Deep multistage System for Liver and Lesion Segmentation Using Computed Tomography Images

Accurate tumor segmentation plays a pivotal role in cancer diagnosis and treatment planning. This study introduces an automated system for segmenting liver and tumors in computed tomography (CT) images through a multistage process. Initial preprocessing improves image quality, followed by feature extraction using the pretrained VGG16-SegNet and FCN-AlexNet models. Outputs from both models are combined using a parallel fusion operation. Finally, pixel-wise classification designates pixels as liver, tumor, or background. Evaluation on the MICCAI'2017 LiTS database yielded a Dice coefficient of 95.3 % for the liver and 78.1 % for tumors using fivefold cross-validation. Comparative studies highlight the superior accuracy of our approach in liver and tumor segmentation, promising advancements in clinical diagnosis and treatment strategies.