SIGNATURE WORK
CONFERENCE & EXHIBITION 2023

Liver cancer is a serious condition with a low 5-year survival rate. Accurate segmentation of liver images is crucial for effective treatment planning. Current methods to conduct liver segmentation, such as manual and semi-automated segmentation, are time-consuming and prone to error. Therefore, deep learning methods in automatic segmentation have shown promise in achieving higher accuracy and precision in liver segmentation. The proper selection of ResNet and hyper-parameter with tuning is critical for optimal performance. However, many researchers overlook this step, which can have significant consequences for model accuracy and reproducibility.

This paper aims to use transfer learning to develop an efficient neural network based on pre-trained ResNet that fine-tunes existing parameters to improve segmentation accuracy and provides a detailed explanation of the fine-tuning process to identify potential future improvements. The study’s outcomes show that ResNet34 with a weight decay value of 0.5, 16 batch size, and 3 epochs is the most efficient network which can shed light on future research.