Some Aspects in Medical Imaging Debasis Mitra Computer
Some Aspects in Medical Imaging Debasis Mitra Computer Science Florida Institute of Technology Acknowledgement: Grant T. Gullberg Radiotracer Department Life Sciences Division Lawrence Berkeley National Lab & Unknown sources from the Web 9/26/2020 CS Seminar, FIT 1
Co-ordinates o. Why this talk? o. Where am I now? o. What does this lab do? 9/26/2020 CS Seminar, FIT 2
Lawrence Berkeley National Lab 9/26/2020 CS Seminar, FIT 3
Center for Functional Imaging 9/26/2020 CS Seminar, FIT 4
Biomedical Imaging is the Engineering behind Radiology • 9/26/2020 CS Seminar, FIT 5
Types of Imaging Instruments Computer Tomography (X-ray) Magnetic Resonance Imaging (MRI) Single Photon Emission Computed Tomography (SPECT): gamma ray of 100 -few hundred kev Positron Emission Tomography (PET): gamma ray from in situ positron annihilation, 500 kev Ultra Sound Optical or Laser Tomography (Infrared) Fluoroscopy, Opto-acoustic, Electron, Atomic-force, Radio-frequency, … 9/26/2020 CS Seminar, FIT 6
CT 9/26/2020 CS Seminar, FIT 7
GE VG 3 Millennium Hawkeye SPECT/CT collimators -ray detectors Resolution Sensitivity 9/26/2020 CS Seminar, FIT Acquisition system 8
Scintillation Camera and Collimator Patient Collimator localizes events in object and determines sensitivity and spatial resolution of the camera 9/26/2020 CS Seminar, FIT 9
Collimator converging parallel hole 9/26/2020 pinhole CS Seminar, FIT 10
Positron Emission Tomography Does Not Need a Collimator Positron annihilates with electron two gamma photons each at 511 ke. V leave under 180 9/26/2020 Coincidence detection (“electronic CS Seminar, FIT collimation”) 11
PET 9/26/2020 CS Seminar, FIT 12
MRI Epilepsy: MRI, PET-time 1, 2 Brain tumor 9/26/2020 CS Seminar, FIT 13
A B Fiber Tracking of DTMRI Data C D E Rohmer D, Sitek A, Gullberg GT: Reconstruction and visualization of fiber and laminar structure in the normal human heart from ex vivo DTMRI 9/26/2020 data. Investigative Radiology, CS Seminar, FIT 42: 777 -789, 2007. F 14
Ultrasound 9/26/2020 CS Seminar, FIT 15
Cardi. ARC 9/26/2020 CS Seminar, FIT 16
Clinical Feasibility Results Spectrum Dynamics Conventional 1. 45 Mcounts total (heart 10%, backgnd 90%) Pixel size 6. 91 mm × 6. 91 mm Iterative reconstruction Total acquisition time: 17. 5 min 9/26/2020 0. 8 Mcounts total (heart 60%, backgnd 40%) Pixel size 2. 46 mm × 6. 91 mm Iterative reconstruction Total acquisition time: 2. 2 min CS Seminar, FIT 17
Radiopharmaceuticals for Cardiac Imaging 201 Tl 99 m. Tc-sestamibi isonitrile) (2 -Meth 0 xy-2 -methylpropyl 99 m. Tc-tetrafosmin 99 m. Tc-teboroxime 123 I-iodorotenone 123 I-BMIPP 123 I-IPPA 9/26/2020 (fatty acid) CS Seminar, FIT 18
Targets of Study • Heart, • Lungs, liver, other organs in torso • Brain: Alzheimer’s Disease Neuroimaging Initiative (ADNI) • Breast • Tumor 9/26/2020 Breast Cancer CS Seminar, FIT 19
Physics behind Models • Emission tomography: SPECT, PET, MRI • Transmission tomography: X-ray, Optical • Reflection: Ultra Sound, Total Internal Reflection Fluoroscopy (TIRF for single cell visualization) • Scattering: Muon tomography? 9/26/2020 CS Seminar, FIT 20
Mathematical Problem Formulation • Forward Problem (modeling): How the data would look like given probe and the model • D = F(M): Forward project • An implementation is a Simulation software • Inverse Problem (tomography): What the model would be given the probe and data • M = F ~ (D): back-project • An implementation is a Reconstruction software • Noise in data makes it a hard statistical problem • Data volumemay be additional computational challenge • http: //en. wikipedia. org/wiki/Inverse_problem 9/26/2020 CS Seminar, FIT 21
Reconstruction Algorithms • Analytic-inverse: E. g. , Radon transformation for emission/absorption (mostly useless except for theoretical purpose) • Algebraic Reconstruction: voxel by voxel reconstruct the model • Iterative Reconstruction using Expectation Maximization • Ordered Set – EM • Maximum A Posteriori (MAP-EM) • Penalized Least Square (PLS): 1. 5 iteration! 9/26/2020 CS Seminar, FIT 22
Dynamic Imaging • Problem: Objects move during data gathering • Question: How to reconstruct (1) Object, (2) Motion • A successful approach: Level Set • For blood concentration change in tissues: • Temporal B-spline • Tensor imaging with MRI 9/26/2020 CS Seminar, FIT 23
Fit the 123 I-BMIPP Data to a Compartment Model Need to estimate an input function. bloo Time activity curves have to be d estimated directly from the projections. A methyl group on the position of the carbon chain limits the oxidation of 123 I-BMIPP. IPPA 123 Differs from IPPA which is completely metabolized to benzoic acid. k 21 k 32 C 2(t) k 12 C 3(t) k 23 TG Model of IPPA Metabolism Benzoic acid 9/26/2020 CS Seminar, FIT 24
Spatiotemporal Modeling Using A Small Number of Splines to Represent Realistic Physiological Curves –Quadratic B-Spline Temporal Basis Functions –Zero Order (voxels) B-Spline Spatial Bases 9/26/2020 CS Seminar, FIT 25
Slow-Rotation Dynamic Pinhole SPECT 0 o 72 o 8 o 16 o 24 o 32 o 80 o 88 o 96 o 104 o Blood Activity Curve o 40 o Time 48 56 o Estimated from Projections Using Factor Analysis 112 o 120 o 128 o 64 o 136 o 352 o 9/26/2020 1 sec frames, 180˚ rotation of one head CS Seminar, FIT Recirculation time is 6 -8 seconds 26
Results — Dynamic Early Data 9/26/2020 CS Seminar, FIT 27
Image Spatial Representations Pixels / voxels regular Blobs Linear B-splines Cubic B-splines sparse “Custom-made” shapes Irregular meshes. . .
Metabolic Rate of BMIPP Normal Ki=0. 40 min-1 SHR Ki=0. 15 min-1 9/26/2020 CS Seminar, FIT 29
Metabolic Rate: FDG vs BMIPP FDG Ki=0. 40 min-1 Ki=0. 15 min-1 9/26/2020 CS Seminar, FIT 30
SHR: hypertensive rat model (genetically modified)
WKY: normal rat 9/26/2020 CS Seminar, FIT 32
Flow rate changes SHR: Hypertensive, WKY: normal Age (months) 7 SHR WKY A (min-1) 0. 94 B (min-1) A (min-1) 1. 44 B (min-1) 14 21 9/26/2020 0. 22 0. 60 CS Seminar, FIT 33
Temporal Comparison of 1 st Principal Strain for SHR and WKY anterior wall septum A A SHR red 0. 6 0 SHR 6/18/20 B 03 8/06/20 10/01/2 Normal 03 003 6/18/20 C 03 8/06/20 10/01/2 SHR red 03 003 6/18/20 D 03 8/06/20 10/01/2 Normal 03 003 12/2/20 03 7/14/20 04 0. 6 0 1. 2 5 12/2/20 7/14/20 03 1. 2 04 0. 8 9/21/20 5 04 5 9/26/2020 8/06/20 03 10/01/2 003 0. 6 0 0. 3 0 0. 0 4/27/20 0 04 0. 8 4/27/20 5 04 0. 9 0 0. 0 1 st PS 0 9/21/20 04 WKY 6/18/20 03 C B FS CS Seminar, FIT 0. 0 0 For war d War ping Veress A et al. : Regional changes in the diastolic deformation of the left ventricle for SHR and WKY rats using 18 FDG based micro. PET technology and hyperelastic warping. Annals of Biomedical Engineering 36: 1104– 1117, 2008. 34
Parametric Imaging Summed Images (between 2 and 12 min) Parametric Images of k 21 Sitek A, Di Bella EVR, Gullberg GT, Huesman RH: Removal of liver activity contamination in teboroxime dynamic cardiac SPECT imaging 9/26/2020 using factor analysis. J Nucl Cardiology 9: 197205, 2002. CS Seminar, FIT 35
SUMMARY • The SHR shows increased glucose metabolism and reduced fatty acid metabolism. • The reverse is true for the nomotensive WKY rat. • The SHR model is used to develop techniques for analysis of imaging data of heart failure related to metabolism. • Molecular Insight Pharmaceuticals is now evaluating 123 I-BMIPP in clinical trials. • These results of fatty acid metabolism correlate with those in humans with hypertensive left ventricular hypertrophy. (de las Fuentes et al. J Nucl Cardiol 13: 369, 2006) 9/26/2020 CS Seminar, FIT 36
COMMENTS • The SHR has a defective gene (CD 36) on chromosome 4. • The defect is associated with compromised long-chain FA transport across the cell membrane. • The defect causes insulin resistance, alteration in basal glucose metabolism. • Short-chain FA diet decreases glucose uptake, alleviates hypertrophy, but hypertension is not improved. • Proposed research will compare 123 I-BMIPP with 18 FTHA. Hajri T et al. : Defective fatty acid uptake in the spontaneously hypertensive rat is a primary determinant of altered glucose metabolism, hyperinsulinemia, and myocardial hypertrophy. J Biological Chem 276: 23661 -23666, 2001. 9/26/2020 CS Seminar, FIT 37
MRI is way advanced in Dynamic Imaging Diffusion Tensor Imaging A high-resolution diffusion tensor imaging scan reveals differences between healthy tracts of axons, at left and in the lower enlargement, and tracts of injured axons, at right and in the top enlargement, in a person who sustained a moderate to severe traumatic brain injury. Such damage has been shown to correlate with cognitive impairment. (Image courtesy of Dr. Deborah Little) 9/26/2020 CS Seminar, FIT 38
Diffusion Tensor Diffusion within a single voxel. (a) Diagram shows the 3 D diffusion probability density function in a voxel that contains spherical cells (top left) or randomly oriented tubular structures that intersect, such as axons (bottom left). This 3 D displacement distribution, which is roughly bell shaped, results in a symmetric image (center), as there is no preferential direction of diffusion. The distribution is similar to that in unrestricted diffusion but narrower because there are barriers that hinder molecular displacement. The center of the image (origin of the r vector) codes for the proportion of molecules that were not displaced during the diffusion time interval. 9/26/2020 CS Seminar, FIT 39
A B Sheet Tracking of DTMRI Data C E 9/26/2020 CS Seminar, FIT D F 40
Fiber Tracking of Right and Left Ventricle B A Cardiac Band Hypothesis: The four chamber heart is built from a single continuous band of muscle. Torrent-Guasp F, Kocicab MJ, Cornoc AF, et al. Towards new understanding of the heart structure and function. Eur J Cardiothorac Surg. 2005; 27: 191 -201. 9/26/2020 CS Seminar, FIT 41
Advancement of Data Acquisition Technology • List mode: acquire data for recording time for each track and reconstruct with it: a computational challenge • Time-of-flight: Acquire event versus data collecting time: new type of detectors needed • Compton gamma camera: provides some measure of angle of a track • Newer Technology: Opto-acoustic, Fluorescence, … • Target-specific detectors: e. g. , Cardiac-Spect, faster and cleaner data with higher resolution 9/26/2020 CS Seminar, FIT 42
Molecular Imaging • Medical imaging is primarily at organ-level • With more genetic information available today it is usual to think in terms of metabolism behind images, and target cellular-level processes • Current focus is to develop ligands that are (1) tagged with imaging agents, (2) binds to some protein or metabolite that we want to visualize with imaging • Understanding dynamic organ-level images from metabolic point of view is another new area 9/26/2020 CS Seminar, FIT 43
Total Internal Reflection Imaging TIRF imaging of actin networks and their reorganization in the cortex of Dictyostelium cells. 9/26/2020 CS Seminar, FIT 44
Auto-diagnosis/prognosis: Machine learning • Images are still used by radiologists for diagnosis/prognosis, or by biologist for doing science: technology targets exclusively to improve image quality, and nothing more • It is quite possible to use machine learning algorithms to help the process: image is input, zones of interest with annotations are output 9/26/2020 CS Seminar, FIT 45
Thanks! Debasis Mitra dmitra@cs. fit. edu 9/26/2020 CS Seminar, FIT 46
- Slides: 46