AnnularCut: a graph-cut design for left ventricle segmentation from magnetic resonance images

摘要: Clinician-friendly methods for cardiac image segmentation in clinical practice remain a tough challenge. Larger standard deviation accuracy may be expected automatic when the input dataset is varied; also at some instances radiologists find them hard case any correction desired. In this context, study presents semi-automatic algorithm that uses anisotropic diffusion smoothing and enhancing edges followed by new graph-cut method, `AnnularCut', three-dimensional left ventricle (LV) from selected slices. Unlike conventional cellular automata, where performance depends solely on features, method simultaneously considers minimal energy between two adjacent regions thus mitigating convergence problem. The main contributions can summarised as (i) dynamic automation approach to integrate distinct labels, (ii) generation of missing contours subject slices using level set construct volumetric LV. Both qualitative quantitative evaluation performed publicly available databases reflect potential proposed method.