Spectral CT with a Sparsely Sampled Silicon Strip - I

A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Spectral CT with a
Sparsely Sampled Silicon Strip
Photon Counting Detector
A. Sisniega1
W.
J. W. Stayman1
J. Xu1, K. Taguchi2
E. Fredenberg3, M. Lundqvist3
J. H. Siewerdsen1,2
Zbijewski1,
1Dept
.of Biomedical Engineering, Johns Hopkins University
2Dept. of Radiology, Johns Hopkins University
3Philips Women’s Healthcare, Solna, Sweden
Acknowledgments
The I-STAR Lab
Imaging for Surgery, Therapy, and Radiology
http://istar.jhu.edu
Philips Healthcare
Erik Fredenberg, Karl Berggren
Funding Support
NIH-AR-R21-062293
NIH-CA-2R01-112163
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Sparse Sampling in CT
Motivations for sparse sampling
Dose or time reduction: repeat scans, near-real-time imaging
Technical limitations: detector design, mechanical constraints
Types of sparse sampling
Sparse view sampling*: IGRT, C-arm
Short scan protocols*: C-arm, application-specific scanners
Detector gaps: Tiled and sparse detectors (CMOS, PCDs)
Reconstruction gaps / Sparse angular sampling
Conjugate rays
Sinogram in-painting
Consistency conditions
Model-based reconstruction (MBR)
Compressive sensing and related approaches
Advanced regularization
XI-Europe
*J. Bian et al., PMB, 2010
Medpix2
Si-Strip Photon Counting CBCT
Silicon strip PCD
Philips MicroDose Si-strip PCD
Mature technology
High-quality - fast charge transport
Low absorption – edge-on
For example: MicroDose (Philips)
Characterization
70 kVp (+ 2 mm Al + 0.2 mm Cu)
0.025 mGy/mAs (in-air)
Modest energy resolution
No pulse pileup within the exposure range
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Philips
Threshold=100
70 kVp, coincidence det. OFF
Counts #
Scanning-slot mammography
Pre-collimator matched to PCD array (not shown)
Two thresholds (DE)
Coincidence detection in adjacent pixels pairs
Minimizes charge sharing effects
Edge-on creates large gaps in projection
70 kVp, coincidence det. ON
0 0.015 0.03 0.045 0.06 0.075 0.09 0.105
mAs/frame
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Sparse Sampling in CBCT
Philips MicroDose
Very sparse sampling in the projection domain
Array of Si-strip detectors: 6 – 7 arrays per row
Gap between detectors / rows: 2 – 5 mm
Pixel size: 0.05 mm x 0.5 mm
Detector size: 250 mm x 50 mm
Overall fill factor: ~25 %
Precollimator matched to detector matrix minimizes unnecessary patient dose
vFOV = 5 cm
Philips Microdose
Continuous detector
uFOV = 25 cm
uFOV = 25 cm
Scanning Trajectories
Simulation Study
Soft-tissue phantom with low contrast spherical inserts
Inserts positioned randomly without overlap
Insert diameter: 1 – 5 mm
Scanning Trajectories
Axial Scanning
Helical Scanning
Axial Scanning
Pitch
1 – 10 Stacked circular trajectories
Helical scanning
2 helical rotations
Pitch range 0.065 (2.5 mm) – 0.65 (25.2 mm)
Dz
Dz
Evaluation metrics
Sampling density
Number of samples per voxel
Sampling uniformity
Image fidelity
Structural similarity index matrix: 𝑆𝑆𝐼𝑀 𝑥, 𝑦 =
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
(2𝜇𝑥𝜇𝑦 𝐶1)(2𝜎𝑥𝑦 +𝐶2)
+
2 + 𝐶 )(𝜎 2 + 𝜎 2 + 𝐶 )
+ 𝜇𝑦
𝑥
𝑦
1
2
(𝜇𝑥2
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Image Reconstruction in Sparse Sampling
Penalized likelihood (PL)
𝜇 = argmax 𝐿 𝜇; 𝑦 − 𝛽 ∙ 𝑅(𝜇)
𝜇
System Model (Likelihood)
Smoothness Penalty (Regularization)
Projection noise model
Geometry + gaps
“Optimal” use of projection samples
Accommodates complex sampling patterns
Sensors
Assumptions on image smoothness
Huber penalty
Dominates in projection gaps
Penalty map
Penalty Strength
(1) Constant across volume
(2) Spatially varying
Certainty-based*
*Fessler & Rogers, TIP, 1996
Experimental Setup
Data acquisition
40 kVp / 70 kVp + 1.6 mm Al, 5 mA
Axial scanning: 360 projections / rotation
Helical scanning: 720 projections (720o)
Silicon-strip
detector
X-ray source
u
SDD = 65 cm
v
Evaluation of image quality
Soft-tissue simulating phantom
Gelatin background
Spherical tissue-equivalent inserts
Anthropomorphic phantom
Natural skeleton (hand / wrist)
Soft-tissue equivalent plastic
Tissue equivalent inserts
Bone, blood, adipose
1.5 – 12 mm
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Helical
stage
Lateral
stage
Gelatin
Background
Polyethylene inserts
Random positions
2 – 12 mm diameter
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Volumetric Spectral Photon Counting CT
Iodine
Three material decomposition
(5 mg/mL)
Water – Bone – Iodine
𝝁𝑳𝑬
𝒘𝒂𝒕𝒆𝒓
𝝁𝑯𝑬
𝒘𝒂𝒕𝒆𝒓
𝟏
𝝁𝑳𝑬
𝑰𝒐𝒅𝒊𝒏𝒆
𝝁𝑯𝑬
𝑰𝒐𝒅𝒊𝒏𝒆
𝟏
𝝁𝑳𝑬
𝒃𝒐𝒏𝒆
𝝁𝑯𝑬
𝒃𝒐𝒏𝒆
𝟏
𝒇𝒘𝒂𝒕𝒆𝒓
𝝁𝑳𝑬
𝒇𝑰𝒐𝒅𝒊𝒏𝒆 = 𝝁𝑯𝑬
𝒇𝒃𝒐𝒏𝒆
𝟏
Bone
(100 mg/mL)
PMMA
Iodine
(10 mg/mL)
Calibration and evaluation
Bone
Evaluation of threshold position valid range
Inter-pixel threshold position calibration
Selection of optimal threshold position
Maximum separation between materials
(50 mg/mL)
Evaluation
Accuracy of concentration (RMSE)
Material classification
Iodine
(5 mg/mL)
Water
Results: Axial Scanning
3 Rotations
1 Rotation
Sampling
Sampling
Reconstruction
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Reconstruction
6 Rotations
Sampling
Reconstruction
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Results: Helical Scanning
Pitch 0.05 / 2.5 mm
Pitch 0.22 / 10.8 mm
Sampling
Sampling
Reconstruction
Pitch 0.5 / 25.2 mm
Sampling
Reconstruction
Reconstruction
Scanning Trajectories: Axial vs Helical
Sampling density
Helical
Normalized voxel samples
Normalized voxel samples
Axial
Number of stacked rotations
Helical Pitch
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Scanning Trajectories: Axial vs Helical
Sampling density
Axial
Normalized voxel samples
Normalized voxel samples
Helical
Number of stacked rotations
Helical Pitch
Scanning Trajectories: Axial vs Helical
Sampling density
Axial
Normalized voxel samples
Deviation from uniform sampling
Helical
Number of stacked rotations
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Helical Pitch
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Effect of Penalty on Image Fidelity
Constant b
Spatially varying penalty
Sagittal
b=3
Coronal
Sagittal
b = 30
Coronal
Sagittal
b = 30
Sagittal
Sagittal
b = 60
Coronal
Sagittal
b = 60
Coronal
SSIM
Sagittal
b = 120
Coronal
Effect of Penalty on Image Fidelity
Constant b
Spatially varying penalty
Sagittal
b=3
b = 30
Coronal
SSIM
SSIM
Coronal
Sagittal
Sagittal
b = 30
Sagittal
Sagittal
b = 60
Sagittal
b = 60
Coronal
b
b
Coronal
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Sagittal
b = 120
Coronal
SSIM
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Volumetric Photon Counting CT
Gelatin phantom - Helical pitch = 0.28 (10.8 mm)
Spatially varying penalty
b = 100
b = 175
b = 500
b = 1000
mm-1
Volumetric Photon Counting CT
Gelatin phantom - Helical pitch = 0.28 (10.8 mm)
Spatially varying penalty
b = 100
b = 175
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
b = 500
b = 1000
mm-1
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Volumetric Photon Counting CT
Wrist - Helical pitch = 0.28 (10.8 mm)
Axial
Sagittal
Coronal
Spectral CT Calibration
Material separation as a function of threshold position
Iodine
a
Bone
Soft tissue
m low bin (mm-1)
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
Low Threshold = 60
a (deg)
m high bin (mm-1)
ROIs
Low Threshold = 100
High Energy threshold
mm-1
A. Sisniega et al. (presented at SPIE Medical Imaging 2015)
Volumetric Spectral Photon Counting CT
Material decomposition
Calibration
b = 100
0.62
0.03
b = 50
b = 100
b = 500
0.0
0.93
0.001
0.01
0.5
0.002
0.0
1.0
0.001 0.006
0.20
0.001
0.0
0.20
0.003 0.03
0.0
0.001
0.96
0.02
0.49
0.002
0.99
0.008
g/mL 0
0.12 0 0.01
Bone Iodine
0.0
0.04
0.48
0.002
Bone Iodine
0.48
0.001
1.0
0.006
0.0
0.05
1.0
0.006 0.61
0.001
0.73
0.001
Bone Iodine
1.0
0.008
Bone Iodine
Conclusions
Acquisition trajectories
Sparse detector matrix
Multi-axial and helical trajectories investigated to improve sampling
Helical trajectories yielded superior sampling density (for equal # projections)
Image reconstruction
Model based reconstruction successfully compensated for sparse sampling
Regularization strength affects image fidelity
Spatially varying penalty strength further improved fidelity and helped to
overcome irregularity in sampling density
Dual-energy decomposition
Three-material decomposition achieved for water, bone, and iodine
Careful selection of PCD threshold parameter is required for proper material
separation
I-ISTAR Lab, Johns Hopkins University
(www.jhu.edu/istar)
1.0
0.003
0.71
0.001