Automated and Large-Scale Liver Segmentation Using Deep Learning: a Promising Approach for Accurate Diagnosis

EasyChair Preprint 10485

Abstract

In this paper proposes an automated and large-scale segmentation of the liver based on deep learning. Segmentation of the liver and liver lesions is very important in the initial diagnosis of the doctor. In the past, manual segmentation was usually used, but it took too long and there was human error. The automatic liver and lesions segmentation on a large scale can greatly reduce the diagnosis time. First of all, we will do a variety of pre-processing of the slice map, including preliminary organ differences and histogram equalization. Furthermore, due to the lack of pre-processing training data, we use data augmentation methods to increase our training data. We divide the model into two parts. The first part focuses on the prediction of the liver and the second part focuses on the segmentation of the liver. We trained more than 30,000 liver slice maps. Experiments show that our DICE Score can exceed 89% in the liver segmentation, and the lesion prediction is 65%.

Keyphrases: CNN, LITS dataset, Liver tumors, automatic segmentation, deep learning

@booklet{EasyChair:10485,
  author    = {Hesham Mostafa},
  title     = {Automated and Large-Scale Liver Segmentation Using Deep Learning: a Promising Approach for Accurate Diagnosis},
  howpublished = {EasyChair Preprint 10485},
  year      = {EasyChair, 2023}}