Automatic segmentation of the epicardium in late gadolinium enhanced cardiac MR images

A novel method for fully automatic segmentation of the epicardium of the left ventricle in Late Gadolinium Enhanced Cardiac MR images (LGE-CMR), short axis view is presented. No other images or measurements are used. The method uses prior information about typical heart size and shape making an a priori probability map around the automatically detected heart center (HC). An a posteriori probability map is made iteratively by combining the current probability model with the preprocessed image and low-pass filtering over consecutive slices. The a posteriori probability map is used as the input for segmentation. Using the detected HC as origin, a radial evaluation is performed for all possible angles [0-2π], giving a candidate border point on the epicardium. The final segmentation combines the information from all slices. The presented method is tested on all slices of 54 patients, and gives a mean Dice index of 0.86 and mean Jaccard index of 0.76.