An Ensemble Deep Learning Framework for Automated Detection, Segmentation, and Classification of Liver Cancer in CT and MRI Images

Liver cancer, particularly hepatocellular carcinoma (HCC), poses a significant global health challenge due to late diagnosis and limited accuracy of conventional imagingbased assessments. This study proposes a novel ensemble deep learning framework for the automated detection, segmentation, and classification of liver cancer using CT and MRI images. The model integrates U-Net++/ResNet50, Swin Transformer, and DenseNet121 architectures in a multi-phase pipeline to enhance tumor localization, boundary segmentation, and malignancy classification. Adaptive weighted fusion is employed to optimize ensemble performance, while Grad-CAM and SHAP-based visualizations ensure interpretability for clinical decision-making. Experimental results on benchmark datasets LiTS demonstrate that the proposed model achieves superior accuracy, with a Dice score of 0.92 and classification accuracy exceeding 96%, outperforming state-of-the-art single models. This ensemble framework provides a robust, explainable, and generalizable approach for AI-assisted liver cancer diagnosis and can significantly contribute to precision oncology and radiological automation.