Inside a CT Scanner Web Stayman, Advanced Imaging Algorithms and - PDF document

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" J. Webster Stayman Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) Johns Hopkins University October 16, 2019 Inside a CT Scanner Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 1

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" CT in Operation https://www.youtube.com/watch?v=2CWpZKuy-NE Forward Model Mean measurements as a function of parameters Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 2

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Noise in Projection Data 10000 0.08 9000 100 100 0.06 8000 200 200 0.04 7000 300 300 0.02 6000 5000 0 400 400 4000 -0.02 500 500 3000 -0.04 600 600 2000 -0.06 1000 700 700 -0.08 0 100 200 300 400 500 600 700 100 200 300 400 500 600 700 Beam Shaping – Bowtie Filters X-ray Source Patient Detector Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 3

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Multiple Aperture Devices (MADs) X-ray Design Target: X-rays Flatten fluence behind 1) a QRM phantom (30 x 20 cm), Source 2) cylinders 20~50 cm in diameter MAD Manufacturing: Tungsten powder, Laser Sintering, EDM Wire cutting Filter MAD0 (Stayman et al., SPIE 2016) Patient 15 mm 135 mm Detector MAD1 Spacing: 10 mm Thickness: 2mm Moiré patterns Mathijs Delbaere, behance.net Stayman, SPIE 2016 ; Mathews, CT Meeting 2016, SPIE 2017 ; Gang, PMB 2019 Experimental CBCT Bench Diagnostic CT Scanner Motion system on CT gantry SDD = 108 cm SAD = 80 cm SMD = 34 cm 0 cm 1 cm MAD1 MAD0 Actuation Stages Linear motors Flat Panel X-ray Motion Stage MADs Detector Source Beam shape: Miscentered object* and VOI imaging**: Relative displacement between MADs Both MADs moving to change beam center Amplitude: View-dependent mA and/or ms modulation *Mao, SPIE 2018, JMI 2018; ** Wang, CT Meeting 2018, SPIE 2019, JMI 2019 Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 4

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" MAD Acquisition - Ellipse Phantom MAD gain in air “One period” of MAD profiles Uniform Elliptical acrylic phantom (25.8 x 14.1 cm) Actuation for elliptical phantom Phantom acquisition Customized CT Workflow Customized Customized Initial Tube Current Model-based 3D Scout Volume Ultra-Low Dose Modulation Reconstruction Scan 𝜈 � = argmax Φ 𝑧; 𝜈 ? Φ = 𝑀 𝑧, 𝜈 − 𝑆(𝜈) r 1 r 2 Patient-Driven Diagnostic CT r 3 r 4 Prior Information Custom Custom about Patient Acquisition Regularization Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 5

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Image Quality Task-Based Performance Added noise realizations with different correlation while maintaining same variance True Signal “noise masquerading as signal” Contrary to CNR, task-based metrics: Accounts for spatial frequency characteristics of task, noise and system response Accounts for observer detection strategy and detection threshold Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 6

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Task-Performance: Detectability Index Detectability Index (non-prewhitening observer) Hypothesis #1 Hypothesis #2 Spatial resolution � � 2 Ω � , Ω � ⋅ 𝑋 ���� 2 𝑒𝑔 ∫ ∫ ∫ 𝑁𝑈𝐺 � 𝑒𝑔 � 𝑒𝑔 � �� Ω � , Ω � = 𝑒 � � 2 Ω � , Ω � ⋅ 𝑋 ���� 2 𝑒𝑔 ∫ ∫ ∫ 𝑂𝑄𝑇 � Ω � , Ω � 𝑁𝑈𝐺 � 𝑒𝑔 � 𝑒𝑔 � Noise Imaging task Task Function Hypothesis #1 Hypothesis #2 𝑋 ���� vs 𝑧 𝑔 � 𝑦 𝑔 � System Modeling Local NPS and MTF Penalized Likelihood Estimation (PLE) 𝜈 � = argmax log 𝑀 𝜈; 𝑧 − 𝛾𝑆 𝜈 (3) Likelihood Quadratic Penalty (1) Forward Model: (2) � = 𝐽 � 𝑣, 𝜄 𝑓 ��� 𝑣 : Detector location, q : Projection angle 𝑧 Object dependence ℱ A � D 𝑧 � A𝑓 𝑧 � : 1.2 � (via the projection data) NPS 𝑂𝑄𝑇 � ≈ -0.4 x10 -4 � ℱ A � D 𝑧 � A𝑓 � + 𝛾𝐒𝑓 � f y 0 𝛾𝐒 : Regularization dependence (1) (2) (3) 0.4 0 ℱ A � D 𝑧 � A𝑓 � 1.0 𝑁𝑈𝐺 � ≈ MTF -0.4 ℱ A � D 𝑧 � A𝑓 � + 𝛾𝐒𝑓 � e j : Location Dependence f y 0.5 0 0.4 0 -0.4 0 0.4 -0.4 0 0.4 -0.4 0 0.4 f x f x f x Fessler, IEEE-TIP 5(3), (1996); Stayman and Fessler, Trans. Med. Im. 23(12),2004; Gang et al., Med Phys 41(8) 2014 Zhang-O’Connor and Fessler, IEEE, 2007; Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 7

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Task-Driven Imaging Framework Task-Driven Imaging ∗ 𝛁 𝑩 Imaging Task 0 o Objective Optimizer Task-driven Location 𝑒 � Ω � , Ω � Detectability index argmax Conventional Contrast 𝑒′ Ω � , Ω � � � ,� � Spatial frequency 270 o 90 o System Model 0.2 0.4 Ω � Ω � 0.6 Anatomical 0.8 Model mAs 180 o Kernel (FBP) mAs, kV, Orbit Fluence field Regularization (MBIR) ∗ 𝛁 𝑺 r 1 r 2 Spatial resolution: 𝑁𝑈𝐺 Ω � , Ω � Noise: 𝑂𝑄𝑇 Ω � , Ω � r 3 Low Dose 3D Scout Task-Driven Optimization r 4 Traditional and Task-Driven Acq/Recon Uniform Minimum Task-driven Task-driven Task-driven Unmodulated Variance* (FBP) Regularization TCM + Reg. Signal TCM 0° 0° 0° 0° 0° 0° 90° 270° 180° 180° 180° 180° 180° 180° 𝑠 0 0 0 0 0 0 0 0 0.65 0.37 1.12 0.50 0 0 0 0 -0.91 −1.07 0 0 0 -0.71 0 −0.05 *Gies et al, MedPhys, 1999 Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 8

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Sample Reconstructions Task-driven Minimum Task-driven Task-driven Anisotropic Unmodulated Uniform Signal Variance (FBP) TCM TCM + Aniso Reg. Regularization � 𝑒 ��� = 1.0 𝑒′ = 0.77 𝑒′ = 0.90 𝑒′ = 1.08 𝑒′ = 1.0 𝑒′ = 1.10 All recons: are quadratic penalized likelihood have the 3 target stimulus have optimal b selection G. Gang et al. SPIE Medical Imaging, March 2016, 9783, PMCID: PMC4841467 G. Gang et al. Physics in Medicine and Biology, December 2017, PMCID: PMC5738673 Traditional versus Task-Driven Fluence-Field Modulation Design Objective Task and Phantom Definition Task Function Stimulus Phantom � Ω � , Ω � 𝑒 � � Ω � , Ω � 𝑒 � argmax min ⋮ � � ,� � � Ω � , Ω � 𝑒 � b ( x,y ) Flattened Fluence Min. Mean Variance (FBP) Task-Driven r ij ( x,y ) Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 9

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Reconstructions of Simulated FFM Data G. J. Gang, J. H. Siewerdsen, and J. W. Stayman, "Task-driven optimization of fluence field and regularization for model-based iterative reconstruction in computed tomography", IEEE Transactions on Medical Imaging (Special Issue on Low-Dose CT), 36(12), 2424-35 (December 2017) Detectability Maps a = 1.0 a = 0.5 Unmodulated Task-Driven Other Novel Acquisition: Orbits Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 10

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Task-Driven Interventional Imaging Conventionally Ignored Conventional by Interventional Devices Interventional Imaging Intraoperative CT Preoperative Planning Diagnostic Flat-Panel Detector Image Data Imaging ? Task-Driven Trajectory Task Traditional Definition X-ray Circular Source Patient- and Trajectory Task-Driven Prior Information Intraoperative CT about Patient and Task Trajectory Design Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 11

IMA Workshop - "Novel CT Data Acquisition 10/16/2019 and Processing" Emulated Workflow on CBCT Bench Anthropomorphic Head Phantom and Synthetic Vasculature CBCT Testbench with 6DOF Object Platform Preoperative Scan CBCT Bench Results Task-Driven Trajectory Circular Scan J. W. Stayman and J. H. Siewerdsen, Int'l Mtg. Fully 3D Image Recon. June 16-21, 2013. S. Capostagno et al. and Stayman et al.,”Task-driven source–detector trajectories in cone-beam computed tomography: I and II” J. Med. Im. 2019 Web Stayman, Advanced Imaging Algorithms and Instrumentation Lab (aiai.jhu.edu) 12