J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Task-Based Optimization of Source-Detector Orbits in Interventional Cone-beam CT J. Webster Stayman, Grace Gang, and Jeffrey Siewerdsen

Acknowledgements

AIAI Laboratory Advanced Imaging Algorithms and Instrumentation Lab aiai.jhu.edu [email protected] I-STAR Laboratory Imaging for Surgery, Therapy, and istar.jhu.edu [email protected]

Faculty and Scientists Clinicians Students Tharindu De Silva John Carey Qian Cao Grace Gang Gary Gallia Hao Dang Aswin Mathews A Jay Khanna Sarah Ouadah Amir Pourmorteza Martin Radvany Sureerat Reaungamornrat Jeffrey Siewerdsen Doug Reh Steven Tilley II Alejandro Sisniega Marc Sussman Ali Uneri Shiyu Xu Jennifer Xu Wojciech Zbijewski Thomas Yi

Funding NIH U01EB014964, NIH R21EB014964, NIH KL2TR001077, NIH R01CA112163 This work was supported, in part, by the above grants. The contents of this presentation are solely the responsibility of the authors and do not necessarily represent the official view of Johns Hopkins or the NIH.

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 1 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Task-Driven Interventional Imaging

Conventionally Ignored by Interventional Devices Conventional Interventional Imaging Intraoperative CT Preoperative Planning Diagnostic Flat-Panel Detector Image Data Imaging

Task-Driven Trajectory Task Traditional ? Definition X-ray Circular Source Trajectory Patient- and Task-Driven Prior Information Intraoperative CT about Patient and Task

Optimization Framework

Anatomical Patient Model Optimal Patient Imaging Parameters Volume Task (W*)

Imaging System Model Imaging Observer Parameters Data Image Imaging Task (W) Acquisition Formation Model Performance Performance

Adjust Imaging Parameters for Increased Performance

G. Gang, J. W. Stayman, T. Ehtiati, J. H. Siewerdsen, “Task-driven image acquisition and reconstruction in cone-beam CT,” Physics in Medicine and Biology, 60 3129-3150 (March 2015).

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 2 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Performance Prediction for Penalized-Likelihood Reconstruction

Detectability Index – Non-prewhitening observer: Spatial Resolution

Noise Imaging Task Spatial Resolution (MTF) 1 2 3 Consider local Fourier approximation of MTF and NPS:

T

ADAye Theory F     j  Empirical MTF  j T F ADARy ej  ej 

T Noise Power Spectrum (NPS) F ADAye j  NPS j  2 T F ADARye jj e  yb DA exp ,  

J A Fessler and W Leslie Rogers, “Spatial resolution properties of penalized-likelihood image reconstruction methods: Space-invariant tomographs,” IEEE Trans. Im. Proc., 5(9):1346-58, Sep. 1996.

Performance Prediction as an Acquisition Design Objective

Detectability Index – Non-prewhitening observer: Spatial Resolution

Noise Imaging Task Acquisition Design Objective:

Consider local Fourier approximation of MTF and NPS: ˆ ˆ 2 ,   arg max d ';, WTask ,   ,

J. W. Stayman and J. H. Siewerdsen, “Task-Based Trajectories in Iteratively Reconstructed Interventional Cone-Beam CT,” Int'l Mtg. Fully 3D Image Recon. in Radiology and Nuc. Med., Lake Tahoe, (June 16-21, 2013).

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 3 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Orbit Parameterization / Optimization Orbit specified by a low-dimensional parameterization:

Single location, single task optimization: ˆ ˆ ˆ 2

,  arg max d ';,   WTask 

,

200 150 100 50 200 150 100 50

0.047 0.047

Multiple-location and/or multiple-task optimization: 500

450 0.048

ˆ ˆ ˆ 2 20.048 2 ,  arg max mind ' ,,,  ;WWW ,200 d '   ; ,..., d '    ;

    12     Task(1)      Task (2)  L      Task ( L ) 

, 400

0.049 0.049

Solve using a nonlinear, nonconvex optimization strategy: 350

0.05 0.05 150

CMA-ES (Covariance Matrix Adaptation Evolution Strategy) 300

0.051 0.051 Hansen N, Müller SD, Koumoutsakos P (2003). Reducing the time complexity of the derandomized evolution strategy with covariance 250

matrix adaptation (CMA-ES). Evolutionary Computation, 11(1) pp. 1–18. 100

200

0.052 0.052

150

0.053 0.053

50 100

0.054 0.054

50

0.055 0.055

200 150 100 50 200 150 100 50

0.047 0.047

500

450

0.048 0.048

200

400

0.049 Optimization for a Simple0.049 Object

350

0.05 0.05 Location #1 150

Orbit #1 300 Object:

0.051 0.051

250 10 cm cylinder,  = 0.05 mm-1

100

200 0.052

Optimization:0.052

150

0.053

90.053 orbital bases

50 100

0.054  q Iter #50(1/1) Time:2.6e+03 s -500.054 ° ≤ ≤ 50°, 0° ≤ ≤ 360°

0.055 50 CMA 0.055-ES (pop=40)

50

0.055 0.055

Location #1 0.054 0.054 50 s 100 Time:2.6e+03 #50(1/1) Iter 0.053 0.053 150

0.052 0.052 100 Fluence200 of rays through Location #1 0.051 250 0.051 40 300 150 0.05 0.05 20  350 0.049 0.049 0 400 200 (degrees) 0.048 -20 0.048 450 -40 0.047 500 0.047 50 100 150 200 50 100 150 200 0.055 0 100 200 3000.055

50 0.054 q (degrees) 0.054

100 50 0.053 0.053 150

0.052 0.052 200 100

250 0.051 0.051

300 0.05 150 0.05

350 0.049 0.049 AIAI Laboratory (aiai.jhu.edu)400 and 200 0.048 0.048 I-STAR Laboratory (istar.jhu.edu),450 500 0.047 0.047 Dept of Biomedical Engineering,50 100 150 200 50 100 150 200 Johns Hopkins University 4

200 150 100 50 200 150 100 50

200 150 100 50 200 150 100 50

0.047 0.047 0.047 0.047

500 500

450

450

0.048 0.048 0.048 0.048

200 200

400 400

0.049 0.049 0.049 0.049

350 350

0.05 0.05 0.05 0.05 150

150

300 300

0.051 0.051 0.051 0.051 250 250

100 100

200

200

0.052 0.052 0.052 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-0.052 June 4, 2015)

150 150

0.053 0.053 0.053 0.053

50 50 100 100

0.054 0.054 0.054 0.054

50 50

0.055 0.055 0.055 0.055

200 150 100 50 200 150 100 50

200 150 100 50 200 150 100 50

0.047 0.047 0.047 0.047

500

Iter #50(1/1) Time:2.6e+03 s 500 Iter #50(1/1) Time:2.7e+03 s

0.055 0.055 0.055 0.055

450

450

0.048 0.048 0.048

50 0.048 50

0.054 0.054 0.054 0.054 200 200

400

50 50100 400 100

0.049 0.049 0.049 Simple Object – Location0.049 -dependence 0.053 0.053 0.053 0.053

150 350 150 350

0.052 0.052 0.052 0.052

0.05 0.05

0.05 0.05

200 200 150

100 100 Location #1 150 300 300

Orbit0.051 #1 250 0.051 0.051 250Orbit #2 0.051

0.051 0.051 0.051 0.051 250

300 250 300

150 0.05 150 0.05 0.05 0.05

100 100

200

350 200 350 0.052 0.052 0.052

0.049 0.049 0.049 0.052 0.049

400 400

150

200 200 150

0.053 0.053 0.053

0.048 0.048 0.048 0.053 0.048

450 450

50 50 100 100

0.047 500 0.047 0.047 500 0.047

0.054 0.054

50 100 150 200 50 100 150 200 0.054

50 100 150 200 Iter #50(1/1)50 100 Time:2.6e+03150 200 s 0.054 Iter #50(1/1) Time:2.7e+03 s

50 50 0.055 0.055 0.055 0.055 0.055 Location 0.055#2 0.055 0.055

50 50 50 50

0.055 0.055 0.055

0.055 0.054 0.054

0.054 Location #1 0.054Location #2 0.0540.054 Location #1 0.054 Location #2 0.054

Iter #50(1/1) Time:2.7e+03 s Time:2.7e+03 #50(1/1) Iter 100 50 50 100 s 50 100 Time:2.6e+03 #50(1/1) Iter 50 100 0.053 0.053 0.0530.053 0.053 0.053 0.053 0.053 150 150 150 150

0.052 0.0520.052 0.052 0.052 0.052 200 0.052 200 0.052 200 100 100 100 200 100

250 0.051 0.051 250 250 0.0510.051 0.051 0.051 250 0.051 0.051 -4.63396079e+00 -3.24941049e+00 -2.09240068e+01 300 300 300 300 150 0.05 150 0.05 150 0.050.05 0.05 150 0.05 0.05 0.05 200 200 200

100 350 350 350 100 100 350 0 0.0490 0 0.049 0.049 0.049 -100 0.049 0.049 0.049 0.049 -100 -100

-200 400 400 400 -200 -200 400 200 200 200 200 200 0.048200 2000.048 0.048 0.048 0.048 200 0.048 0.048 0.048 0 200 200 450 450 0 0 450 0 450 0 0 -200 -200 -200 -200 Fluence through Location #1 Fluence through Location #2 Fluence through-200 Location #1 Fluence through-200 Location #2 Φ Φ -4.63396079e+00 Φ Φ -3.24941049e+00 -2.09240068e+01 500 50 500 500 0.047 0.047 0.047 500 50 0.0470.047 0.047 50 0.047 0.047 50 100 150 200 50 100 150 200 40 50 100 50150 100 150200 200 40 5050 100 150100 200 150 200 50 100 150 200 50 100 150 200 40 40 40 4040 0.055 30 30 0.0550.05530 0.055 20 2020 2020 2020 10 10 10 0 50 0 0 0 0 50 0 0 0.054 -10 -10 0.0540.054-10 0.054 -20 -20-20 -20-20 -20-20 -30 100 -30 100 -30 50 -40 -40-40 50 -40-40 -40-40 -50 -50 0.053-50 0.053 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 0.0530 50 100 150 200 250 300 350 0.053 0 150 100 200 300 0 100 200 300 0 150 100 200 300 0 100 200 300

0.052 0.052 Θ 0.052 Θ 200 Θ 0.052 Θ 200 2.2x d’ 100 2.5x d’ 100

250 0.051 250 0.0510.051 0.051

300 300 0.05 150 0.05 0.05 150 0.05

350 350 0.049 0.0490.049 0.049 400 400 200 200 0.048 0.0480.048 0.048 450 450

500 0.047 500 0.0470.047 0.047 50 100 150 200 50 100 50 150100 150 200 50 100 150 200 Simple Object – Task-dependence High-Frequency Symmetric Task Asymmetric Line Pair Task Orbit #1 Orbit #2

0.06 0.06

0.055 0.055

0.05 0.05

0.045 0.045

50 50

40

30

20

10

0 0

-10

-20

-30

-40

-50 -50 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 5 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Intuition on Task-dependence

Spatial Resolution WXY WXZ WYZ -0.4

-0.2

fY 0 0.2

Noise Imaging Task 0.4

-0.4 -0.2 0 0.2 0.4 fX Symmetric Task Orbit Asymmetric Line Pair Task Orbit MTFXY MTFXZ MTFYZ MTFXY MTFXZ MTFYZ

NPSXY NPSXZ NPSYZ NPSXY NPSXZ NPSYZ

Realistic Object – Simulation Experiment Object: Anthropomorphic Head Phantom, Platinum Embolization Coils 4 3 5 Simulated Bleeds 2 Optimization: 6 1 9 orbital bases; -50° ≤  ≤ 50°, 162° ≤ q ≤ 378°; CMA-ES (pop=40) Six individual location designs, acquisitions, reconstructions Location #1 Location #2 Location #3 Location #4 Location #5 Location #6 50 50 50 50 50 50

0 0 0 0 0 0

-50 -50 -50 -50 -50 -50

200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 Circular Orbit Circular

50 50 50 50 50 50

0 0 0 0 0 0

-50 -50 -50 -50 -50 -50

200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350

Driven Trajectory Driven

- Task

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 6 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Realistic Object – Simulation Experiment Object: Anthropomorphic Head Phantom, Platinum Embolization Coils 4 3 5 Simulated Bleeds 2 Optimization: 6 1 9 orbital bases; -50° ≤  ≤ 50°, 162° ≤ q ≤ 378°; CMA-ES (pop=40) Six individual location designs, acquisitions, reconstructions Location #1 Location #2 Location #3 Location #4 Location #5 Location #6 50 50 50 50 50 50

0 0 0 0 0 0

-50 -50 -50 -50 -50 -50

200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 Circular Orbit Circular

50 50 50 50 50 50

0 0 0 0 0 0

-50 -50 -50 -50 -50 -50

200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350 200 250 300 350

Driven Trajectory Driven

- Task

-3.19138243e+08 -4.72467680e+08

Testbench Investigations

Anthropomorphic Head Phantom Modified CBCT Testbench and Synthetic Vasculature with Tilt Platform

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 7 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Results: Testbench Studies Preoperative Scan Task-Driven Trajectory Circular Scan

Multi-Location Optimization (In-plane)

ˆ , ˆ ˆ arg max mind '2 ,,,  ;WWW , d ' 2   ; ,..., d ' 2    ;     12     Task(1)      Task (2)  L      Task ( L )  ,

Optimization: 4 3 6 stimulus locations in an axial slice centered on coil 5 2 9 orbital bases; -50° ≤  ≤ 50°, 0° ≤ q ≤ 360°; CMA-ES (pop=40) 6 1

50

0 d’1/d’0 = 1.18 -50

0 d’2/d’0 = 1.07 -50

0 d’3/d’0 = 1.73 -50

0 d’4/d’0 = 1.61 -50

0 d’5/d’0 = 1.21 -50

0

-50 d’6/d’0 = 1.15 0 50 100 150 200 250 300 350

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 8 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Multi-Location Task-Driven Results(In-plane) Circular Scan

0.03

0.025

0.02

Task-Driven Trajectory 0.015

0.01

0.005

Multi-Location Optimization (3D Shell)

Optimization: 30 stimulus locations on ellipsoid surrounding embolization coil 9 orbital bases -50° ≤  ≤ 50° 0° ≤ q ≤ 360° CMA-ES (pop=40)

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 9 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Circular vs. Task-Driven

Multi-Location Task-Driven Results(3D Shell) Circular Scan

0.028

0.026

0.024

0.022

Task-Driven Trajectory 0.02

0.018

0.016

0.014

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 10 J. Webster Stayman ([email protected]) Fully 3D 2015 (May 31-June 4, 2015)

Conclusions and Future Work

Demonstrated: Patient- and task-specific acquisition design Location-dependence, Task-dependence Single- and Multi-location design objectives

Ongoing Work: Evaluation in physical experiments Practical workflow issues (registration) Other acquisition parameters (e.g., mA modulation)

AIAI Laboratory (aiai.jhu.edu) and I-STAR Laboratory (istar.jhu.edu), Dept of Biomedical Engineering, Johns Hopkins University 11