle contracts, the tricuspid valve closes and Fig 2.2 Blood Flow the pulmonary valve opens. The closure of the tricuspid valve prevents blood from backing into the right atrium and the opening of the pulmonary valve allows the blood to flow into the pulmonary artery toward the lungs. Left atrium: It receives oxygenated blood from the lungs through the pulmonary vein. Then, the blood passes through the mitral valve into the left ventricle. Left ventricle: It receives oxygenated blood as the left atrium contracts. The blood passes through the mitral valve into the right ventricle. The aortic valve leading into the aorta is closed, allowing the ventricle to fill with blood. Once the ventricles are full, they contract. As the left ventricle contracts, the mitral valve closes and the aortic valve opens. The closure of the mitral valve prevents blood from backing into the left atrium and the opening of the aortic valve allows the blood to flow into the aorta and flow throughout the body. That s why the muscle of left ventricles has more thickness so that it can pump the blood away to the whole body. 1 Most of materials used in this section is summarized from [1],[4] including some parts of text and pictures 4

10 Chapter 2. Anatomy of Heart Beside that 4 main chamber of the heart, the rest detail part of heart is also important to ensure the heart works properly. Some of these parts are described as follow: Papillary muscles: They attach to the lower portion of the interior wall of the ventricles. They connect to the chordae tineae, which attach to the tricuspid valve in the right ventricle and the mitral valve in the left ventricle. The contraction of the papillary muscles opens these valves. When the papillary muscles relax, the valves close. Chordae tineae: They are tons linking the papillary muscles to the tricuspid valve in the right ventricle and the mitral valve in the left ventricle. As the papillary muscles contract and relax, the chordae tineae transmit the resulting increase and decrease in tension to the respective valves, causing them to open and close. Tricupsid valve: It separates the right atrium from the right ventricle. It opens to allow the de-oxygenated blood collected in the right atrium to flow into the right ventricle. It closes as the right ventricle contracts, preventing blood from returning to the right atrium; thereby, forcing it to exit through the pulmonary valve into the pulmonary artery. Mitral valve: It separates the left atrium from the left ventricle. It opens to allow the oxygenated blood collected in the left atrium to flow into the left ventricle. It closes as the left ventricle contracts, preventing blood from returning to the left atrium; thereby, forcing it to exit through the aortic valve into the aorta. Pulmonary valve: It separates the right ventricle from the pulmonary artery. As the ventricles contract, it opens to allow the de-oxygenated blood collected in the right ventricle to flow to the lungs. It closes as the ventricles relax, preventing blood from returning to the heart. Aortic valve: It separates the left ventricle from the aorta. As the ventricles contract, it opens to allow the oxygenated blood collected in the left ventricle to flow throughout the body. It closes as the ventricles relax, preventing blood from returning to the heart. Superior vena cava: It is one of the two main veins bringing de-oxygenated blood from the body to the heart. Veins from the head and upper body feed into the superior vena cava, which empties into the right atrium of the heart. Inferior vena cava: It is one of the two main veins bringing de-oxygenated blood from the body to the heart. Veins from the legs and lower torso feed into the inferior vena cava, which empties into the right atrium of the heart. Aorta: It is the largest single blood vessel in the body. This vessel carries oxygen-rich blood from the left ventricle to the various parts of the body. Pulmonary artery: It is the vessel transporting de-oxygenated blood from the right ventricle to the lungs. Pulmonary vein: It is the vessel transporting oxygen-rich blood from the lungs to the left atrium. 2.2 Coronary Arteries As the heart continuously contracts and relaxes, the muscle needs a constant supply of nutrients and oxygen. The coronary arteries are the network of blood vessels that carry oxygen- and nutrient-rich blood to the cardiac muscle tissue. The blood leaving the left ventricle exits through the aorta, the body s main artery. Two coronary arteries, referred to as the "left" (Left Main Trunk - LM) and "right" (Right Coronary Artery- RCA), emerge from the beginning of the aorta, near the top of the heart. 5

11 Chapter 2. Anatomy of Heart The initial segment of the left coronary artery is called the left main coronary. This blood vessel is approximately the width of a soda straw and is less than an inch long. It branches into two slightly smaller arteries: the left anterior descing (LAD) coronary artery and the left circumflex coronary (CX) artery. The left anterior descing coronary artery is embedded in the surface of the front side of the heart. The left circumflex coronary artery circles around the left side of the heart and is embedded in the surface of the back of the heart. Just like branches on a tree, the coronary arteries branch into progressively smaller vessels. The larger vessels travel along the surface of the heart; however, the smaller branches penetrate the heart muscle. The smallest branches, called capillaries, are so narrow that the red blood cells must travel in single file. Considering the smallness of the coronary arteries, it should give us the idea that the reconstruction of images with good quality for the coronary is a great technical challenge [1]. The anatomy of the whole heart, including the placement of the coronary arteries is illustrated in the following figure. Fig.2.3 Anatomy of the heart 6

12 Chapter 2. Anatomy of Heart 2.3 Cardiac electrical activity Electrical conduction system: a brief overview The heart is composed primarily of muscle tissue. A network of nerve fibres coordinates the contraction and relaxation of the cardiac muscle tissue to obtain an efficient, wave-like pumping action of the heart. The sinoatrial node (often called the SA node or sinus node) serves as the natural pacemaker for the heart. Nestled in the upper area of the right atrium, it ss the electrical impulse that triggers each heartbeat. The impulse spreads through the atriums, prompting the cardiac muscle tissue to contract in a coordinated wave-like manner. The impulse that originates from the sinoatrial node strikes the atrioventricular node (or AV node), which is situated in the lower portion of the right atrium. The atrioventricular node in turn ss an impulse through the nerve network to the ventricles, initiating the same wave-like contraction of the ventricles. The electrical network serving the ventricles leaves the atrioventricular node through the right and left bundle Fig.2.4 Electrical Conduction System, (1) SA Node, (2) AV Node, (3) Common AV Bundle, (4) Right and Left Bundle branches branches. These nerve fibres s impulses that cause the cardiac muscle tissue to contract The standard 12-lead ECG The electrocardiogram (ECG) consists of recording the bioelectrical signal of the heart muscles. The heart s electrical activity is recorded from electrodes on the body surface. ECG is a powerful clinical tool for diagnosing cardiac abnormalities. In the standard 12-lead ECG, electrodes have to be placed in a particular spatial orientation [5] as indicated below. Fig.2.5. Standard Limb Leads 7

13 Chapter 2. Anatomy of Heart This diagram illustrates ECG waves and intervals as well as standard time and voltage measures on ECG paper. Fig.2.6 ECG Wave, and its interval The meaning of ECG waves and intervals are here explained [5]. P-wave: It represents the sequential activation (depolarisation) of the right and left atriums. For normal ECG, the P-wave duration is inferior to 120 ms. QRS complex: It represents the right and left ventricular depolarisation. ST - T-wave: This wave represents the ventricular repolarisation. U-wave: This wave is not always observable and its origin is not clear, but it ay represent after the polarisations in the ventricles. PR-interval: This is the time interval from onset of atria depolarisation (Pwave) to onset of ventricular depolarisation (QRS complex). QRS duration: This is the duration of ventricular muscle depolarisation. It is normally inferior to 100 ms. QT-interval: It represents the duration of ventricular depolarisation and repolarisation. RR-interval: It represents the duration of ventricular cardiac cycle. This is an indicator of ventricular rate. For a normal ECG, the heart rate is about bpm PP-interval: It represents the duration of atria cycle. 2.4 Heart phase 2 The right side receives oxygen-poor blood from the various regions of the body and delivers it to the lungs. In the lungs, oxygen is absorbed in the blood. The left side of the heart receives the oxygen-rich blood from the lungs and delivers it to the rest of the body Human heart: two pumps in one The cardiac cycle is divided in two phases, the systole and the diastole. The contraction of the cardiac muscle tissue in the ventricles is called systole. When the ventricles contract, they force the blood from their chambers into the arteries leaving the heart. The left ventricle empties into the aorta and the right ventricle into the pulmonary artery. The relaxation of the cardiac muscle tissue in the ventricles is called diastole. When the ventricles relax, they make room to accept the blood from the atriums. 2 Most of materials used in this section is summarized from [6] including some parts of text and pictures 8

14 Chapter 2. Anatomy of Heart In the following sections, the different cardiac phases are described [6]. The corresponding ECG and some physiological values (aortic, ventricular and atria pressures, and also ventricular volume) are supplied Systole Atrial systole Aortic press Ventricular press Atrial press Ventricular volume Fig.2.7 Atrial Systole Prior to atrial systole, blood has been flowing passively from the atrium into the ventricle through the open atrioventricular valve. During atrial systole the atrium contracts and tops off the volume in the ventricle with only a small amount of blood. Atrial contraction is complete before the ventricle begins to contract. The "a" wave occurs when the atrium contracts, increasing atrial pressure. Blood arriving at the heart cannot enter the atrium so it flows back up the jugular vein, causing the first discernible wave in the jugular venous pulse. Atrial pressure drops when the atriums stop contracting. Concerning the ECG, an impulse arising from the sinoatrial node results in depolarisation and contraction of the atriums (the right atrium contracts slightly before the left atrium). The P-wave is due to this atrial depolarisation. The PR segment is electrically quiet as the depolarisation proceeds to the atrioventricular node. This brief pause before contraction allows the ventricles to fill completely with blood Isovolumetric contraction The atrioventricular valves close at the beginning of this phase. Mechanically, ventricular systole is defined as the interval between the closing of the atrioventricular valves and the opening of the aortic and pulmonary valves. Aortic press Ventricular press Atrial pressure Ventricular volume The atrioventricular valves close when the pressure in the ventricles exceeds the pressure in the atriums. As the ventricles contract isovolumetrically (their Fig.2.8 Isovolumetric Contraction 9

15 Chapter 2. Anatomy of Heart volume does not change) the pressure inside increases, approaching the pressure in the aorta and pulmonary arteries. The electrical impulse propagates from the atrioventicular to allow the ventricles to contract from the apex of the heart towards the coordinate system. The QRS complex is due to ventricular depolarisation, and it marks the beginning of ventricular systole. It is so large that it masks the underlying atrial repolarisation signal Rapid ejection Aortic press Ventricular press Atrial press Ventricular volume At the beginning of this phase the aortic and pulmonary valves open. While the ventricles continue contracting, the pressure in the ventricles exceeds the pressure in the aorta and pulmonary arteries; the aortic and pulmonary valves open, blood exits the ventricles, and the volume in the ventricles decreases rapidly. Fig.2.9 Rapid Ejection As more blood enters the arteries, pressure built until the flow of blood reaches a peak. The "c" wave of atrial pressure is not normally discernible in the jugular venous pulse. Right ventricular contraction pushes the tricuspid valve into the atrium and increases atrial pressure, creating a small wave into the jugular vein. It is normally simultaneous with the carotid pulse Reduced ejection The reduced ejection is the last phase of the systole. At the of this phase the aortic and pulmonary valves close. After the peak in ventricular and arterial pressures), blood flow out of the ventricles decrease and ventricular volume decreases more slowly. Aortic press Ventricular press Atrial press Ventricular volume Fig.2.10 Reduced Ejection When the pressure in the ventricles falls below the pressure in the arteries, blood in the arteries begins to flow back toward the ventricles and causes the aortic and pulmonary valves to close. This marks the of ventricular systole mechanically. The T-wave is due to 10

16 Chapter 2. Anatomy of Heart ventricular repolarisation. The of the T-wave marks the of ventricular systole electrically Diastole Isovolumetric relaxation Isovolumetric relaxation indicates the beginning of the diastole. At the beginning of this phase the atrioventricular valves are closed. Aortic press Ventricular press Atrial press Ventricular volume Throughout this and the previous two phases, the atrium in diastole has been filling with blood on top of the closed atrioventricular valve, causing atrial pressure to rise gradually. Fig.2.11 Isovolumetric Relaxation The "v" wave is due to the back flow of blood after it hits the closed atrioventricular valve. It is the second discernible wave of the jugular venous pulse. The pressure in the ventricles continues to drop. Ventricular volume is at a minimum and is ready to be filled again with blood. As illustrated in the following ECG, no cardiac electrical activity is recorded during the isovolumetric relaxation Rapid ventricular filling Once the atrioventricular valves open, blood that has accumulated in the atriums flows rapidly into the ventricles. Ventricular volume increases rapidly as blood flows from the atriums into the ventricles. Aortic press Ventricular press Atrial press Ventricular volume Fig.2.12 Rapid ventricular filling As well as during the previous phase, no cardiac electrical activity is recorded during the rapid ventricular filling in the ECG 11

17 Chapter 2. Anatomy of Heart Reduced ventricular filling Compared to the previous phase, ventricular volume increases more slowly during the reduced ventricular filling. The ventricles continue to fill with blood until they are nearly full. Aortic press Ventricular press Atrial press Ventricular volume As illustrated in the following ECG, no cardiac electrical activity is recorded during this phase. Fig.2.13 Reduced ventricular filling ECG signal, Heart Phase and Motion pattern of the heart The movement of the chamber during its activity phase determines the motion pattern of heart. The pattern is repeated from cycle to cycle following the activity phase described before. From the explanation above, it is clear to see that the ECG signal represents exactly the electrical activity of the heart, but only corresponds relatively to the motion of the heart. The ECG information does not always represent the heart phase with an adequate accuracy [12]. During the diastole phase, for example no signal recorded related to the heart activity of ventricular filling. Nevertheless, the ECG signal can be used as the hint of determining phase of the heart during its activity. Extracting the R-peaks from the signal would give indication to relative phase at certain heart activity, thus the indication for motion pattern of the heart. 12

18 Chapter 3. CT Overview 3 Computed Tomography Overview In general terms, the principle of computed tomography (CT) consists of measuring the spatial distribution of the attenuation coefficient from the scan object, which is examined from different directions and of computing images from these data. 3.1 Historical overview Computed tomography became a real feasible technique of imaging with the development of the modern computer technology (1970). But the physical principles, which are used for the CT, were discovered at the of the 19 th century and developed during the 20 th century. The main facts concerning the development of computed tomography are here given [16] 1895: W.C. Röntgen discovers a new kind of rays, later referred to as X-rays or Röntgen rays in his honour. 1917: J.H. Radon develops the mathematical foundation for reconstructing cross-sectional images from transmission measurements. 1963: A.M. Cormack describes a technique for calculating the absorption distribution in the human body. 1972: G.N. Hounsfield and J. Ambrose conduct the first clinical CT examinations. 1974: 60 clinical CT installations (head scanners). 1975: First whole body CT scanner in clinical use. 1979: Hounsfileld and Cormack awarded the Nobel Prize. 1989: W.A. Kaler and P. Vock conduct the first examinations with spiral CT. 1994: Introduction of EBCT for cardiac imaging 1998: Introduction of multi-slice detector systems. 2000: Approximately clinical CT installations (whole body scanners). 2002: Multi-slice scanning with 16 slices per rotation 2004: Advanced 64 slices Scanner and dedicated cardiac CT scanner introduced: including Z-sharp technology. 3.2 CT Basic Principle Measurement principle 3 In computed tomography, the primary intensity I 0 of X-rays and the intensity attenuated by the object have to be recorded [7], [8]. Then, they are used to calculate the attenuation value along each ray from source to detector. The integral along the ray path of the attenuation coefficient µ is given by I = I 0 exp( µ ( x, y) dxdy ) (3.1) The reconstructed function I is the distribution of linear attenuation coefficient µ since, for monochromatic X-rays, the logarithm of the transmitted intensity is proportional to the integral of the attenuation coefficient along the path. Theoretically, to compute the exact two-dimensional distribution µ(x, y), an infinite number of line integrals have to be recorded. However, a high finite number of measurements are sufficient to compute an image to a good 3 Most of materials used in this section is summarized from [7], [8] including some parts of text and pictures 13

19 Chapter 3. CT Overview approximation. It is necessary to carry out measurements in all directions, i.e. at least over an angular range of 180, and to determine many narrowly spaced data points for each projection. A simple measurement set up fulfilling this purpose is sketched in the following figure. Fig.3.1 Measuring an Object in CT The main advantage of CT as compared to projection imaging is the ability to separate objects according to their position in the projection direction, i.e. to avoid the confusion that arises when the shadows of multiple objects are superimposed. This, combined with high precision measurements and digital displays, gives CT the ability to resolve objects with extremely small contrast. For example, conventional X-ray CT systems are able to easily distinguish objects whose relative difference in attenuation coefficient is a fraction of one percent. In the 70 s, a radiation source emits a pencil beam and the detector placed opposite registers the intensity, attenuated by the object. For a given angular position, this set up of radiation source and detector is moved linearly. These results in an intensity profile recorded for parallel rays. By determining the ratios of the primary intensity and the attenuated intensities recorded behind the object and taking their logarithms, an attenuation profile results which is generally termed a projection. Projections are measured for successive angular positions. In Fig.3.2 Fan Beam geometry the previous figure, the set of projections is determined in parallel ray geometry over 180. CT scanners today measure typically in fan-beam geometry over an angular range of Image Generation Information on the as yet unknown distribution of the attenuation coefficient µ(x,y) is only given in form of a set of projection values, which is also termed the Random transform of the image. An inverse transformation has to be carried out to determine µ(x,y). In today s CT scanners the so-called convolution-back projection procedure is usually utilised. This is illustrated in the following figure. 14

20 Chapter 3. CT Overview Convolution X-ray 0 projection 1 projection 3 projections No conv The starting point is always a matrix, which contains only zeros. Then, each projection value is added to all the picture elements along the direction in which it has been measured. In general, each detail in the object and represented in the attenuation profile does not only contribute to the pixel value at the desired image point, but to the entire image as well. N projections Image obtained N projections CT value profil Fig.3.3. Image Reconstruction in CT by convolution and backprojection Even when considering only three projections it becomes apparent that an unsharp image will result. To avoid this, each projection has to be convoluted before the backprojection with a convolution kernel. After convolution, the resulting values are added. In essence, this represents a high pass filter procedure, which generates over- and undershoots at object boundaries. Such a method enables to counteract the unsharpening. Convolution additionally offers the possibility to image characteristics by the choice and design of the convolution kernel. A relatively weak high pass filter reduces spatial resolution as well as image noise, whereas a strong high pass filter has the opposite effect. For cardiac imaging for example, soft kernel B30f is used in order to get the optimum spatial resolution. As explained above, CT measures and computes the spatial distribution of the attenuation coefficient µ(x,y). However, the physical quantity µ is not very descriptive and is strongly depent on the spectral energy used. By displayingµ, a direct comparison of images obtained on scanners with different voltages and filtration would be limited. Fig.3.4 HU scale for several organs Therefore the Hounsfield scale (Fig 3.4) is used in CT. In these HU units, CT values characterise the attenuation coefficient of the tissue in each volume element relative to the µ-value of water. µ µ water µ HU = *1000 (3.2) µ water The obtained CT values of different tissues are therefore defined to be relatively stable and to a high degree indepent of the X-ray spectrum. 15

21 Chapter 3. CT Overview Setup of a CT System A CT System comprises several components. These basically include [15]: - The scanning unit, including the gantry with tube and detector system. - The patient table - Image Processor for image reconstruction - The console (represents the man-machine interface) Scanning Unit (Gantry) A CT scanning system consists of an X-Ray unit, which functions as a transmitter, and a data acquisition unit, which function as the receiver. In Commercial CT system these two components are housed in a ring shaped unit called Gantry X-Ray Components (Tube) Manufacturers of CT systems use X-Ray tubes with variable focal spot sizes. Volumes for which good low-contrast resolution is essential need to be scanned with a large focal spot and high power, whereas the high-resolution images with thin slices requires a small focal spot. Tubes used in the CT scanners have a power rating of kw at voltages of kv. The system can however be operated at maximum power for a limited time only. These limits are defined by the properties of the anode and the generator. X-Ray Shielding CT scanner is equipped with grids, collimators and filters to provide shielding against scattered radiation, to define the scan slice and to absorb the low energy portion of the x-ray spectrum. In this way, both patient and examiners are protected. Detector The detector system plays an important role in the interaction of the CT components. It converts the incident x-rays of varying intensity to electric signals. These analog signals are amplified by downstream electronic components and converted to digital pulse. Fig.3.5 inside a CT, For example Sensation Cardiac 64 Scanners. (a) Tube (b) Beam Collimator (c) Generator (d) UFC Detector Picture is taken from [15] Scanner Parameters 16

22 Chapter 3. CT Overview Scanner parameters determine the image quality. Optimal performance of CT systems can be achieved only with the optimal combination of parameters. These parameters differ form one application to another. Set of parameters called the scan protocol. Observing heart in the cardiac application would need different set of parameters compared to the CT application of brain perfusion. Collimation The collimation and together with focal size determine the quality of the slice profile. There are 2 terms of collimator: Source collimator placed directly in front of the radiation source to form maximum required fan beam radiation and determining the dose, and another term Detector Collimator positioned directly in front of the detectors used to shield against scattered radiation and preventing image artifacts. Slice Thickness Images can be reconstructed with slice thickness following the detector collimator. The widest range of possibilities in the selection of collimation and reconstructed slice thickness is only possible in spiral CT using multi detector system. With Multi-slice detector slice thickness equal to or larger than detector collimator can be used. Fig.3.6 Slice Profile in relation with collimation Increment Increment determines the distance between images reconstructed from data volume. If an appropriate increment is used, overlapping images can be reconstructed. A clinical useful overlap is about 30%-50%. Pitch An important parameter for spiral CT scanning is the spiral pitch p. The spiral pitch is defined as the ratio of the table feed per rotation d to the collimated slice-width Scol. This table feed is given in mm per 360 rotation, and for single slice scanners with a rotation time of 1 s the table feed corresponds to the table feed in mm/s. Pitch = table feed per rotation / collimation. The pitch is a dimensionless quantity and is of great importance for image quality and dose considerations. The pitch value should be selected in the range 1 and 2. It should be larger than 1 in order to cover a given scan volume as fast as possible and to reduce the dose compared with sequential CT, and should not exceed 2 to exclude gaps in sampling the object along the z-axis. The term pitch is changing for multi-slice application to adapt the number of slices used in reconstruction. (See chapter 3.4) Rotation Time Rotation Time is time interval needed for a complete rotation of the tubedetector system around the patient. It affects the spiral scan length and thus the coverage of scan range during a certain period of time. Rotation times determine the temporal resolution of the reconstruction and deal with the motion artifacts. This applies especially for instance, to constantly moving organs such as the heart. 17

23 Chapter 3. CT Overview mas The mas value is the product of the tube current (e.g. 200 ma) and the rotation time (e.g. 0.5 s). This selected mas determine the dose used, thus deal with the image noise and detectability of scanner. For example, higher mas values will reduce the noise, and improving the detectability of lower contrast, that is used for visualization of soft tissue. Visualizations of bone or lungs as well as contrast study require lower dose from lower mas. 3.3 Scan modes The routine operation of a CT scanner requires only a few scan modes: taking a survey radiograph for orientation over in the anatomy in question, selecting slices or scan regions and scanning them in the sequential or spiral CT mode[7] Survey Topogram To select the position of single slices or complete scan regions it has proven very helpful to generate a survey radiograph similar to a conventional radiograph. For this purpose the X-ray tube is kept in a fixed angular position and the patient is transported through the field of measurement at low speed, with radiation emitted continuously or in pulsed mode. This topogram is also useful for the dose reduction strategy Sequential CT For more than two decades CT examinations consisted of scanning single slices sequentially. A cross-sectional image is produced by scanning a transverse slice of the body from different angular position while the tube and detector rotate around the patient with the table being stationary. After scanning a single slice, the patient is transported for a defined distance or scans increment, mostly selected equal to the chosen slice thickness. Then the next scan is taken and Fig.3.7 Sequential CT the procedure is repeated. This examination mode, which meanwhile has been largely replaced by spiral CT, is relatively time demanding since time is required for table feed. Examination of complete organs typically takes from five to twenty minutes in a way. Nowadays, sequential scanning mode is still used in application for head scan Volume scanning Spiral CT For sequential CT, two basic requirements hold true which cannot be neglected without negative implications for image quality: the object to be scanned must not move during data acquisition, and the scan geometry must be perfectly planar. Indeed, if the patient moves during the scan, motion artifacts result. In the same way, when the focus and detector do not travel in the same plane, artifacts also arise. However, spiral CT builds precisely upon violating these two principles. As a matter of fact, it no longer requires a planar geometry and it moves the patient during scanning. 18

24 Chapter 3. CT Overview Spiral CT constitutes a volume-scanning mode in non-planar geometry with the patient scanned continuously in space and in time. Spiral scans encompass many rotations of the tube-detector system, while the patient is transported continuously through the gantry. The focus of the X-ray tube of course continues to travel on a circular path; however, relative to the patient it follows a spiral trajectory. This is illustrated in the following figure. Fig.3.8 Scanning Principle for Spiral CT The actual image reconstruction in spiral CT is in principle the same as in sequential CT. Identical algorithms; convolution kernels and the same hardware are used. However, an additional pre-processing step is required, the so-called z-interpolation. This is inted to generate a consistent planar data set from the spiral data for an arbitrary image position. The principal difference relative to sequential CT and a significant advantage at the same time is the inherent possibility of choosing image positions and reconstruction increments freely and retrospectively. The direct coupling between the scan position and the image position, which is unavoidable in sequential CT, no longer exists for spiral CT. 3.4 Multi-slice Spiral Scanning 4 Until 1998 in practice all CT scanners were equipped with single-row detector systems. Consequently, the scanned volume in a given time was limited, as far as the table feed was restricted. Four-slice CT systems have been introduced since 1998, eighth- and sixteen-row detector systems since Afterwards, 32, 40 slices scanner is introduced and for 2004, 64 slices scanner is available. It meant an important reduction of volume scan times by making possible faster table feed. Fig.3.9 Illustration for Development in CT system, from sequential scanning, to single slice spiral scanning, and now the multi-slice spiral CT scanning. *ISD = Inter Slice Delay 4 Most of materials used in this section is summarized from [1], [3] including some parts of text and pictures 19

25 Chapter 3. CT Overview Multi-slice Pitch As mentioned above, in multi-slice scanners, with more than one slice (typically M= 2, 4, 16, 32 or 64 simultaneously scanned slices) and with rotation time between 0,375s and 2s, the previous simple relation of pitch no longer applies. d The ratio of table feed d to total slice collimation S is termed the pitch: p ' = S For multi-slice scanner number of simultaneously scanned slices (M) has to be d taken into account for this parameter. p = M. S col It is obvious that p ' = M. p. E.g. a p =4 means a p= 1 for a 4-slices CT and this could produce misunderstandings. Therefore now the term of feed per rotation is used instead of pitch in the user interfaces Basic scanner geometry The following figure shows a schematic view of the n-slices CT scanner [1]. The multi-row detector forms a cylindrical surface with the radius R f + R d. With R f is the focus-isocenter distance, and R d is the detector-isocenter distance. Focus and detector are fixed with respect to each other. During a spiral scan, the focus moves along a spiral (helical) path with the radius R f. The z-position of the focal spot is given by the following equation: pscol z F ( α ) = ( α α 0 ) + z0 (3.3) 2π z 0 is the start z-position. α 0 is the start angle. By definition, z F =z 0 for α=α 0. Scol is the slice-width of one of the M = n simultaneously acquired slices, i.e. the centre-to-centre distance S det of the corresponding detector rows scaled to the centre of rotation. In the isocenter, the z-position of slice k, is given by the following equation: M 1 z( α, k) = z F ( α ) + k S col (3.4) 2 y-axis y-axis Detector k = n 1 0 Detector R d z F(α) β R f α x-axis z F (α,k) γ z-axis Focus Focus Fig 3.10 Schematic view of the geometry of n-slices CT scanner Projection onto the x-y plane (left) and projection onto the y-z plane for α=π/2(right) 20

26 Chapter 3. CT Overview During the scan, N P,2π projections per rotation are measured, which are characterised by their projection angle α n. Within a projection, each individual ray is characterised by its fan angle β m and by its index slice k. The sampled angles are given by the following equations. α n = α 0 + n α N + 1 (3.5) β m = m + a β 2 2π α = is the projection angle increment. N P,2π ( N 1) n 0, K, is the projection index, and N P is the = P total number of projections acquired during a spiral Fig.3.11 Sampled Angles β projection in spiral scanning scan. fan β = is the fan angle increment. β fan is the N total detector fan-angle, and N is the number of channels in each detector row, with m = 1K,, N. a is the so-called alignment, which takes into account that the central ray does not necessarily have to pass through the isocenter Rebinning The plane perpicular to the z-axis that contains the focus is referred to as the midplane. α and β uniquely specify the projection of a ray into the midplane. It is possible to replace, through rebinning, the projection angle α with another angle, π which is noted θ. θ is the azimuthally angle. θ = α + β 2 α is used to label rays when projection data are in the form of fan-beam projections, whereas θ is used when projection data are in the form of parallel projections, which are easier to use for computing than the fan-beam projections. The general approach for z-interpolation is to use the data redundancy in 360 CT scanning, because in any rotation of the X-ray tube, each line integral or projection value is measured one time in one direction with the angle α and one time in the opposite direction with the angle α+π [1], [7]. With the use of these opposite projections and of the rebinning methods it is possible to determine the projection at an arbitrary angular position, and with repeating this method to synthesize a second spiral with both data sets offset with a angle difference of 180. The logarithmic attenuation values, i.e. the line integrals of the object s attenuation coefficient µ along the measured rays, are denoted by f(n,m,q). Using the equation relating α and θ, a partial rebinning from fan-beam multi-slice projections Fig.3.12 Rebinning process: From fan-beam projections to parallel projections 21

27 Chapter 3. CT Overview f(n,m,q) to no equidistantly sampled parallel multi-slice projections g(l,m,q) by resorting and interpolation in the projection angle direction can be performed. Thus, a parallel projection is obtained by assembling rays from several fanbeam projections. This rebinning process generates a spiral of multi-slice non-equidistantly parallel projection. The rays in these parallel projections are tilted against the x-y plane by the cone-angle γ. Because the rays in a parallel projection correspond to different locations of the X-ray tube on the spiral focus path, the rays within one parallel projection also have a channel-depent z-position. The z-positions of the rays in the generated sampled parallel projections are given by the following equation. M 1 z( l, m, k ) = z F ( l, m) + k S col cos( β m ) (3.6) 2 The term M 1 k S col cos( β m ) takes into account the fact that the distance 2 between the focus and the azimuthal direction deps on the fan angle. z F (l,m) is the focus z-position of the ray corresponding to the azimuthal angle θ l and to the fan angle β m. It can be calculated using the equation relating α n and θ l. (, pscol π z l m) z F = θl + β m α + (3.7) π In this way, the z-positions for a parallel multi-slice projection at a given azimuthal angle θ can be determined Multi Row Detector Design Multi row detector utilizes the radiation delivered from x-ray tube more efficiently than single row detector. By simultaneously covering several slices, the scan time can be reduced significantly, or the smallest can be scanned within the practicable scan times. Using the adaptive array detectors, the rows inside the detector are very narrow, becoming wider toward its outer edges in the z- direction (longitudinal axes). A combination of collimation and electronic interconnections provide considerable flexibility in the selection of the slice thickness. At the same time the space required by the detector septa, and therefore the unused space is minimized. Fig.3.13 Multi row detector (Left) and the adaptive array arrangement (case 16 slices) 22

28 Chapter 3. CT Overview 3.5 Image Evaluation and Image Post Processing 5 The first obvious results of any CT examination are the axial cross-sectional images. Since these images are already available in digital form on a storage medium, they can be processed immediately by the processor. The evaluation of geometrical parameters such as distance, area, angle, and volume as well as density measurements are part of clinical routine. The tissue density for example can be determined by using intensities value averaged over a defined area. The term 2D and 3D refer to the image content. Views showing entire volumes are referred to as 3D Display [15]. Examples of the 2D post processing capabilities for instance is zoom and shift image segments. CT mainly used the transverse as the imaging plane. Therefore views of other orientation (axial or coronal, see figure 3.13) have to be reconstructed from original images. Some possibilities of further reconstruction is introduced: Fig.3.14 Reconstruction Orientation MPR Multi Planar Reconstructions Transaxial images are combined to form volume stack. The volume can be reformatted to secondary images in selected planes (sagittal, coronal or oblique) MIP (or MinIP) - Maximum (or Minimum) Intensity Projection Maximum (or Minimum) projection through the entire volume, which will show the part contain the maximum (or minimum) attenuation value. SSD Shaded Surface Display Surface images of tissue structure are created out of the volume dataset. A three dimensional object is calculated from voxels, whose threshold values are within a specific density range VRT Volume Rering Technique Possibility to rer different tissues, which have different densities, as a 3D object in different colors and with different brightness and opacity Fig.3.15 Some example of further reconstruction; from left to right: image result from MPR, MIP, SSD and VRT 5 Most of materials used in this section is summarized from [15], [16] including some parts of text and pictures 23

29 Chapter 4. Cardiac CT 4 Cardiac CT Due to the motion characteristic of the heart, it is very difficult to make cardiac image reconstruction with a reasonable result based on standard methods. The motion causes artifacts in the image, which is of limited use in diagnostic imaging. Therefore, a special algorithm dedicated for this cardiac imaging is absolutely required. 4.1 ECG-correlated cardiac image acquisition The Use of ECG Information As mentioned in the previous chapter, the movement pattern of the heart varies within the cardiac cycle. The strongest movement is present during contraction of the atriums and ventricles in systole. In the diastolic phase of the cardiac cycle less movement is present during the filling phase. Image acquisition and reconstruction need to be synchronised as accurately as possible with the movement of the heart, i.e. by using ECG information that is recorded in parallel with the scan data acquisition. ECG information can also be used to perform dedicated exposure-reduction for cardiac applications. It is also important to limit radiation exposure while maintaining the diagnostic quality. The intensity of the ECG is representative to the global electrical activity of the heart and correlated to its movement (see chapter 2.3). The following sections provide information about ECG-correlated data acquisition and ECG-controlled tube output acquisition developed in the CT system Prospective multi-slice ECG triggering For ECG-synchronised sequential imaging in phases of slow cardiac motion (i.e. the diastolic phase), a prospective trigger is derived from the ECG trace to initiate the CT scan, with a certain, user-selectable delay time after the R-peak. The delay time for scan acquisition after an R-peak is calculated from a given phase parameter for each cardiac cycle individually, (e.g. a percentage of the RR-interval time as delay after an R-peak, see next part in this chapter) based on a prospective estimation of RR-intervals. Usually, the delay is defined such that the scans are acquired during the diastolic phase of the heart. For instance, a sixteen slices CT scanner allows for simultaneous acquisition of sixteen adjacent slices per prospective ECG trigger for sequential coverage of the heart volume. As the patient table has to move sixteen slices in between the scans, one heartbeat has to be skipped for typical heart rates. This process is illustrated in the following figure 4.1. The exact examination time deps on the patient s individual heart rate. For diagnosis of dynamic processes in a specific region, ECG-triggered acquisition can also be performed without table feed in between the scans. The 6 Most of materials used in this section is summarized from [1], [3] including some parts of text and pictures 24

30 Chapter 4. Cardiac CT same volume is then acquired in corresponding phases of consecutive heartbeats. As no table feed is needed, scans can be acquired within every heartbeat for moderate heart rates. Fig 4.1 Sequential volume coverage with prospectively ECG-triggered four-slice scanning The two different filter techniques that are commonly used for prospective estimation of the position of the following R-peak are mean filtering (e.g. three previous RR-intervals) and median filtering (e.g. five previous RR-intervals). The median filter approach shows increased robustness for patients with moderated arrhythmia as single extra-beats are eliminated Retrospective multi-slice ECG gating ECG-triggered acquisition cannot be applied to continuous spiral acquisition with mechanical CT, which represents an important step towards true volumetric imaging. With spiral scanning, 3-dimensional evaluation based on overlapping trans-axial image slices becomes feasible due to considerably improved spatial z-resolution; faster scan speed and volume coverage. Retrospective ECG-gated spiral scanning is an attempt to synchronise the reconstruction of a continuous spiral scan to the movement of the heart by using an ECG-trace that is recorded simultaneously. The acquired scan data are selected for image reconstruction with respect to a pre-defined cardiac phase with a certain temporal relation to the onset of the R-peak, which defines the start point of data that are used for image reconstruction. Image reconstruction during different heart phases is feasible by shifting the start point of image reconstruction relative to the R-peak. For a given start position, an image stack at different z-positions covering a small sub-volume of the heart can be reconstructed owing to multi-slice data acquisition. The following figure 4.2 shows an example of how the cardiac volume is successfully covered with stacks of images reconstructed in consecutive heart cycles. All image stacks are reconstructed at specific timepoints or percentage phase during the cardiac cycle. Continuous volume coverage can only be achieved when the table moves slowly. Table feed is limited by the patient s heart rate. Indeed, if the table moves too fast, volume gaps between image stacks that are reconstructed using data from different heart cycles are present. The table speed has to be selected according to the minimum heart rate that is expected during the scan. 25

31 Chapter 4. Cardiac CT Fig 4.2 Continuous volume coverage with retrospectively ECG-gated four-slice scanning ECG gating and ECG triggering: advantages and trade-offs With the advent of multi-slice acquisition, ECG-gated spiral scanning has become feasible with significant advantages over prospective ECG triggering that are important for clinical applications. ECG-gated spiral scanning provides continuous volume coverage and better spatial resolution in patients longitudinal direction as images can be reconstructed with arbitrary, overlapping slice increments. Instead, ECGtriggered sequential scanning is usually restricted to scanning with nonoverlapping adjacent slices or slice increments with only small overlap. The scan time to cover the heart is here directly proportional to the slice increment. Retrospective analysis of the ECG results in less sensitivity to heart rate changes during the scan. The ECG trace can be retrospectively analysed and extra-systolic beats can be eliminated for reconstruction. With prospective ECGtriggering, the estimation of the next RR-interval may be wrong when heart rate changes are present (e.g.. arrhythmia) and scan may be placed in inconsistent heart phases. ECG-gated spiral scanning provides faster volume coverage than ECG-triggered sequential scanning as spiral scans data can be acquired continuously and images can be reconstructed in every cardiac cycle. Relatively long travel distances and travel times of the table are present for multi-slice acquisition in between two consecutive scans. This limits the scan cycle time (minimum time between start of two consecutive scans) and ECG-triggered scans my only be obtained in every second heart cycle for higher heart rates. ECG-gated spiral acquisition allows for imaging in a complete cardiac cycle using the same scan data set. ECG-triggered acquisition targets only one specific phase of the cardiac cycle and requires additional examinations to cover more phases of the cardiac cycle. During ECG-gated spiral imaging of the heart, data are acquired with a slow table feed and continuous X-ray exposure. Thus, ECG-gated spiral acquisition requires higher patient dose than ECG-triggered sequential acquisition. All spiral 26

32 Chapter 4. Cardiac CT data can be used for image reconstruction in different cardiac phases and no data have to be omitted. However, if only one dedicated cardiac phase (i.e. diastolic phase) needs to be targeted by retrospective data selection, the specific requirements of the clinical application should indicate whether ECGtriggered sequential scanning with less radiation exposure could provide sufficient performance and image quality. However, due to numerous benefits of ECG-gated spiral acquisition, dedicated development is under way with the goal to reduce radiation exposure via modified continuous spiral acquisition techniques. These techniques are supposed to maintain the important benefits of ECG-gated spiral scanning with X-ray exposure comparable to ECG-triggered sequential acquisition (see section 4.1.6) ECG synchronisation strategies With both prospective ECG triggering and retrospective ECG gating, the start points of data acquisition or, respectively, data selection for reconstruction have to be defined within each cardiac cycle during the acquisition. These start points are determined relative to the R-peak of the ECG signal by a phase parameter. The following phase determination strategies can be used. Relative delay: A temporal delay T del relative to the onset of the previous R- wave is used for determining the start point of the ECG-triggered acquisition or the start point of the reconstruction data interval. The delay time T del is determined individually for each heart cycle as a given percentage δ RR of the RR-interval time T RR. For ECG triggering the RR-interval times have to be prospectively estimated based on the prior RR-interval times. T del,1 =δ RR T RR,1 T del,2 =δ RR T RR ECG R R R T RR,1 T RR,2 Time Fig 4.3 Phase definition strategies: Relative Delay Absolute delay: Fixed delay times T del after onset of the R-peak defines the start point of the ECG-triggered acquisition or the start point of the reconstruction data interval. T del T del ECG R R R T RR,1 T RR,2 Time Fig 4.4 Phase definition strategies: Absolute delay 27

33 Chapter 4. Cardiac CT Absolute reverse: Fixed times T rev prior to the onset of the next R-wave define the start point of the ECG-triggered acquisition or the start point of the reconstruction data interval. For ECG-triggering the position of the next R-wave has to be prospectively estimated based on the prior RR-interval times. T rev T rev ECG R R R T RR,1 T RR,2 Time Fig 4.5 Phase definition strategies: Absolute reverse Different approaches are in use in clinical application practice today for different clinical applications. For free motion imaging of small anatomical structures (e.g. coronary arteries) in the diastolic phase with less cardiac motion, the relativedelay and absolute-reverse approaches are most frequently used. For functional imaging with retrospective ECG-gated, images need to be reconstructed in phases of maximum and minimum filling of the ventricles (diastole and -systole). End-diastolic reconstruction is feasible with the absolute-reverse approach, while the absolute-delay approach allows for most consistent reconstruction in -systolic phase. Retrospective ECG gating allows for viewing and analysis of the ECG signal after scan. Data are available during all phases of the cardiac cycle. This offers the possibility of retrospective modification of the synchronisation of the ECG trace and data reconstruction. Editing of R-peak positions that are detected inappropriately or that represent extra-systoles will have a beneficial effect on phase-consistent volume imaging Radiation exposure reduction approach Limiting radiation exposure while maintaining the diagnostic quality of ECGgated spiral scanning is an important goal of research and development. The relatively high radiation exposure for ECG-gated multi-slice spiral imaging of the heart is caused by continuous X-ray exposure and data acquisition at slow and highly overlapping table feed. The slow table feed is a consequence of the phase-consistent coverage of the heart volume in specific phases of the cardiac cycle. However, if the table feed is limited such that one phase can be covered consistently, all other cardiac phases of the cardiac cycle can be covered as well. If reconstruction in different cardiac phases is not needed, but rather only a very limited interval (i.e. diastolic phase) in the cardiac cycle is targeted during reconstruction, a significant portion of the acquired data and radiation exposure is redundant. On-line reduction of the tube output in each cardiac cycle during phases that are of less importance for ECG-gated reconstruction has a high 28

34 Chapter 4. Cardiac CT potential for exposure reduction. The nominal output is only required during the phases of the cardiac cycle that will be reconstructed. A straightforward attempt is acquisition of limited scan data intervals in every cardiac cycle with the X-ray radiation switched on and off (using prospective ECG triggering) while the table is moved continuously with appropriate speed. This approach enables continuous coverage of the heart volume with overlapping slices. However, similar to ECG-triggered sequential scanning, the risk of phase-inconsistent acquisition is implied if only moderate heart rate changes are present during the scan. Thus, another approach that maintains the benefits of ECG-gated spiral scanning is proposed. Instead of switching the X-ray radiation on and off, the tube output is modulated on-line with prospective ECG control. Within every cardiac cycle the tube output is raised to the nominal level during a limited interval in the diastolic phase. During the remaining part of the cardiac cycle the tube output can be reduced by about 80%. Thus, continuous volume reconstruction is still possible in all phases of the cardiac cycle. 100% 20 Fig 4.6 Retrospectively ECG-gated four slice spiral scanning with prospectively ECG controlled tube current modulation for reduced radiation exposure As illustrated in the previous figure, due to certain delays in tube-current adaptation, the function of tube current over time forms a trapezoid curve. The position of the windows of nominal tube output within the heart cycles needs to be defined prior to the scan. For normal heart rates between bpm (beat per minute), the exposure is reduced by 35-50%. For increasing heart rate the relative reduction decreases, as the time intervals of low tube output are shorter. 4.2 State of the Art: Cardiac CT Scanner 7 Development in cardiac imaging is done not only in the reconstruction technique, but also supported by CT scanner setup system itself. Increasing the time and spatial resolution would give benefit in reducing the artifacts mainly caused by the continuous motion of the heart. Several developments in the gantry system and the detector slice are done for this purpose. Time resolution is mainly determined by the rotation time of the tube along the patient body. Usually time resolution is defined as a fractional (half or one eight) of the rotation time deps on segment reconstruction technique used. Reduced the rotation 7 Most of materials used in this section is summarized from [14], [15] including some parts of text and pictures 29

35 Chapter 4. Cardiac CT time will provide better time resolution. Meanwhile, the spatial resolution can be achieved by using the multi-slice scanner due to the overlapping data scanned. One example of the implementation for cardiac-dedicated CT system is Sensation Cardiac 64, provided by Siemens Medical. The scanner is introduced worldwide at the beginning of It has the ability to provide the rotation time of 0.33 s/360 0, acquisition time of 8 s for the whole heart and spatial resolution less than 0.4 mm. It would give the possibility to freeze motion of the heart and capture the cardiac image much better. Robust cardiac acquisition is achieved even for the patient with high heart rate. Fig 4.7 Sensation Cardiac 64, CT system dedicated for cardiac application. The most important development applied in this CT system is the use of the special STRATON-tube, together with the Z-sharp double sampling technique and the UFC detector system supporting 64 DAS (Data Acquisition System) Straton X-Ray Tube The unique STRATON X-ray tube utilizes an electron beam that is accurately and rapidly deflected to the rotating Anode, creating two precise focal spots alternating 4,640 times per second. Straton is the smallest tube size that enables the fastest rotation time of 0.33 s. Using direct cooling oil, the tube enables the extremely high cooling rate of 5 MHU/min (heating unit loss) resulting in compact design and no cooling delay needed. Fig 4.8 STRATON tube, creates 2 focal spots rapidly, right picture describe the size of the tube. 30

36 Chapter 4. Cardiac CT Z-Sharp Double Sampling Straton x-ray tube produces double X- ray projections reaching each detector element. The two overlapping pro jections result in an oversampling in z- direction, known as double z-sam pling. Z sharp technology significantly increases resolution by doubling the number of measurement points utilizing the maximum 0.6 mm detector width. The resulting measurements interleave half a detector slice width, doubling the scan information without a corresponding increase in dose. Fig 4.9 Z-sharp technique Thus the spatial resolution of smaller than 0.4 mm is achieved. This finest resolution is produced without any increasing in dose, and can be done for all scan speed. As the result images come with previously unknown sharpness and clarity with spiral artifacts can be eliminated UFC Detector Siemens proprietary, Ultra Fast Ceramic (UFC ) Detector and the corresponding 64-slice detector electronics, enables a virtually simultaneous readout of these two projections for each detector element 2 x 32 slices for every viewing angle, resulting in a full 64-slice acquisition with speed data acquisition 2.5 Gbit/s. Speed and efficiency is based on Ultra Fast Ceramic (UFC) Detector which has the ultra short afterglow. Fig 4.10 Detector System The scanner is also supported by comprehensive software solution to help the clinician to do the measurement better. This includes the Calcium scoring accurate risk assessment coronary heart disease, Heart View - for better look of cardiac image, Vessel View semi automatic seg mentation of the coronary artery tree, Stenosis quan tification, ventricular con tour detection and cine display for 4D CT data. The whole application is applied in order to get the most from the cardiac CT scanner in analyzing the cardiac function Fig 4.11 Sensation Cardiac 64, Cardiac image with fine details. 31

37 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging 5 Automatic Optimum Phase Selection in Cardiac CT Imaging As mentioned in previous chapters, an optimal phase determination for reconstruction is a prerequisite for good image quality in phase-correlated cardiac imaging. Current cardiac reconstruction only allows for the use of a global phase. However the cardiac motion pattern is very complicated. A better approach is to find a local phase determination that might change from cycle to cycle, or even from slice to slice in image stacks. The algorithm for automatic optimum phase selection globally or locally is described in the diagram below: Fig 5.1 Flowchart depicts the proposed procedure for automatic determination of optimum phase in reconstruction. The Optimum phase can be derived for both overall heart motion and for specific structure, produce both globally and locally optimum phase. The automatic phase selection in cardiac CT imaging is first introduced by Phillips Research Laboratories by the of 2004 [2]. The proposed algorithm was used for globally characterizing the motion pattern of the heart. The method successfully shows motion pattern of the heart, and delivers the systole-diastole phase directly from a motion map (flowchart marked with ) 32

38 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging Projection data from gated cardiac CT scan are used, including all slice samples from a multitude of cardiac phases (at %RR peaks). The motion information is then extracted by simply detecting the changes of pixel values between images at consecutive phases from all data set reconstructed. But for specific part like coronary arteries, the result is not accurate enough. The motion of coronary arteries doesn t exactly coincide with the motion of the heart. Coronary arteries have more complicated motion pattern. The artery lies in the surface of the heart wall, from aorta to the epicardium. The motion of the artery is overshadowed by the chamber motion, thus the real motion pattern of the coronary arteries is not presented well. The motion map algorithm should be improved and applied in an optimal way. Introducing the subset voxel definition and histogram-weighting vector is one solution proposed for this improvement. Here the algorithm is developed for the purpose of analyzing motion pattern of specific structure and defining local phase selection (flowchart marked with ). 5.1 Volume Dataset and Image Slice Reconstruction In spiral acquisition, the acquisition time is related to a specific z-position by: z z o t z = TROT p * S col (5.1) with p = pitch, Zo = start position, Scol = collimation width of one slice, Trot=rotation time of the scanner [7]. In parallel with the spiral scanning, a synchronous ECG signal is recorded with the same time domain e(t). Therefore, for each image slice position, the corresponding cardiac cycle is known. (see fig 4.2 for more detail retrospectively multi-slice ECG gating). A certain delay related to R-R peaks is used to correlate projection data to a preselected cardiac phase. This reconstruction produces volume dataset V ( P u, i),, where P denotes a certain cardiac phase, e.g. as a percentage of R-R cycle, ( u ) T x, u y u z = ( i, i i ) T V denotes a discrete voxel { 1,2 n } u =, defines the centre of the image volume, and i, x y z dimension of image data used. i x, y, z,... x, y, z = respectively describe the For good spatial resolution, normally the voxel reconstructed with the size of n x * n y = 512 x 512 pixel, while n Z defines the number of image slices reconstructed with certain increment dz, given byn z Z lastpos Z0 = + 1. (5.2) dz The shift along the z-axis is considered, while the centre position in x- and y- T direction is maintained. u k = ( u, u, u 0 + kd ) ; for k =,1,2... n, with slice x y z z 0 z index k indicates a specific z-position of the patient with the relationship: Z = Z k * (5.3) + k 0 d z 33

39 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging The orientation of the reconstruction volume coincides with the orientation of the acquisition geometry, where the z-axis is parallel to the axis of rotation. The reconstruction of volume dataset is done by using the CardioRecon tool. The parameters are set to specify the reconstructions: Field Of View, x-axis and y-axis centre, kernel used, slice increment, and also the cardio parameters: gating strategies, and phase %RR selected. More details about the CardioRecon tool are explained at the of this report. 5.2 Subset Voxel Definition A subset voxel is needed when it is necessary to extract the motion information in the region of interest of specific structure of the heart. A voxel mask is created to cover a certain area at every slice that contains parts that of interest. In this thesis work the analysis of the Right Coronary Artery and its motion pattern is considered. Right coronary artery is the coronary vessel showing most complicated motion. To obtain a motion profile for small anatomic structures like coronary artery, it is necessary to use volumes with an increased resolution. One thing that can be done is by selecting the Field of View of appropriate size and centre in the reconstruction, focusing to the parts of interest. Subset Voxel S ( P, v, j) are created which define sub volumes S ( P, v, j) V ( P, u, i) to spatially resolve the motion information of specific structure contained. In the same manner, v = ( v, v, v ) T defines the centre of the subset volume specific for each slice, and ( ) T x x y y z S = denotes a discrete voxel { 1,2 n } j j, j, j Voxel masking z j x, y, z,... x, y, z =. A voxel mask is needed to define the centre of the Subset Volume and the size of the area of region of interest. The mask is derived from axial images reconstructed at common gating phase 60% of RR peaks within the EGC signals and will then be applied identically to volume dataset at all other phases. With the positioning of the voxel mask (shifting-masking), the motion information extraction now examined specifically for a certain anatomic structures (fig. 5.2). T A voxel mask characterizes the subset voxel s centre v k = ( v, v, u + kd ), which varies for each slice k deping on the position of the target object in the axial slice image. The centre position must be determined in order to follow the structure over the spatial position slices by slices. Here the centre position in x- and y- direction is no longer maintained. The cubic subset volume has dimension S n x = xk S n y, which describes the size of S the subset area -the region of interest. And n z defines the number of slices along the z-direction which contains the specific object concerned, clearly S V n n. z z yk z z 34

40 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging Deping on the position of the voxel mask within the reconstruction volume, S T T S S the subset voxel S n n y n n y is created with x S vk n z 2 2 S S S T V V V dimension ( ) ( ) T n x n n n n n. y z x y z x S jk v k + n z 2 2 Fig 5.2 Voxel mask derived from each axial slices reconstructed at common 60%, the mask defines the centre of the subset voxel and the radius There are 2 methods developed to determine the centre and the radius of the region of interest in the subset voxel definition: Manual definition and semi auto definition Manual Definition of Subset Volume The centre and the radius of the subset voxel are set manually for each slice. User has to click twice on each axial image: the centre position and another point represent the radius area of the subset. The Algorithm then calculated the cubic radius automatically and set the maximum value from all slices as the common radius. Fig 5.3 Manual Definition of Subset Volume Semi Auto Definition of Subset Volume With semi automatic definition user just need to click a seed point on each axial slice as the centre position of the object concerned, then the algorithm automatically will grow the region by analysing pixel value around the seed point with iteration process. Region growing is done by simply calculating the difference between the pixel value at centre point and pixel value around its neighbourhood. Certain limit range is set to ensure only pixel with certain value is included. Fig 5.4 Region Growing applied for Semi automatic definition 35

41 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging Only pixels containing contrast agent in the range of 300 HU ± 50 HU are considered. Commonly the Iodine concentration is used as the contrast medium. The algorithm will set the maximum and minimum size of region grown to ensure that the subset area is wide enough to capture the motion of the coronary and small enough to focus the region of interest only on specific area concerned. Again, the maximum value from all slices is set as the common radius. After a certain region is grown, the centre of the subset voxel is modified based on the size of the region. This re-defining centre process will ensure that the subset voxel focus the covered area exactly into the coronary artery part. 5.3 Motion Information extraction The reconstructions are made using several phases with certain step, e.g. for each 2% or 5% of RR-peaks within ECG signals recorded. This produces 4D data as whole cardiac image volume dataset or subset volume created with certain voxel mask. To obtain information about the motion pattern, the similarity between images at the same slice index k for each consecutive phase is calculated. Similarity between image slice Vk ( Pl, u, i) reconstructed at certain cardiac phase P l and Vk ( Pl 1 u, i), reconstructed at phase P l 1, or V k ( P l 1 u, i), reconstructed at phase P + l + 1 is calculated. In case of coronary artery motion analysis, it is the subset volume S P, v j and the subset voxel reconstructed at the consecutive phases. ( l ) k, To optimize the efficiency of the algorithm, several simple methods can be used in order to calculate the similarity between images, which is developed by considering only the differences in pixel value. There are two methods introduced here [2], [9]: 1. Subtract Method /Energy-based n n x y Mean Absolute difference, MAD (A, B) = 1 A( ix, i y ) B( ix, i y ) (5.4) ( n * n ) 2. Correlation Method Correlation value, COR (A, B) = x y x y ix i y A( i, i ) A * B( i, i ) B n n n n x y x y 2 2 ( A( i, i ) A) ( B( i, i ) B) x y x y ix iy n ix n iy (5.5) With A and B are two image slices at two consecutive cardiac phases: between S ( Pl, u i) and S k ( Pl 1 u, i), or between S k ( Pl u, i) k, x y ix iy, and Sk ( Pl 1 u, i), x y +. The method simply shows the similarity of two image data sets (with COR)- or the difference (with MAD or 1-COR). This value is used as the indication about the motion state of the heart during that phase. Consequently, within the phase where heart has its high motion, the images would have great difference at the consecutive phases. In contrary, when heart 36

42 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging is in its stable phase, images within consecutive phases will have great similarity. The differences between images in consecutive phases indicate the amount of the motion of the heart within that phase. This linear relation simplifies the algorithm to set the relative motion value directly equal to the difference value between images in consecutive phases Histogram Weighting Function The difference between images is calculated basically just by comparing pixels between two image data. In order to focus the difference information only on the coronary arteries part, a histogram weighting function is needed. The HU-based histogram-weighting vector enables the algorithm to get rid of irrelevant pixel values influenced by motion from other parts. The weighting function is applied by multiplying the pixel value with certain weighting vector based on the image histogram. It gives maximum weight factor for pixel with certain important value and slightly reduced the weight factor to zero for other pixel value. Coronary arteries or other parts containing blood with contrast media will have the pixel value about 300 (contrast media has HU 300). Therefore the weighting vector used should have maximum value set to intensity HU 250 to 350. It will ensure that only pixels containing contrast media are considered. Several type of weighting function can be applied for this purpose. Both Gaussian function and trapezoid are proposed. Using the trapezoid function, which is usually used in the post processing reconstruction, would give benefit as the high flexibility to modify, not only in setting the centre but also the width of the vector, slope, and range of the weighting vector. Fig 5.5 Histogram Weighting function (a) Gaussian Function, (b) Trapezoid function. the trapezoid weighting function gives more flexibility to set the parameters. The weighting value is set to max equal 1 for pixel value of Contrast media The Motion Information The motion information is calculated by measuring the similarity between image slices reconstructed at consecutive phases. This calculation delivers the motion value M ( P, k) measured at specific phase P l for the specific slice k. l As mentioned before, there are 2 methods that can be used for this calculation: Mean Absolute Difference (MAD) and Correlation (COR). Results obtained from 37

43 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging both methods are quite similar to each other. Thus, we decide to use the Subtract method MAD since the technique is more robust and simpler to be applied in the algorithm. Using the MAD value, the motion information can be extracted by the following formulas: M ( P l, k) = MAD ( Vk ( Pl u, i) V with the phase 0 % P < 100%. l,, k ( Pl 1 u, i), ) + MAD ( Vk ( Pl, u, i), V k ( P l u, i) + 1, ) (5.6) Since the heart motion is periodic, for the first phase at 0%, the similarities are calculated between the last and the next phases, e.g., phase 99% of RR peaks and phase 2% of RR peaks. The same method can also directly be applied for the subset volume of specific anatomy structure like coronary artery, using the Subset volume Sk ( Pl, v, j) developed from the voxel mask created before. The motion information follows the same formula: M ( P l, k ) = MAD ( Sk ( Pl v, j) S,, k ( Pl 1 v, j), ) + MAD ( Sk ( Pl, v, j), Sk ( Pl 1 v, j), + ) (5.7) Picture 5.6 describes the scheme of motion information extraction by comparing image data of the same slice reconstructed from consecutive phases. Fig 5.6 a volume data set contains the Subset voxel. Motion value is calculated by comparing image slices between consecutive phases. 5.4 Motion curve and Motion Map Motion curve and Motion map are the basic means for stable cardiac phase selection [2]. Both are the graphical display of the motion value that visualize the motion pattern of the cardiac images, and directly provides information about the optimum phase with minimum motion Motion Curve Motion curves are obtained by simply plot all motion values M ( P, k) [2] for all slices reconstructed at each phase, in one plot figure. (for example fig 6.3) l 38

44 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging From the motion curve, the average of the motion value for each phase overall slices can be measured. nz 1 M mean ( Pl ) = M ( Pl, k) (5.8) nz k= 0 This mean value is needed to see which cardiac phase shows the high similarity averaged over all slices. By comparing the mean motion value M mean ( P l ) from all phases, the global optimum phase can be derived. This is the phase with minimum motion value, thus cardiac phase with less motion of the heart, e.g. at the systolic and diastolic phase of the heart. A filter can be applied in order to reduce noise and enhance sensitivity. The filter is simply done by replacing M mean ( P l ) with the average of 3-point running value in three consecutive phases. 1 1 old M mean( Pl ) = M mean( Pl i ) (5.9) 3 i= 1 A minima search then applied to define the phase, which has minimum motion value for overall slices. P global = min( M mean ( Pl )), for all P l, where 0% P l <100% RR (5.10) The value then becomes the global stable phase. This value determines the stable phases averaged for the entire cardiac volume. By limiting the range of phases when doing the minima search, two general stable phases relate to systole or diastole of cardiac cycle could also be derived. Based on the fact that the heart typically sps about 1/3 of its time in systole and 2/3 of its time in diastole, both stable phases are derived as follow: P globalsystole = min( M mean ( Pl )), for all P l where 0%< P l <35% RR) (5.11) P globaldiastole = min( M mean( Pl )), for all Pl where 35%< P l <100% RR) (5.12) Motion Map Motion maps are obtained if the motion values M ( P l, vk ) are grey-coded [2] in the contour form. Along the horizontal axis of the map, the slice k is noted versus the cardiac phase P with certain step in %RR cycle along the vertical l axis. The area with the same value will be coded with the same colour. Dark area corresponds to phases with less motion and bright area corresponds to phases with high motion. The Motion pattern of the heart could be evaluated by analysing the contour of this motion map. (for example fig 6.4) Phases with minimum motion can be determined for a specific slice. Minima search is calculated among motion value of the specific slice for all reconstruction phases. It gives a stable phase regarding to specific z-position of the patient s body in the predefined neighbourhood of the previously determined global stable phases. The range of ±10% is used to avoid the big gap between each step in further reconstruction. P opt ( k) = min( M ( Pl, k)), for all P l, where Pglobal -10%< P l < P global +10% (5.13) 39

45 Chapter 5. Automatic Optimum Phase Selection in Cardiac CT Imaging 5.5 R-shift vector modification By using the CardioRecon program, modification for the delay phase is not possible to be done for each slice, but only for each cardiac cycle, which contains several slices. The information about gating phases used for reconstruction is included in the R shift vector. A local phase can be realized in cardiac reconstruction by modifying the R-shift vector and introducing a local delay with respect to the R-peak to the related cycle. Up to now, only a global delay has been used. Fig 5.R Shift vector contains the gated phases used for each cycle. Up to now, only global phase used, means R_shift (1)= R_shift (2)= R_shift (3)=global phase. Cycle-Depant optimum phase is done by modifying the R_shift vector, which introduces indepent local delay for each cycle. The stable phase for each slices then need to be transformed into cycledepent optimum phase. From the motion map obtained before, mean value of several slices contained in same cycle can derived as follow: Ck 1 M mean ( Pl, cycle) = M ( Pl, k), (5.14) Ck with Ck defines amount of slices contained in the related cycle. The cycle-depent optimum phase then derived with the minima search between these mean values: P opt ( cycle) = min( M mean ( Pl, cycle)), for all P l, Pglobal -10%< P l < P global +10% (5.15) The information regarding amount of slices contained in one cardiac cycle is gathered from the reconstruction data recorded before. One cardiac cycle might have more slices compared to another cardiac cycle. The specific slices contained in a cycle is derived analytically by comparing the acquisition time of each slice and its relation to R-R time interval of ECG signal in time domain (equation 5.1). 40

46 Chapter 6. Result and Analysis 6 Result and Analysis The aim of this section is to analyze how the developed algorithm works step by step and at the try to validate it with several patient data. The algorithm is developed using MATLAB 7.0 utilize functions in image processing toolbox 4.2 [9]. Script of the written function is attached at the of this report. Each process is simulated separately to see the functionality of the algorithm. Both a test data and patient data are used for this investigation. 6.1 Simulation with Test Data First, the simulation is focused on the similarity calculation. To analyze the functionality of similarity calculation, a test data set is created. The test dataset consists of 10 images containing a sphere with different sizes. To simulate the behavior of the heart motion, in temporal position 3 and 6 represent the phase of 30% and 60% RR, the circles are changing slowly within the consecutive positions. Meanwhile, at other positions the radius of the circle changes rapidly. Temporal Position Radius (pixel) Fig 6.1 Test Data, consists of 10 images contain sphere with different size The test dataset is analyzed with the similarity calculation method introduced in chapter 4. Both method Mean Absolute Difference (MAD) and Correlation (COR) are applied to see how the algorithm measures the similarity between images in consecutive positions. Fig 6.2 Motion curve simulated using test data. Left is result for MAD, right is result for COR (blue line) and 1-COR (black line). Both MAD and 1-COR give quite similar result From the curve, both the MAD and COR value show quite the same result for measuring the similarity between images. At temporal position 3, the radius of the circle changes slowly from position before and after ( ) and similarly in temporal position 6 ( ). They are represented as two minimum values 41

47 Chapter 6. Result and Analysis in the motion curve. At temporal position 4, the radius of the circle changes faster ( ) which is expressed as a maximum in the motion curve. From the simulation, it is clear that the algorithm is able to detect the change of object size by calculating the similarity between images in consecutive phases. Both methods are applicable, but for simplicity the MAD method is chosen in the algorithm. 6.2 Analysis with Patient Data Patient data are analyzed in order to demonstrate the potential of the algorithm using clinical data. Several processes are tested separately to see the functionality of the algorithm. Parameter settings of the algorithm are optimized Overall Motion Analysis of the Heart The 4-D data of Volume data set with grid size of 512x512x223 voxel times 50 phases (from 0% to steps) are used, applying the mean absolute differences calculation for the determination of the difference between image slices. The difference value is directly related to the motion value. The motion value averaged over all slices is plotted in one figure (fig. 6.3). The black thick line represents the mean motion values for each specific phase M mean ( P l ), which is used to determine the global Stable Phase P opt. The white lines represents the filtered mean value using 3 points running averaging. Analyzing this mean value by minimum search will provide the phase with minimum motion, defined as the global stable phase. In this case, it is determined at phase 64% RR. Fig 6.3. Motion curve; plot all motion values for all slices reconstructed with all phases. Black thick line represents mean motion value for each phase, and white thin line represents mean values after averaging. Doing the minima search the global stable phase can be derived. Systolic stable phase is reached at 26%, and diastolic phase at 64% By limiting the range when observing the global stable phase, two global stable phases for systole and diastole period can be defined. The diastolic stable phase is determined at 64%RR-cycle. And at 26%RR-cycle, another cardiac phase with little motion has been determined, corresponding to systolic state. 42

48 Chapter 6. Result and Analysis With 2 general stable phases at systole and diastole, the reconstruction with minimum motion can be done for any state of the heart. For example, in order to evaluate the ESV ( systolic volume) of the left ventricle -which indicates percentage of blood ejected to whole body, reconstruction should be gated at systole stable phase to ensure the good quality considering reconstruction of the image in heart s minimum motion. In the motion map the motion value is mapped onto a greyscale in the contour form. Dark area corresponds to phases with less motion and bright area corresponds to phases with high motion. The motion map shows optimum phase with little cardiac motion at systole (26%RR-cycle) and another optimum phase is shown at phases representing diastole (64%RR-cycle). It can be clearly seen that the global stable phase (red line in the motion map) is obtained just before the high motion of systolic, represented by brightest areas in the motion map. It shows exactly the same result to what clinician did in defining the stable reconstruction phase manually. Fig 6.4. Motion Map; plot grayscale- contour of motion value for all slices reconstructed for all phases. Two Red lines represent systole-diastole global stable phase, and white markers represent optimum phase related to each z-position in predefined range. Along the red line, white markers indicate optimum phase for every slice k within the range of ±10% from each global phase. Furthermore, phase with minimum motion for each slice would be useful in order to optimize the image quality concerning a specific z-position within the heart, for example the valves or specific location of the heart wall Analysis of Coronary Arteries Motion One important thing in the coronary arteries motion analysis is to define a subset voxel containing the coronary artery. Using a manual definition, a subset volume is selected with grid size 43x43x102. The region of interest is determined so that 43

49 Chapter 6. Result and Analysis it is wide enough to capture all motion information about the coronary arteries concerned, in this case the RCA (Right Coronary Artery). Axial image slices are reconstructed at normal gating phase 60%RR and used for the subset definition. The scheme below describes the workflow of Subset Voxel definition. Fig 6.5. Defining a subset voxel of RCA, (a) axial slice with improved resolution contains coronary artery part (b) centre of the subset manually selected (c) region of interest is grown that is wide enough to capture the motion of the arteries (d) histogram weighting vector to weight the pixel value, here the trapezoid function is used (e) Weighted subset voxel which will be used in motion information extraction. Simulation on Manual and Semi auto Subset definition Both methods can be applied in defining the subset volume. The benefit of using the semi auto method is the possibility to reduce the interference from the user in defining the radius of the subset area. And it is also possible to re-define the centre of the coronary arteries right into the middle point of the subset slices. Meanwhile, using the manual method, the centre and size of subset can be wrong, affected by the human error due to the inaccuracy of the user in defining the centre point. Fig 6.6. Defining subset voxels containing the RCA, (A) Manual Definition of Subset, rad=71 pixel. (B) Semi Auto definition of Subset, rad = 67 pixel. With region growing and re-defining centre process, the coronary part lies exactly in the middle of the subset slice. 44

50 Chapter 6. Result and Analysis From the simulation, both methods produce quite the same area of subset slices. But of course with additional region growing and centre re-defining process, the semi auto definition requires more computational effort compare to the manual definition. And it is also considered as a time consuming process in the algorithm. Application on Histogram weighting vector Subset volume need to be weighted with certain histogram weighting vector in order to emphasize the difference calculation only for the arteries which contain contrast agent and eliminate motion that is influenced from other structure. Using the trapezoid function, more flexibility gained in designing the weighting vector. Here maximum value set to intensity (HU 250 to 350) and slightly reduced to zero at the intensity 1200 and This parameter is acquired with trial and error process in order to get the subset slice containing only important pixel value (the contrast agent in Coronary artery). Fig 6.7 Simulation using several histograms weighting function, (a) with Gaussian function centre=1300, (b) with trapezoid: , (c) with trapezoid: trapezoid function is used as the weighting vector in the algorithm, the weighted subset now contains only the coronary parts pixel value. The similarity calculation is later applied to the weighted subset voxel of the coronary arteries. Only the motion of the coronary arteries part now has impact on the difference in image pixels. Extracting the motion value of this subset volume would give more reliable information about the motion pattern of the coronary artery. Modification of the R-shift vector The same technique using the motion curve and the motion map is applied to find the optimum phase based on the subset voxel. Global stable phase, and slice depant phase can be derived from these 2 plots as the same way like the heart overall motion analysis. The cycle-depent optimum phase is derived from the motion value for all slices containing the related cycle. This local delay is introduced in the reconstruction by modifying the R-shift vector. Here, a simulation is conducted to check whether the R shift modification in the reconstruction works properly or not. By manually modify the R-Shift vector in the reconstruction, the image is observed too see the change of each segment that is reconstructed in different cycle with different local delay. 45

51 Chapter 6. Result and Analysis Fig 6.8 Simulation by locally changing the phase of one cycle and observe the segment affected on the MPR. Relation between segments reconstructed in a cycle is shown in the below figure. Upper left is normal reconstruction use global phase; upper right is reconstructed with modification at cycle 9, lower left at cycle 10 and lower right at cycle 11. Several reconstructions are made by introducing local delay only for a certain cycle. By comparing the image with the global-phase reconstruction, it is seen that the modification takes place only on a segment image related to the cycle phase that has been modified. This simulation proves that the R shift vector, which is modified by the algorithm, works properly in the reconstruction. 6.3 Validation with Patient Data In this section, results for cardiac patient data set are presented. The purpose of the validation is to analyze motion pattern of the coronary artery and to get the local phase, which the coronary artery can be reconstructed with, segment by segment in its minimum motion. Systole and diastole state is no longer important since the motion of the coronary artery has its own characteristic and indepent to the motion of the overall heart. Thus, only one global phase is derived Three patient data sets are used for the validation purpose. The projection data were acquired using Siemens Sensation 16 with parallel ECG recording and R- peak detection. The patient data is obtained from University of Tübingen, with scan parameter [10] described in next table. Table 1. Scan and Reconstruction Parameters CT system Sensation 16 Scan Mode Spiral Collimation 16 x 0.75 mm Pitch 4 Rotation Time s Tube Setting 120kV/433mAs Kernel Soft B30f Version VA40 Contrast Medium Yes For each patient, the cycle-depant optimum phase is applied. New reconstruction is made applying the local delay for each cycle. The result is compared with the reconstruction using conventional technique to see the improvement have been made by the algorithm. 46

52 Chapter 6. Result and Analysis Result for Patient 1 Table 2. Data Patient 1 Image Slice Reconstruction Field of View 181x181 mm 2 x,y centre (9, -3) Coverage in z-dir mm Reconstruction Slice 1mm increment 0.6 mm Volume grid size 512x512x237 Subset Volume grid size 43x43x102 Phase step 50 steps (@2%) Sex, Age Mean heart rate Maximum heart rate Minimum heart rate Patient Data Male, 66 yrs 58.6 bpm 60 bpm 56.6 bpm The algorithm is implemented to find the global and local optimum gating phase automatically. The motion value is calculated using Mean Absolute Difference and plot in the Motion Curve and Motion Map. Automatic Global phase is derived from the motion curve at 56% RR. Analyzing the motion map, it is clear that the global phase lies in the darker area 40%-70% indicates the phase with minimum motion. Cycle to cycle local phases are derived within the range 10% in motion map. The maximum local delay is 60% and the minimum is 52% RR The cycle-depant local phase is implemented by changing the R shift vector. Here 9 cycles are affected and used as the local delay in the reconstruction Fig 6.9. Motion curve and Motion map of right coronary artery for patient1. The R-Sift Modification is plot at the right figure. 47