Automated Segmentation of Subcutaneous and Visceral Adipose Tissues from MRI

Our objective is to create an automatic image segmentation system of the abdomino-pelvic area allowing quick adipose tissue segmentation demanding minimal intervention of the human operator. The algorithm tackling with this problem is a process encompassing several image processing techniques including morphological operators (erosion, dilatation), binarization, connected-component labeling, holes filling techniques, watershed transform, achieving accurate segmentation results. We have performed computational experiments over 24 anonymized human magnetic resonance image datasets, consisting of 60 slices each, of the abdomino-pelvic area. Right now the evaluation is visual, because we still haven’t got the manual delineation serving as ground truth, so that we can’t calculate the accuracy, sensitivity or specificity values. Medical expert examination of the obtained results are encouraging, comparable to manual segmentation. This automatic method could save a lot of time allowing the realization of large scale studies.