Automatic Multi-organ Segmentation on Abdominal CT scans using Deep U-Net Model

Reliable, automated segmentation of the abdominal organs is often critical for computerized diagnosis and assistance in laparoscopic surgery. Many of the methods used so far specialize in the segmentation of individual organs or it is not possible to influence the variability of the shape and position of the abdominal organs. In this article, a fully-automated registration free U-Net architecture for multi-organ segmentation of abdominal CT scans using deep learning is presented. The method relies on U-net architecture. This approach allows for high inter-individual variability, while being flexible enough to be applied to different organs. U-Net for each organ are intensively trained around the organs, where the starting balance is sent from the previous stage. Segmentation in the proposed method consists of several stages. The effectiveness of the proposed method is tested on 50 data from abdominal computed tomography and it gives better results than traditional methods of 2D segmentation. Our data-set scan results are compared with state-of-the-art values with the dice/recall/precision/sensitivity values of 99% for spleen, left kidney, right kidney, right adrenal gland, left adrenal gland, gallbladder, esophagus, portal vein and splenic vein, liver, stomach, aorta, inferior vena cava, and pancreas. The effectiveness of the proposed method is measured by abdominal computed tomography data and gives better results than traditional 2D segmentation methods.