LEARNING-BASED VENTRICLE DETECTION FROM CARDIAC MR AND CT IMAGES

The objective of this work is to investigate the issue of automaticall y detecting regions of interest (ROI's) in medical images, It is assum ed that the regions to be detected can be roughly segmented by a thres hold based on a likelihood measure of the ROI, First, an analysis of t he global histogram is used to compute a preliminary threshold that is likely near the optimal one, The histogram analysis is motivated by t he analytical result of a bell image intensity model proposed in this work, Then, the preliminary threshold is used to segment the input ima ge, resulting in an attention map, which contains an attention region that approximates the ROI as well as many spurious ones, Due to the no noptimality of the preliminary threshold, it can happen that the atten tion region contains a part of, or more regions than, the ROI, Learnin g takes place in two stages: 1) learning for automatic selection of th e preliminary threshold value and 2) learning for automatically select ing the ROI from the attention map while dynamically tuning the thresh old according to the learned-likelihood function, Experiments have bee n conducted to approximately locate the endocardium boundaries of the left and right ventricles from gradient-echo magnetic resonance (MR) i mages, Cardiac computed tomograph (CT) images have also been used for testing, The boundary of the segmented region provided by this algorit hm is not very accurate and is meant to be used for further fine tunin g based on other application-specific measures.