Correction of Ring Artifacts in Micro CT Wesley Armstrong, Matthew Teeter & Dr. David Holdsworth General Biophysics Laboratory 3970Z Dr. Ian MacDonald April 1, 2011
Introduction X ray computed tomography (CT) is a powerful medical imaging modality, with many applications. It collects x ray images in various planes, then these images are compiled to form a three dimensional image which can be analyzed. CT images have high resolution and are usually very useful for making measurements and comparing objects. As powerful as CT imaging is, there are still many problems; one of these is the occurrence of ring artifacts. Ring artifacts are a series of concentric rings that appear superimposed on the image, and interfere with image analysis (Kinney et al, 1989). This experiment applies a potential correction to these images to reduce the severity of the ring artifacts. The main goal of this experiment is to determine if images up to a year old can be corrected, and to what degree. This experiment will provide insights into the operation of the scanner and lead to improvement in image quality. Theory Micro computed axial tomography is a common technique for examining the interior of objects and small animals. The images are collected by an X ray source opposite a detector that rotate around the specimen. In this case the detector is made of cadmium zinc telluride. This collects two-dimensional X ray images based on the specimen s ability to absorb the X rays. These images are then reconstructed by the computer into a single volume of data which is a three dimensional representation of the specimen (GE Healthcare).
Ring artifacts are a major problem in CT scanning. They appear as a series of concentric rings on the image collected (Kinney et al, 1989). They are caused by imperfect detector elements such as gain error in the detector array. These errors include: changes in detector element sensitivity in between bright field calibrations caused by changes in temperature, non linear detector element response caused by beam hardening effects, and drifts in the detector bright field correction caused by hardware shortcomings in the X ray tube, scintillator, and read out electronics (Sijbers & Postnov, 2004). These errors over or under estimate attenuation values throughout the scan (Kyriaku et al, 2009). Ring artifacts interfere with image analysis and post processing such as subsurface analysis or surface rendering when studying musculoskeletal implants. Subsurface analysis is affected when looking for cracks in the implants. The ring artifacts can distort or hide the cracks. Surface rendering is superimposing two images to compare the failure points (Teeter et al, 2010). Ring artifacts can interfere with measurements as they may be on the surface of the object. The correction applied in this experiment was a bright field calibration. A bright field is a scan with nothing in the field of view. This collects information about the noise generated by the scanner. When reconstructing the threedimensional image, the computer corrects this noise in each of the images using the information from the bright field scan. The current bright field calibration techniques scan for a few seconds before the scan, and are insufficient to cover all the error sources involved in producing ring artifacts (Riess et al, 2004). A scan of 100 minutes was run with nothing in the field of view. By studying the individual
acquisition frames from the scan, it was determined that they changed from frame 0 to frame 1199, which demonstrates that the detector changes over time (since nothing in the scan volume was changing). A new bright field was constructed by averaging the last 100 frames of the scan, which is when the detector has stabilized. This new bright field was then used as described in the next section. In this experiment, standard deviation of the grey scale values was used as the quantitative measurement. In the images, the standard deviation of the grey scale values is due to noise. Since the images are not calibrated the standard deviation is unitless. In these images the standard deviation adds in quadrature. This is represented by the following equation: SDTotal 2 = SDRings 2 + SDPhoton 2 (1) Where SDRings is the standard deviation due to the ring artifacts, and SDPhoton is the standard deviation due to photon counting noise. Microview, from GE Healthcare, was the image processing software used in this experiment. It was used for measuring the standard deviation in various regions of interest (ROI) in the reconstructed images. Methods The images used for analysis were collected between April 2010 and January 2011. The scanner used was the General Electric explore SpeCZT at the Robarts Research Institute. The objects in these images are acetabular liners and tibial inserts. Each of the images contained severe ring artifacts. The images were then reconstructed using a new bright field as described above. To do this, the original
bright field had to be manually replaced by the new one and then the reconstructions had to be run. This was all done on the console of the scanner. Using the imaging program, Microview, the standard deviation of the grey scale value in this image was measured in four different ROI s. The ROI s were 3 mm cubes and the measurements were made at the center of the ring artifacts, above the center of the rings, in the object, and outside of the object. The measurements were made before and after the bright field correction was performed. In this type of image the noise adds in quadrature, and the measured value is SDtotal. To determine the SDrings, we found regions in the images with a very low severity on the periphery. The standard deviation of this region was measured and this value was used as the estimate for the photon counting noise (SDphoton). With this data, SDrings was determined using equation (1). Results Standard Deviation 180 160 140 120 100 80 60 40 20 0 ROI 1 (Center of the Ring Artifacts) April July November December January January Figure 1 This shows the standard deviation due to the ring artifacts before (green) and after (purple) the bright field correction was applied. The measurements were made using a 27 mm 3 ROI with Microview. This ROI was placed at the center of the ring artifacts.
Standard Deviation 100 90 80 70 60 50 40 30 20 10 0 ROI 2 (Above Center of Ring Artifacts) April July November December January January Figure 2 This shows the standard deviation due to the ring artifacts before (green) and after (purple) the bright field correction was applied. The measurements were made using a 27 mm 3 ROI with Microview. This ROI was placed slightly above the center of the ring artifacts. Standard Deviation 50 45 40 35 30 25 20 15 10 5 0 ROI 3 (In Object) April July November December January January Figure 3 This shows the standard deviation due to the ring artifacts before (green) and after (purple) the bright field correction was applied. The measurements were made using a 27 mm 3 ROI with Microview. This ROI was placed inside the tibial insert or acetabular liner.
Standard Deviation 45 40 35 30 25 20 15 10 5 0 ROI 4 (Outside Object) April July November December January January Figure 4 This shows the standard deviation due to the ring artifacts before (green) and after (purple) the bright field correction was applied. The measurements were made using a 27 mm 3 ROI with Microview. This ROI was placed outside the object. Figure 5 Image of acetabular liner collected using the General Electric Explore SPeC7T at the Robarts Research institute, University of Western Ontario, London, ON. This image was collected before the bright field correction was applied.
Figure 6 Image of acetabular liner collected using the General Electric Explore SPeC7T at the Robarts Research Institute, University of Western Ontario, London, ON. This image was collected after the bright field correction was applied. ROI Average Reduction Average Percent P Value in SD Reduction in SD Center of Rings 79.04 60.43 3.89298 x 10-7 Above Center of 28.05 52.97 0.007266483 Rings In Object 13.58 34.72 0.00673873 Outside Object 15.51 49.25 0.000117016 Table 1 This table is a summary of the information provided by Figures 1 4. The average reduction in SD is the average value of the differences between the SD due to the ring artifacts before and after the bright field correction. The average percent reduction is the percentage of the uncorrected SD due to the ring artifacts that has been eliminated by the brightness correction. A paired two tailed t test was performed on the data to determine if there was a significant reduction in SD due to ring artifacts. This study was carried out in order to determine the improvement in the standard deviation due to the ring artifacts. Figures 1 4 compare the SD due to the ring artifacts before and after the bright field correction. From the graphs it is clear that there is a significant improvement in each of the images in each ROI. This was verified by performing a paired, two tailed t test for the data from each ROI.
Each ROI had a significant improvement in SD and the p values are displayed in Table 1. Figures 5 and 6 show a before and after image of an acetabular liner, and it is clear that there is a significant improvement in the ring artifacts in Figure 6. Discussion This experiment was carried out to test if ring artifacts in micro CT images can be corrected by using a bright field calibration. A bright field collected over a 100-minute scan was used to reconstruct images collected up to a year old. The standard deviation of gray scale values was measured before and after the correction to determine reduction of ring artifacts. The experiment was successful as there was significant improvement in four regions of interest in each image tested. Figures 1 4 compare the standard deviations before and after the correction for each image. Table 1 shows a summary of the results and shows the results of a paired two tailed t test that was performed to prove the significance of the reduction in standard deviation. Figures 5 and 6 show the qualitative improvement in image quality. It is clear that both qualitatively and quantitatively, the severity of the ring artifacts in these images had been improved. These results are important as they show this correction protocol can be used to correct ring artifacts in old images. Dr. Holdsworth and his team are working on a protocol that will allow the computer to use a bright field collected in the manner described in the theory section to reduce ring artifacts in new scans. Dr. Holdsworth has met with representatives from the scanner manufacturer (Global
Product Manager and Global Applications Specialist, GE Healthcare Pre-Clinical Imaging). They are very interested in the results and are planning to use this protocol at some of their other customer sites in the United States. Prell et al used a different method of correcting ring artifacts. Their correction uses post processing software to determine the location and thickness of the ring artifacts and uses simple image processing for correction. They tested their protocol in polar and Cartesian coordinates. This achieved about 20 percent reduction in the ring artifacts (Prell et al, 2009). The correction presented here is much more effective, as it corrects at the source of the noise, rather than in post processing. To improve the results of this experiment further, more experiments should be conducted. One area would be to determine if averaging the last 100 images in the bright field is optimal. Perhaps the last 50 or 150 would produce better results. This protocol could also be tested on images older than a year to determine how old of an image can be corrected. These investigations will improve value and may improve the performance of this bright field correction protocol. Conclusion The purpose of this experiment was to determine if ring artifacts in micro CT images up to a year old could be corrected using a bright field calibration. It was demonstrated that the standard deviation due to the ring artifacts was significantly reduced in all regions of interest in each image. These results prove the validity of this correction protocol.
References GE Healthcare. "Computed Tomography." Web. 14 Mar. 2011. <https://www2.gehealthcare.com/portal/site/usen/menuitem.f76842a5b0 610162d6354a1074c84130/?vgnextoid=79bda52fcea2d110VgnVCM100000 258c1403RCRD> Kinney, J. H., Johnson, Q. C., Nichols, M. C., Bonse, U., Saroyan, R. A., Nusshardt, R., & Pahl, R. (1989). X ray microtomography on beamline X at SSRL. Rev. Sci. Instrum, 60 (7), 2471 2474. Kyriakou, Y., Prell, D., & Kalender, W. A. (2009). Ring artifact correction for high resolution micro CT. Physics in Medicine and Biology, 54, 385 391. Prell, D., Kyriakou, Y. & Kalender, W. A. (2009). Comparison of ring artifact correction methods for flat detector CT. Physics in Medicine and Biology, 54, 3881 3895. Sijbers, J. & Postnov, A. (2004). Reduction of ring artifacts in high resolution micro-ct reconstructions. Physics in Medicine and Biology, 49, 1 8. Teeter, M. G., Naudie, D. R., Charron, K. D. & Holdsworth, D. W. (2010). Three Dimensional Surface Deviation Maps for Analysis of Retrieved Polyethylene Acetabular Liners Using Micro Computed Tomography. The Journal of Arthroplasty, 25 (2), 330 332.