Translating Protocols Between Scanner Manufacturer and Model Cynthia H. McCollough, PhD, FACR, FAAPM Professor of Radiologic Physics Director, CT Clinical Innovation Center Department of Radiology Mayo Clinic, Rochester MN
Disclosures NIH: Research Support: EB 079861 DK 083007 DK 059933 EB 004898 RR 018898 Siemens Healthcare Off Label Usage None
Disclosures Our team currently supports 25 CT systems. Presently they are from only two manufacturers (GE and Siemens) 10 distinct scanner models (all multi-slice) 12 distinct multi-slice models in our practice since 1998 Examples come from the systems I know best Am an equal opportunity critic, so if I know about weaknesses (or strengths) of other systems, I m happy to share those also.
The first thing you need is a dictionary
http://mayoresearch.mayo.edu/ctcic/educational-resources.cfm
Clarity, Transparency, and Uniformity
Collimation vs. Slice Width
The next thing you need is data When is a 5-mm not a 5-mm? Which pitch values give best images? Which collimations are most/least dose efficient? Which reconstruction algorithms/kernels have special features or alter CT number accuracy? Which bowtie is used for which scan modes? Which focal spot is used when? etc.
Measured width of 5 mm image Full vs. Plus reconstruction option
Full or Plus Mode Full mode will retain the prescribed slice thickness Plus mode will give you a thicker slice thickness than prescribed (about 20% thicker, e.g. a 5 mm becomes a 6 mm). Correspondingly, noise level will be about 10% lower.
Pitch vs. Image Quality In spiral CT, image noise is dependent on pitch To compensate, mas must be changed as pitch is changed Relationship is linear on some systems, but not all Relationship is different for cardiac reconstructions Noise is INDEPENDENT of pitch in cardiac CT Image width can be affected by pitch Windmill and cone beam artifacts affected by pitch Windmill artifacts discussed in talks by Gupta, Morin
Study to evaluate cone beam artifacts A thin-walled object with edges at an angle to the scan plane Rate of change of funnel shape is constant along the z-axis Scanned in air, the funnel has high contrast (~ 500 HU) Courtesy D. Platten et al. ImPACT (RSNA 2003)
Courtesy D. Platten et al. ImPACT (RSNA 2003) Example images Single slices through the funnel appear as rings MIP image of many slices results in a wider ring If perfect the images should be uniform
Cone-beam algorithm on and off Low pitch (0.5), Siemens Sensation 16 Standard Cone-beam (AMPR) Courtesy D. Platten et al. ImPACT (RSNA 2003)
Cone-beam algorithm on and off High pitch (1.5), Siemens Sensation 16 Standard Cone-beam (AMPR) Courtesy D. Platten et al. ImPACT (RSNA 2003)
Cone-beam algorithm on and off High pitch (1.5), Philips Mx8000 IDT Standard Cone-beam (COBRA) Courtesy D. Platten et al. ImPACT (RSNA 2003)
Cone-beam algorithm on and off High pitch (1.5), Toshiba Aquilion 16 Standard Cone-beam (TCOT) Courtesy D. Platten et al. ImPACT (RSNA 2003)
Cone-beam algorithm with pitch GE LightSpeed 16, cone-beam reconstruction always on 0.562 0.938 1.375 1.735 Courtesy D. Platten et al. ImPACT (RSNA 2003)
Clinical relevance Standard Courtesy D. Platten et al. ImPACT (RSNA 2003) Cone-beam
Inclined (60 ) Teflon rod High pitch (1.5), Siemens Sensation 16 Standard Cone-beam (AMPR) Standard 13 cm off center Cone-beam 13 cm (AMPR) off center Courtesy D. Platten et al. ImPACT (RSNA 2003)
Dose Efficiency vs. Collimation Siemens Sensation 16
GE Recon Algorithms Soft Standard Detail Lung Bone Edge Bone Plus
CT Number Accuracy Some edge-enhancing algorithms/kernels can alter CT numbers E.g. GE Lung and Bone Plus Boedeker et al. Emphysema: Effect of reconstruction algorithm on CT imaging measures. Radiology 2004 Zhang, McCollough, et al. Selection of Appropriate Computed Tomographic Image Reconstruction Algorithms for a Quantitative Multicenter Trial of Diffuse Lung Disease. JCAT 2008
Boone: Presampled MTF in CT (Med Phys 2000)
Siemens Recon Kernels B10 B90 Body H10 H90 Head U30 U90 Ultra High Resolution T20 T81 Topogram Lower number smoother Higher number sharper Multiples of 10 are the basic kernels In between values are special kernels
Siemens Recon Kernels B18 B20 B25/B26 - cardiac B30 B31 finer grain noise B35/36 - calcium B40 B41 finer grain noise B45 B46 - cardiac/lung B50 B70 H30 H31 finer grain noise H32 no PFO H37 GE like H40 H41 finer grain noise H42 no PFO H47 GE like H48 GE like but sharper U70
Special Body Kernels B25 and B26 are for cardiac with edge-preserving noise reduction. B35 and B36 are for Ca scoring without edge enhancement. B45 is intermediate sharpness between B40 and B50 (e.g. not very special ) B46 is designed specifically for accurate assessment of inside coronary stents with 3D edge preserving noise reduction techniques. B75 is comparable to B70 in sharpness but used 2D edge-preserving noise reduction
B10
B20
B25
B26
B30
B31
B35
B36
B40
B41
B45
B46
Noise and Noise Uniformity B31/41 is like B30/B40 but with finer grain noise and a milder edge enhancement. Noise more uniform over FOV.
Special Head Kernels H21, H31, H41 are like H20, H30, H40 but with finer grain noise and a milder edge enhancement. H22, H32, H42 don t include iterative beam hardening correction (PFO). Reconstruction speed is faster, but the reconstructed images may contain significant beam hardening artifacts. H37 is comparable to GE Soft H45 is intermediate sharpness between H40 and H50 H47 is comparable to GE Standard H48 is like H47 but a bit sharper
H10
H20
H21
H22
H30
H31
H32
H37
H40
H41
H42
H45
H47
H48
H50
H60
H70
Have I made your head spin yet?
GE bowtie and focal spot selection (once upon a time)
Moral of the story There are many good reasons to invoke special features and characteristics Manufacturers often tie these features to protocols where they make sense There are many ways to get these features when you don t want them or to not find them when you do Often users are not even educated about them Don t stop having good ideas and features But make them transparent so user knows what they do and when they are used
Lastly, you need deliverables To design or translate a protocol, you need to know what the final product needs to look like Scan time (total) and per image (temporal resolution) Slice width(s) and image plane(s) required The thinnest image width determines the detector configuration Coronals and sagitals require thinner collimation Image sharpness or smoothness Noise level Target anatomy/patient (pediatric, obese, cardiac, etc) Diagnostic reference level (CTDIvol)
Knowing terminology, operation, features (quirks) and performance of your system(s) You can translate across manufacturer (make) and model to yield the desired deliverables Usually not one way to accomplish the same results Usually not a lot of ways Evaluate options as quantitatively as possible on phantoms and then form a WIP prootcol for clinical evaluation/refinement Some differences between seemingly similar options can show up only in patients, where motion and specific diagnostic criteria (like noise texture or subtle enhancement of small structures) come into play
Routine Chest
Routine Chest
Routine Chest
Routine Chest
Routine Chest
Routine Chest
Routine Chest
Mayo CT Clinic Innovation Center and Dept. of Radiology J. Kofler, L. Yu, S. Leng, M. Bruesewitz, T. Vrieve http://mayoresearch.mayo.edu/ctcic