ML reconstruction for CT derivation of MLTR rigid motion correction resolution modeling polychromatic ML model dual energy ML model Bruno De Man, Katrien Van Slambrouck, Maarten Depypere, Frederik Maes, Jung-ha Kim, Roger Fulton, Johan Nuyts MIRC, KU Leuven & Univ of Sydney 1 Tomography 2 1
CT data recon 3 Maximum Likelihood one wishes to find recon that maximizes p(recon data) data recon computing p(recon data) computing p(data recon) difficult inverse problem easy forward problem Bayes: p(recon data) = ~ p(data recon) p(recon) p(data) 4 2
Maximum Likelihood recon p(recon data) ~ p(data recon) data projection Poisson µ j p(data recon) j = 1..J i = 1..I ln(p(data recon)) = ~ L(data recon) = 5 Maximum Likelihood L(data recon) find recon: Iterative inversion needed 6 3
MLTR Likelihood L(µ+Δµ) T 1 (µ, Δµ) µ 7 MLTR Likelihood L(µ+Δµ) T 2 (µ, Δµ) T 1 (µ, Δµ) µ 8 4
MLTR Likelihood L(µ+Δµ) T 2 (µ, Δµ) T 1 (µ, Δµ) µ 9 MLTR Likelihood L(µ+Δµ) T 2 (µ, Δµ) T 1 (µ, Δµ) 10 5
MLTR 11 MLTR MEASUREMENT COMPARE UPDATE RECON REPROJECTION 12 6
MLTR FBP MLTR 13 MLTR metal artifact reduction projection truncation FBP FBP MLTR MLTR 7
ML reconstruction for CT derivation of MLTR rigid motion correction resolution modeling polychromatic ML model dual energy ML model 15 MLTR for rigid motion correction 1) validation Siemens Sensation 16 Siemens MLTR J-H Kim, Z Kuncic, R Fulton, J Nuyts 16 8
MLTR for rigid motion correction 2) simulation motion rotations: in transaxial plane in sagittal plane in coronal plane 5 s measured rat motion software phantom trans cor CT protocol sag proj translations: along column along row along plane high pitch narrow collimation low tube current high rotation speed low dose 17 MLTR for rigid motion correction relativity : assign inverse motion to CT MLTR modified to support stationary object rigid view-dependent displacement of CT detector-source assembly 18 9
MLTR for rigid motion correction pitch = 2 trans cor sag proj MLTR w/o correction MLTR with correction pitch = 0.5 MLTR w/o correction MLTR with correction 19 3) phantom measurement MLTR for rigid motion correction 20 10
MLTR for rigid motion correction 21 MLTR 22 11
MLTR 23 ML reconstruction for CT derivation of MLTR rigid motion correction resolution modeling polychromatic ML model dual energy ML model 24 12
MLTR 25 MLTR microct Ex vivo - global FBP - global FBP - adap1ve MAPTR global Recon Segment 26 13
ML reconstruction for CT derivation of MLTR rigid motion correction resolution modeling polychromatic ML model dual energy ML model 27 metal artifacts Double knee prosthesis Double hip prosthesis Dental fillings Cause of metal artifacts: Beam hardening Scatter (Non) linear partial volume effects Noise (Motion) Mouse bone and titanium screw (microct) 28 14
metal artifact reduction (MAR) Projection completion Initial filtered backprojection (FBP) reconstruction Segment the metals and project Remove metal projections for sinogram Interpolate (e.g. linear, polynomial, ) Reconstruct (FBP) and paste metal parts 29 Models for iterative reconstruction Poisson Likelihood: Update: Projection model: SKYSCAN SPECTRUM Black = without filter Blue = 0.5 mm Al and 0.038 mm Cu monochromatic: 1 material polychromatic: 30 15
Models for iterative reconstruction Full Polychromatic Model IMPACT SKYSCAN SPECTRUM Black = without filter Blue = 0.5 mm Al and 0.038 mm Cu 31 Models for iterative reconstruction Full Polychromatic Model IMPACT Base substances Material dependence Energy dependence 32 16
Models for iterative reconstruction Φ and θ (1/cm) Base substances µ mono (1/cm) 33 Local models IMPACT is complex and slow, MLTR and MLTR_C are simpler and faster Find the metals PATCH 3 Define patches IMPACT in metals MLTR_C elsewhere PATCH 2 PATCH 1 34 17
simulations Geometry based on Siemens Sensation 16 Included: polychromatic spectrum detector, source and view subsampling afterglow crosstalk source view(k) = a*view(k-1) + (a-1)*view(k) detector 500 µs 35 results PMMA Al Fe 36 18
clinical CT (Siemens Sensation 16) Circular phantom PMMA Al Fe Siemens Sensation 16 (part of Biograph 16 PET/CT) 120 kv, 300 ma 2 x 1.00 mm Circular scan, 0.5 s per rotation (no flying focal spot) 2D reconstruction of 1 slice 37 clinical CT (Siemens Sensation 16) Body shaped phantom 38 19
clinical CT (Siemens Sensation 16) Body shaped phantom 39 SKYSCAN SPECTRUM Black = without filter Blue = 0.5 mm Al and 0.038 mm Cu FDK iterative reconstruction for microct IMPACT Ti-cage, culture of soft tissue and cartilage 40 20
ML reconstruction for CT derivation of MLTR rigid motion correction resolution modeling polychromatic ML model dual energy ML model 41 Dual energy CT Dual energy CT: exploits dependence of linear attenuation coefficient on photon energy to discriminate between materials. Dual energy CT has been widely used to discriminate bone from contrast agent. 42 21
Dual energy microct applications MicroCT: imaging bone and contrast agents in small animals, such as mice. Bone development and repair requires a normal vascular system to supply oxygen and nutrients. Rat skull Detail of trabecular bone structure Mouse bone fracture MicroCT Imaging X-ray energy range: 20 100 kev 43 Post-reconstruction: microct post-reconstruction dual energy for microct problems: Perfused mouse tibia E1: 56 minutes beam hardening due to dense materials contrast agent metal implants Noise. Signal-to-noise ratio is limited by In vivo microct: dose concerns Ex vivo microct: cumbersome long scan times Voxel by voxel comparison is sensitive to erroneous intensity values Noise robustness can be increased by incorporating a noise model resorting to statistical approaches 22
Polychromatic attenuation model Dual energy algorithms exploit the dependency of the linear attenuation coefficient µ on the photon energy E The attenuation can be modeled as a linear combination of b basis functions A well known combination of basis functions is the Compton scatter and the photoelectric effect. Water Bone 45 IMPACT extension to dual energy microct Iodine Iodine Our model consists of a third basis function that models the attenuation of a single contrast material (barium, iodine, lead): 46 23
Results Noiseless simulation Noiseless simulation of water, bone and 0.15 and 0.20 g/ml mixtures of barium sulfate Post reconstruction Beam hardening affects tissue decomposition Polychromatic model accounts for beam hardening Iterative Decomposition 0.1957 g/ml + 0.0024 (0.20) 0.1455 g/ml + 0.0021 (0.15) 47 IMPACT extension to dual energy microct Measurement of polypropene tube, water, bone equivalent material CaHA and a barium sulfate mixture Noisy Post reconstruction Coefficient of variation in BaSO4 region: 0.36 Coeffecient of variation in BaSO4 region: 0.15 IMPACT Decomposition 48 24
IMPACT extension to dual energy microct Measurement of a mouse bone perfused with barium sulfate Post reconstruction Barium fractions Iterative decomposition Barium coefficients Iterative decomposition Coloured overlay 49 thanks 50 25