Segmentation of 4D cardiac MRI: Automated method based on spatio-temporal watershed cuts

Based on discrete mathematical morphology, we introduce in this paper a new watershed framework which allows for segmenting spatio-temporal images, that we apply to medical image analysis. Specifically, we propose a new automated and fast procedure to segment the left ventricular myocardium in 4D (3D+t) cine-MRI sequences. Both quantitative and qualitative evaluation are provided. Thanks to the comparison with manual segmentation performed by two cardiologists, we demonstrate the accuracy of the proposed method. The relevance of the ejection fraction and myocardium mass measured from segmentations is also assessed. Furthermore, we show that the proposed 4D procedure allows to keep the temporal coherency between the successive 3D segmentations obtained along the time sequence. In an effort to promote open science, both the data and the software are available on-line.