Abstract
1634
Objectives Dynamic positron emission tomography (PET) is being increasingly utilized to quantify myocardial blood flow (MBF), and may improve management of coronary artery disease and patient prognosis. Patient body motion during image acquisition can result in severe MBF estimation errors. This work investigates whether image decomposition holds promise for detection of heart motion in dynamic 82Rb PET based by analyzing patterns in the residue image.
Methods 108 noise-free and 108 noise-added NURBS-based cardiac-torso (NCAT) numerical-phantom dynamic image sequences were generated of which half had ±1 cm simulated motion and half had ±2 cm simulated motion. In addition, images from 65 patients that had undergone dynamic cardiac Rubidium-82 PET imaging at rest and stress were visually inspected for the presence of heart motion. All the image sets were decomposed using PCA and the residue images were then visually inspected for any prominent patterns resembling the myocardium, indicating the presence and time of body motion. The root mean square (RMS) of the residue activity in each time frame was calculated, and a threshold of the peak RMS was used as an automatic indicator of the time of motion. All images were processed and the frequency and time of motion were recorded and compared to the reference standard (simulated or visual assessment).
Results For 1 cm motion, noise-free simulations, 42/52 motion times were accurately detected while all 54 motion frames were detected within ±2 time frames. For 2 cm motion noise-free simulations, 52 events were accurately detected and all 54 within ±2 time frames. For noise-added simulations with 1 and 2 cm motion, 28 and 38 events respectively were accurately detected while 50 and 49 motion frames were within ±2 time frames. Of 130 patient images, 60 events were accurately detected while 115 motion events were detected within ±2 time-frames. Missed events tended to occur most frequently in the middle time frames of the dynamic image sequence.
Conclusions Preliminary data suggests that heart organ motion detection in dynamic PET using image decomposition may be feasible, especially in the early and late time frames.