Liver Segmentation from Combined Tomography (CT) Abdominal Images Data Using Deep Neural Networks

Abstract

The field of medical image segmentation is rapidly advancing over time. Segmentation of organs helps medical professionals to plan radiotherapies and track the prognosis of disease in general. Image segmentation of abdominal organs can facilitate clinical procedures for example transplantation surgeries and to determine the precise location of organ in aortic abdominal surgeries. The study focuses primarily on segmentation of liver from computed tomography (CT) dataset of healthy abdominal organs. The dataset taken for the study is from Combined (CT-MR) Healthy Abdominal Organ Segmentation (CHAOS) Challenge. The dataset used three-dimensional data to enrich the level of analysis. The dataset used is in DICOM format. All the patient images were of healthy liver, aligned in the same position while they were given contrast injection for enhanced vascular structure. The automated model used for the segmentation of images was U-NET. Two convolutional layers were used and the size of each image was halved. The expansive and contractive parts of the layers were repeated thrice. PYtorch and FastAl libraries were used for training the data set. 15% of the images were used for validation. The learning rate was altered during training to avoid overfitting. The metrics used for calculating results was dice. The calculated dice with our algorithm is 0.968 while the state-of-the-art dice reported by CHAOS for the given challenge was 0.979.