Cardiac Phase-resolved Blood-Oxygen-Level-Dependent (CP-BOLD) MRI examines changes in myocardial oxygenation in response to ischemia without contrast and stress agents. Since signal intensity changes are subtle, quantitative approaches are necessary to examine variations in myocardial BOLD signals and identify ischemic myocardial territories. Here, using data from animal studies, we extract myocardial time series (BOLD signal as a function of cardiac phase) and explore such variations using a structured dictionary-learning framework, considering shift-invariant learning and spatial priors. We use it: to learn a model of baseline (absence of disease) myocardial time series; and in datasets where disease is assumed, to obtain a spatial map of ischemia presence, identifying myocardial time series from ischemic territories in an unsupervised fashion, by exploiting structural properties, or the lack thereof, in the data. By providing new visualization and quantification approaches, we hope to accelerate the clinical translation of cardiac BOLD MRI for noninvasive ischemia assessment.
|Number of pages||8|
|Journal||Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention|
|Publication status||Published - 1 Jan 2014|