Multi-Phase Attention for Early Stage Hepatocellular Carcinoma Segmentation from Contrast Enhanced Liver MRI

Hepatocellular carcinoma (HCC) is a leading cause of cancer-related deaths globally, primarily due to late-stage detection. Accurate segmentation of HCC from multiphase MRI scans plays a critical role in early detection and treatment planning, but manual segmentation is labor-intensive owing to the heterogenous and diffuse nature of these tumors. In this study, we propose a novel deep learning framework that leverages multi-phase MRI data to perform early stage HCC segmentation. Our method introduces a Multi-Phase Attention (MPA) module to effectively capture and integrate information from arterial, venous, and delayed post-contrast phases. We compare our approach against five state-of-the-art medical image segmentation models, demonstrating its ability to outperform these baselines on small early stage tumors. Our experiments on a real-world pilot clinical dataset of liver transplant patients illustrate the robustness and generalizability of our model, offering a promising tool for improving HCC detection and segmentation accuracy in clinical practice.