CSE 337: Introduction to Medical Imaging Lecture 10: Nuclear Imaging :CSE 337: Introduction to Medical Imaging Lecture 10: Nuclear Imaging Klaus Mueller Computer Science Department Stony Brook University


Overview :SPECT: Single Photon Emission Tomography PET: Positron Emission Tomography Idea: inject (into the bloodstream) a tracer molecule labeled with a radionucletide there are specific tracer molecule for specific targets example: Deprenyl triggers the production of dopamine in the human brain the molecule will go to the target anatomic site with metabolic activity (e.g., a brain area) tracer will give rise to X-rays that can be detected Just like fMRI it is a metabolic imaging modality, but with much higher SNR (orders of magnitude higher) Overview


Relation To Anatomic Imaging :Relation To Anatomic Imaging


PET: Concept (1) :PET: Concept (1)


PET: Concept (2) :PET: Concept (2)


PET: Case Study :PET: Case Study PET scan takes usually 30 min (brain) to 60 min (whole body) Usually displayed pseudo-colored: red, yellow: high activity green, blue: low activity raw Pseudo-colored normal Alzheimer’s


PET: Case Study :PET: Case Study Reduced Cerebral Blood Flow (CBF) and elevated compensatory Oxygen Extraction (OEF) before and after carotid artery angioplasty (stroke risk)


SPECT: Concept :SPECT: Concept A labeled tracer (e.g., glucose) is injected into the blood stream: only a single photon is emitted slower decay than PET study length about 20 min (heart) Applications: measure blood flow through arteries and veins brain, heart, renal gamma cameras


SPECT: Case Studies :SPECT: Case Studies Brain: uncontrolled complex partial seizures left temporal lobe has less blood flow than right indicates nonfunctioning brain areas causing the seizures Heart: perfusion of heart muscle orange, yellow: good perfusion blue, purple: poor perfusion brain metabolism heart


PET vs. SPECT (1) :PET vs. SPECT (1) SPECT: a single photon is produced (need collimator on the detector to determine its path) low resolution (6-8 mm) tracer decay slower therefore longer-lasting effects can be monitored tracers don’t have to be produced on site


PET vs. SPECT (2) :PET vs. SPECT (2) PET: no collimators needed -- annihilated positrons yield detectable dual gamma rays 180 apart tracers decay fast transient processes can be monitored scan time short (less than a minute) tracers must be produced in nearby cyclotrons more expensive equipment (detector hardware) higher resolution than SPECT (2-3 mm) best for the study of brain receptors with particular neurotransmitters (over fMRI) also much better SNR than fMRI


Reconstruction: Iterative Methods :Reconstruction: Iterative Methods Iterative methods are advantageous in these cases: limited number of projections irregularly-spaced and -angled projections non-straight ray paths (example: refraction in ultrasound imaging) corrective measures during reconstruction (example: metal artifacts) presence of statistical (Poisson) noise and scatter (mainly in functional imaging: SPECT, PET)


Simultaneous Algebraic Reconstruction Technique (SART) :Iteratively solves W V=P Simultaneous Algebraic Reconstruction Technique (SART)


SART :Projection (into pixel) Projection vj P SART


SART :Projection (into pixel) Normalization at pixel i Scanned pixel Correction factor computation C SART


SART :Projection (into pixel) Normalization at pixel i Backprojection (into voxel) Scanned pixel Backprojection vj C SART


SART :Projection (into pixel) Normalization at pixel i Normalization at voxel j Backprojection (into voxel) Scanned pixel Voxel normalization vj SART


SART :Projection (into pixel) Normalization at pixel i Normalization at voxel j Backprojection (into voxel) Scanned pixel New (k+1) and previous (k) values of voxel j Voxel update vj SART


SART :Next projection SART


Iterative Reconstruction Demonstration: SART :Iterative Reconstruction Demonstration: SART


Iterative Reconstruction Demonstration: SART :Iterative Reconstruction Demonstration: SART


Maximum Likelihood Expectation Maximization (ML-EM) :Maximum Likelihood Expectation Maximization (ML-EM) Maximizes the likelihod of the values of voxels j, given values at pixels i


Maximum Likelihood Expectation Maximization (ML-EM) :Maximum Likelihood Expectation Maximization (ML-EM) Projection (into pixel i) Backprojection (into voxel j) Normalization at voxel j New (k+1) and previous (k) values of voxel j Maximizes the likelihod of the values of voxels j, given values at pixels i


Algorithm Comparison :Algorithm Comparison SART: projection ordering important ensure that consecutively selected projections are approximately orthogonal random selection works well in practice EM: convergence slow if all projections are applied before voxel update use OS-EM (Ordered Subsets EM): only a subset of projections are applied per iteration