Learning Objectives Optimization of Image Acquisition and Reconstruction in Multi-slice CT • Clinical implementations of reconstruction algorithms on multimulti-slice CT scanners. Lifeng Yu, Cynthia H. McCollough, Shuai Leng, James M. Kofler • How CT acquisition parameters affect image reconstruction and image quality. Department of Radiology, Mayo Clinic, Rochester • How reconstruction parameters affect image quality. From SingleSingle-slice to MultiMulti-slice CT Wide cone beam Narrow cone beam Fan beam I. Clinical Implementations of Image Reconstruction 1x5 mm 0.75 sec 1995 ConeCone-beam Effect 16o 320-slice CT 64x0.6 mm 0.33 sec 2004 >128x0.6 mm 0.27 sec 2008 - Present ConeCone-beam Effect ~40 mm <20 mm 4o 64-slice CT 16x0.75 mm 0.42 sec 2001 z 160 mm <2o 4, 8, 16-slice CT 4x1 mm 0.5 sec 1998 2D FBP 2D13 FBP cm off center AMPR AMPR 13 cm off center Courtesy D. Platten et al. ImPACT (RSNA 2003) 1 Analytical Reconstruction Methods Fan-beam, parallel- beam FBP, rebinning FBP, 180LI, 360LI FDK, generalized FDK, Wedge FBP PI, PI-Slant, ASSR, nPI, AMPR Various 3D weighted FBP or FDK-based approximate methods Analytical Katsevich’s exact method BPF Minimum data exact reconstruction Bontus et al, EnPiT: Filtered Back-Projection Algorithm for Helical CT Using an n-Pi Acquisition, TMI, 2005 Clinical Implementations of Image Reconstruction • All current analytical reconstructions on clinical scanners are approximate methods • Goal is to maintain a good balance between • computationally efficiency • accuracy • dose efficiency – fully utilize redundant data Vendor Variations on Analytical Reconstructions • GE – 3D Weighted FBP (Tang et al, PMB, 2006) • Philips – COBRA, nPI (Kohler et al, Med Phys, 2002; Bontus et al, TMI, 2005) • Siemens – AMPR (Flohr et al, Med Phys 2003) and 3D weighted FBP (Stierstorfer et al, PMB, 2004) • Toshiba – TCOT, modified FDK (Taguchi et al, Med Phys, 2003) What can users control on image reconstruction? II. Acquisition parameters • Scan mode: Axial, helical, cine, shuttle, ECGECG-gated • Rotation time (e.g., 0.28 sec, 0.35 sec, 0.5 sec) • X-ray beam filter (e.g., flat filter, bowtie filter) • X-ray beam collimation • Detector configuration (e.g., 64x0.625 mm, 24x1.2 mm, 128x0.6 mm) • Helical pitch (in helical mode); Table feed (in axial mode) • Tube potential: kV • Tube current: mA • mAs: mAs: Tube current (mA (mA)) x Rotation time (s) • Effective mAs: mAs: mAs/pitch mAs/pitch • Automatic exposure control (AEC) 2 Scan Mode Axial (or sequential) Helical Pitch Cine Continuous axial scan, no table move Helical (or spiral) Pitch = Shuttle or 4D spiral Non-cardiac spiral MDCT: pitch=1 Gantry rotation angle (deg) 270 180 90 1 2 3 Non-cardiac spiral MDCT: pitch=0.75 Pitch and Image Quality 270 • Higher pitch – higher scan speed • For nonnon-cardiac exams, a higher pitch 180 90 0 4 0 Gantry rotation angle (deg) 180 90 1 2 3 3 4 360 270 • Downside of higher pitch • Lower dose capacity • More severe helical artifacts • More severe conecone-beam artifacts If 90 combined with wide collimation 0 4 scan has a higher temporal resolution. (windmill) 180 0 z-position in units of the total nominal beam width 1 2 3 4 z-position in units of the total nominal beam width Primak A et al, RadGraphics, 2006 Detector Configuration GE 64 x 0.6 20 mm Sens-16 4 x 1.5 4 x 1.2 Def or 64 4 x 1.2 20 mm Z-flying focal spot SIEMENS AS+ or Flash LS-8 8x1.25 or 8x2.5 LS-16 4 x 1.25 LS-64 4 x 1.25 16 x 0.625 0 2 Non-cardiac spiral MDCT: pitch=1.5 270 64 x 0.625 Gantry rotation angle (deg) Non-cardiac spiral MDCT: pitch=0.5 360 0 1 z-position in units of the total nominal beam width z-position in units of the total nominal beam width 16 x 0.75 0 Pitch > 1 4 x 1.5 0 Pitch = 1 360 32 x 0.6 Gantry rotation angle (deg) Pitch < 1 360 Table translation per rotation Detector collimation 24 mm 28.8 mm 40 mm 38.4 mm Flohr et al, Image reconstruction and image quality evaluation for a 64-slice CT scanner with z-flying focal spot. Med Phys 32, 2536, 2005. 3 Effect of Z FFS NO Z FFS Detector configuration and electronic noise With Z FFS Flohr et al, Image reconstruction and image quality evaluation for a 64-slice CT scanner with z-flying focal spot. Med Phys 32, 2536, 2005. Wider detector collimation • Routinely used for faster speed and better use of tube power (64x0.625 mm, 128x0.6* mm, …) 128 x 0.6* mm Equal CTDIvol 32 x 1.2 mm ConeCone-beam Artifacts in Wider Collimation (4 cm) • Tends to generate more severe artifacts if not wellwellcorrected • Helical and conecone-beam artifacts, particularly for highhighcontrast object at high pitch • Some applications, narrower collimation is better • e.g., 32x0.625 mm is often a better choice for head CT than 64x0.625 mm • Use slower pitch for wide collimation if there is no scan time concern. Pitch=0.983 Pitch=0.516 Narrower Collimation (2 cm) Pitch=1.375 X-ray Beam Bowtie Filter X-ray source bowtie filter flat filter • Small bowtie • Ped Head, Cardiac Small, Small Head, Ped Body, Small Body • Medium bowtie • Head, Cardiac Medium, Medium Body Patient Pitch=0.562 Pitch=0.969 • Large bowtie • Large body, Cardiac Large Pitch=1.375 Detector 4 mA, mA, mAs, mAs, effective mAs, mAs, mAs/slice mAs/slice • mAs: mAs: Tube current (mA (mA)) x Rotation time (s) • Effective mAs = mAs/pitch mAs/pitch • Effective mAs is proportional to the radiation output (usually expressed as CTDIvol) • mA or mAs cannot be related to CTDIvol without knowing the pitch Effective mAs 70 Effective mAs 450 mAs vs. Noise in Cardiac and NonNoncardiac Exams • In nonnon-cardiac exams, image noise is directly Noise: 22.3 240 effective mAs Noise: 25.3 180 effective mAs related to effective mAs (mAs/pitch). mAs/pitch). • Redundant data is used when pitch is less than 1 • In cardiac exams, image noise is directly related to mAs. mAs. • Independent of Pitch • Each reconstruction uses data from a halfhalf-scan range. • No redundant data is used even when pitch is much less than 1. Noise: 30.8 120 effective mAs Noise: 46.4 60 effective mAs Photon Starvation B10 Adaptive Filter in Reconstruction B18 5 Automatic Exposure Control (AEC) X-ray attenuation • Varies over body region and with projection angle • Image noise is primarily determined by noisiest projections (thick body parts) • More photons (dose) to thinner body parts is unnecessary radiation dose Automatic tube current modulation Automatic Tube Current Modulation • Goal of tube current modulation is to dynamically adjust tube current according to the attenuation level. ⎛ A(d ) ⎞ ⎟⎟ General strategy I (d ) = I (d 0 ) ⋅ ⎜⎜ ⎝ A(d 0 ) ⎠ x=0.5 x=1 – Higher attenuation, higher tube current – Lower attenuation, lower tube current x=0.4 x=0.3 x x=0 A( d ) = exp( µd ) where d0 is the reference attenuation path length. x is the modulation strength If If x=0 x>0 x =1 then then I ( d ) = I (d 0 ) Constant tube current Attenuation decreases tube current is modulated according to attenuation Attenuation increases equal noise for each view independent of attenuation M. Gies, W. A. Kalender, H. Wolf and C. Suess, "Dose reduction in CT by anatomically adapted tube current modulation. I. Simulation studies," Med Phys. 26, 2235-47 (1999) Topogram Evaluation: a.p. and lateral Optimal mA: a.p. and lateral 4000 a.p. (measured) a.p. 3500 lateral (calculated) max 2000 1500 1000 atten u atio n I_0 / I 2500 tub e c u rr e nt in m A lateral 3000 500 0 600 500 400 300 table position 200 100 0 600 500 400 300 200 100 0 table position 6 402 mA 544 mA On-line tube current modulation 400 4000 Attenuation 3500 tube current 300 3000 250 2500 200 2000 150 1500 100 1000 50 500 0 attenuation I_0 / I tube current 350 0 600 500 400 300 200 table position in m m 100 0 678 mA Streaking Artifact Reduction Different Versions of AEC Manufacturer Constant Tube Current AEC 452 mA AEC Trade name Image Quality reference Goal GE Auto mA, Smart mA Noise Index Maintain a constant noise level (defined in noise index), using tube currents within prescribed minimum and maximum values. Toshiba SureExposure Standard deviation (high quality, standard, low dose) Maintain a constant noise level (defined in standard deviation values for each protocol), using tube currents within preset minimum and maximum values. Siemens CARE Dose4D Quality Reference mAs Maintain the same image quality (varying noise target for different attenuation level) with reference to a target effective mAs level for a standard-sized patient. Philips DoseRight Reference Image Keep the same image quality as in the saved reference image. 3.00E+05 80 kVp 100 kVp 120 kVp 140 kVp Tube Potential (kV) x-ray intensity 2.50E+05 2.00E+05 1.50E+05 1.00E+05 5.00E+04 0.00E+00 0 50 100 150 En ergy (keV) 7 Iodine Contrast LowerLower-kV Benefit – Increased Iodine Contrast Iodine Contrast vs. kVp 120 kV, CTDIvol=5.18 mGy 300 100 kV, CTDIvol=3.98 mGy 250 Contrast 200 Small Medium 150 Large 100 50 0 60 80 100 120 140 The same patient scanned with a protocol at 120 kV and a protocol at 100 kV. Note that improved contrast and visualization of mural stratification of the 100 kV image despite a 25% radiation dose reduction. 160 kVp Phantom size: small 30x13 cm, medium 30x21 cm, large, 36x27 cm Limitation of Lower kV – increased noise Noise 80kV CTDIvol=6.4 mGy Noise vs. kVp (CTDIvol=23mGy) 120kV CTDIvol=6.5 mGy 45 40 35 Noise (HU) 30 Small 25 Medium 20 Large 15 10 5 0 60 80 100 120 140 160 kVp Optimal kV • Optimal kV is the kV that uses the minimum radiation dose to achieve the desired image quality • Dependent on patient size • Dependent on diagnostic task A General Strategy for Optimal kV Selection • Iodine CNR with a noise constraint α: CNR (Ω d , kV , D ) ≥ CNR (Ω d , kVref , Dref ) and σ (Ω d , kV , D ) ≤ α ⋅ σ (Ω d , kVref , Dref ) • The corresponding relative dose factor (RDF) at each kV is given by ⎧ C (Ω d , kVref ) 1 ⎫ k (Ω d , kV ) RDF (Ω d , kV ) = max ⎨ , ⎬⋅ ⎩ C (Ω d , kV ) α ⎭ k (Ω d , kVref ) L. Yu, H. Li, J. Fletcher, C. McCollough, Automatic selection of tube potential for radiation dose reduction in CT: A general strategy. Med Phys, 37:234-43, 2010. 8 Optimal kV for Abdominal CT Modulation strength 25 cm 30 cm 35 cm 40 cm 45 cm 50 cm 55 cm Very weak 120 120 120 120 120 120 140 Weak 100 100 100 120 120 120 140 average 100 100 100 100 120 120 140 Strong 80 80 100 100 100 120 140 Very strong 80 80 80 100 100 120 140 • • • • • Very weak: weak: average: strong: Very strong: Routine nonnon-contrast exams Liver/pancreas Routine contrastcontrast-enhanced exams CT enterography, enterography, CTU, stone CT angiography Slice Thickness III. Reconstruction parameters Nominal width of image (z-direction) Slice sensitivity profile • Slice thickness/Increment • Reconstruction kernel • Reconstruction FOV • 3D reformat • Iterative reconstruction or noise (line spread function along Z) reduction options Slice Thickness, Noise, and Dose Slice Thickness – Partial Volume Image (mm): 10 Noise ∝ 5 2.5 1.25 3.84 5.89 7.82 1 # Photons Slice thickness (mm): 5 Relative Noise: 100% Required mAs: 100% (for = noise) 2.5 141% 200% 1.25 200% 400% 0.625 283% 800% Better z-resolution (less partial vol. averaging) Increased image noise Requires more radiation dose to get the same noise level Noise (HU): 2.93 All other parameters are identical Thinner slices => less partial volume effect 9 Slice Thickness – Partial Volume Slice Thickness – Partial Volume 10mm image thickness 5mm image thickness All other parameters are identical All other parameters are identical Slice Thickness – Partial Volume Slice Thickness – Partial Volume 2mm image thickness 1mm image thickness All other parameters are identical All other parameters are identical Slice Thickness – Partial Volume Reconstruction kernel • Reconstruction kernel (filter or algorithm) has a significant impact on spatial frequency and noise characteristics of an image • Smooth kernels reduce high spatial-frequency information and image noise • Sharp kernels increase high spatial-frequency information and image noise 0.6mm image thickness All other parameters are identical 10 Impact of Reconstruction Kernel on In-plane spatial resolution Modulation Transfer Function (MTF) B10 B20 B30 B40 1 MTF 0.8 0.6 0.4 0.2 0 0 A smoother kernel A sharper kernel Soft CT number: 89.7±2.2 2 4 6 8 10 spatial frequency (1/cm) Standard CT number: 89.7±2.6 Detail Bone CT number: 90.7±3.5 CT number: 90.6±8.9 11 Some Considerations on GE Reconstruction algorithms Soft • Lung and bone plus alter CT # values • For high res lung use bone • Standard is most frequently used • Soft is to recover a really grainy exam • Use Full recon for getting the slice width you Standard Detail Lung Bone ask for • Plus mode can help smooth noisy images or Edge Bone Plus Siemens Reconstruction Kernels • B10 Æ B90 Body • H10 Æ H90 Head • U30 Æ U90 Ultra High Resolution • T20 Æ T81 Topogram • Lower number smoother • Higher number sharper • Multiples of 10 are the “basic” basic” kernels • In between values are “special” special” kernels reduce some artifacts, but is up to 20% wider slice thickness Reconstruction FOV & Pixel Size 0.78 mm 0.78 mm 40 cm 40 cm 0.39 mm 0.39 mm 20 cm 20 cm Pixel size & System Spatial Resolution • Pixel size should not be confused with system spatial resolution • Pixel size should be small enough in order not to limit the display display of 3D Reformat system resolution. • 512 x 512 matrix Pixel Size (mm) Pixel size Resolution (lp/cm) 50 1.0 5.1 40 0.8 6.4 30 0.6 8.5 20 0.4 12.8 rFOV (cm) System spatial Resolution at 10% MTF (lp/cm) B20 5.5 B30 B40 6.2 7.0 Reformat from 1 mm axial reconstruction Reformat from 2 mm axial reconstruction 12 Iterative Reconstruction • Benefits over analytical reconstruction • Accurate physical models (imaging system geometry, Vendor implementations of Iterative Reconstruction or noise reduction methods • GE: ASIR (adaptive statistical iterative reconstruction), MBIR (model based iterative scattering, spectrum, etc.) • less bias, improved quantitative accuracy, improved spatial resolution • Appropriate statistical models (FBP treats all data equally) • Lower image noise, lower radiation dose • Object constraints (e.g., nonnon-negativity) • Flexible to handle nonnon-standard scanning configurations • Potential to reconstruct images from sparse or incomplete data reconstruction) • Phillips: iDose • Siemens: IRIS (iterative reconstruction in image space), SAFIRE (sinogram affirmed iterative reconstruction) • Toshiba: AIDR3 (adaptive iterative dose reduction) Source - SPIE Medical Imaging Symposium, 2011 Example – Image quality improvement Standard FBP Noise Texture Change MBIR Images provided by Rendon Nelson, MD, Duke University, to AuntMinnie.com (“MBIR aims to outshine ASIR for sharpness, CT dose reduction,” May 18, 2010) LowLow-contrast Detectability 50% dose reduction possible? Hara, AK et al (2009). "Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study." AJR Am J Roentgenol 193(3): 764-71. Special Considerations in Iterative Reconstruction • Potential for dose reduction ~ 20% - 50% - task dependent • HighHigh-contrast allows more • LowLow-contrast allows less or none • Noise texture could change dramatically • NonNon-linear regularization term leads to nonnon-linear noisenoiseresolution tradeoff • Spatial resolution is highly dependent on contrast level • Could reduce noise substantially without sacrificing highhighcontrast spatial resolution (measured in MTF) • LowLow-contrast detectability needs to be evaluated more carefully. • Quantitative evaluation and optimization of iterative FBP, 25 mGy ASIR, 12.5 mGy reconstruction is still an active research field Hara, AK et al (2009). "Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study." AJR Am J Roentgenol 193(3): 764-71. 13 Summary • Implementations of reconstruction algorithms on clinical scanners • How data acquisition parameters affect image quality • How reconstruction parameters affect image quality • Understanding of these is essential for designing dosedose- Thank you! http://mayoresearch.mayo.edu/ctcic http://mayoresearch.mayo.edu/ctcic// efficient and quality scanning protocols for various diagnostic tasks. 14