Automated Liver Tumor Detection Using Deep Learning for Accurate Segmentation and Boundary Localization

Liver cancer, predominantly due to hepatocellular carcinoma is one of the highest ending causes for cancer-related deaths across the globe. Proper identification and segmentation of liver tumors is a vital step in clinical diagnosis, treatment preparation, and patient monitoring. We propose a deep learning-based liver tumor detection tactic with an emphasis on segmentation (segmenting the liver tumor) and boundary localization. Liver cancer remains one of the most prevalent cancers globally, contributing significantly to cancer-related deaths. This work highlights the limitations of manual diagnostic processes and the potential of automation to address the growing demand for accurate and scalable solutions. By incorporating attention mechanisms into the U-Net architecture, we aim to enhance segmentation precision and reduce diagnostic errors, offering a clinically viable solution that aligns with current technological advancements in medical imaging. We propose U-Net and Fully Convolutional Networks (FCN) combined with an attention mechanism that increases segmentation accuracy by decreasing false positives and decrease false negatives. We conduct extensive experiments on liver scans dataset and show that our approach outperforms the traditional Machine Learning tactics by achieving a Dice Similarity Coefficient of 0.89 and an Intersection over Union (IoU) of 0.85. Deep learning techniques have demonstrated promising outcomes in the realm of medical image analysis, particularly in diagnosing liver tumors.