Liver cancer remains one of the most fatal malignancies worldwide, with high mortality rates largely attributed to late-stage diagnosis and limited accuracy in conventional imaging-based interpretation. Recent advances in deep learning have significantly improved the capabilities of automated medical image analysis, particularly in tumor segmentation and classification tasks. This study proposes a hybrid deep learning framework designed for automated liver tumor segmentation and malignancy prediction using computed tomography (CT) imaging data. The framework integrates convolutional neural networks (CNNs) for spatial feature extraction with deep belief networks (DBNs) and multi-classifier systems for enhanced diagnostic reasoning. Drawing upon established architectures such as U-Net and deep CNN variants, the proposed system emphasizes precision segmentation and robust classification under variable imaging conditions. The study critically synthesizes prior research in medical image analysis and demonstrates how hybridization improves diagnostic performance compared to single-model approaches. Experimental insights suggest that combining feature-rich segmentation networks with probabilistic classification models enhances sensitivity and specificity in tumor detection. The framework is further evaluated in the context of clinical decision support systems, highlighting its potential to assist radiologists in early liver cancer detection and malignancy grading. Limitations such as dataset variability and computational complexity are also discussed, alongside future directions for multi-modal integration.