AI - Powered Liver Cancer Diagnosis and Early Detection using Deep Learning & Image Processing

Liver cancer is a highly fatal disease due to its rapid progression and the difficulty of early detection in medical imaging modalities such as CT, MRI, and ultrasound. Subtle tumor characteristics and reliance on manual interpretation often lead to delayed or missed diagnoses. This study presents an AI-powered diagnostic framework for automated liver tumor detection, classification, segmentation, and clinical interpretation. The system employs Convolutional Neural Networks for benign–malignant classification and an Attention U-Net for precise tumor segmentation, supported by advanced preprocessing techniques including noise reduction, CLAHE, and normalization. Radiomic feature extraction and explainable AI methods (Grad-CAM, SHAP, and LIME) are integrated to enhance interpretability and clinical trust. Implemented using Python with a lightweight Streamlit interface, the system is suitable for real-time clinical deployment. Experimental results on LiTS, TCGA-LIHC, and IRCAD datasets achieve up to 95.2% classification accuracy and an 87.6% Dice coefficient, demonstrating the framework’s effectiveness in supporting radiologists and improving early liver cancer diagnosis.