Efficient Liver and Tumor Segmentation from CT Images Using a Hybrid ResUNet Model

In this paper, we investigated deep learning models for automated segmentation of liver and tumor regions from CT images to improve the diagnostic accu-racy as well as treatment planning in case of liver cancer. The research used a dataset of 3200 CT scans, with about 70% serving as the testing set and 30% for training. Preprocessing steps like Auto Windowing HU (Hounsfield Units) and Histogram Equalization to retain better clarity of the images. Based on the metrics of performance such as accuracy, Dice coefficient, IoU(Intersection over Union), precision and recall were calculated used three types of CNN models (ResUNet, VGG16 and VGG19). Its hybrid architecture combining U-Net and ResNet, called as ResUNet indicated better results with an accuracy of 97.3%, against VGG16 at 93.5% and VGG19 at 91.5%. The model also demonstrated good data loss reduction from 0.35 to about 0.10 over the course of a total for around an epoch angle in enduring caching past phase. VGG16 and VGG19 outperformed the baseline performances with reduced data losses while higher accuracies, although their convergence rate was slightly slower than ResUNet. The results confirmed the high precision of liver/tumor segmentation that ResUNet provides, which would be useful for clinical practice to achieve sufficient diagnostic improvement. The architecture of ResUNet specifically designed for the segmentation tasks in medical imaging from this study, demonstrates to be a versatile framework favoring its potential integration into automated tools within oncological diagnostics and could thus radically transform liver cancer detection.