Ventricle detection based on boolean maps and geometric features
Chen Dongyue, Jingfeng Zhang, Chen Huazhao, Tong Jia, Chengdong Wu · 2016
An adaptive initialization algorithm is proposed to serve the fine segmentation of the left ventricles in MRI images. The proposed algorithm combines boolean maps and geometric features to extract the target region. Boolean maps are obtained by segmenting the input image with an adaptively varying threshold, which promises the better change to distinguish the target from its surroundings. A target detection scheme contains a cascade classifier and a scoring algorithm is constructed based on some trained models of multiple geometric features. By sending hundreds of regions of interest derived from boolean maps into the target detection model, the left ventricle region can be extracted in a fast and reliable way. Experimental results demonstrated that the proposed model can effectively extract the target region as the initialization for the finer segmentation of the left ventricle.