Patient oriented neural networks to overcome challenges of abdominal organ segmentation in CT angiography studies

Segmentation of organs from abdominal Computed Tomography Angiography (CTA) images is one of the essential steps in quantitative measurements (e.g. volume, size etc.). Due to gray level similarity of adjacent organs, injection of contrast media, high variations in organ borders, partial volume effects and atypical shapes, effective segmentation of these organs is a very difficult task. In this paper, we propose a semi automatic and neural network based segmentation method that adapts its parameters according to each dataset by learning the data characteristics in parallel to segmentation process, thus named patient oriented neural networks. Proposed approach makes the design of the overall system fully automatic (if the initial segmentation result is also automatic) without requiring any training set. The segmentation results are evaluated by using area error rate and they show that, the proposed algorithm gives promising results in most of the challenging aspects of abdominal organ segmentation.