MULTI-MODEL AI PIPELINE FOR EARLY-STAGE LIVER CANCER DIAGNOSIS USING CLINICAL AND IMAGING DATA

Early-stage liver cancer diagnosis remains a significant clinical challenge due to subtle disease presentation,
heterogeneous risk factors, and variability in imaging interpretation. This study proposes a robust multi-model
artificial intelligence (AI) pipeline that integrates clinical parameters and medical imaging data to enhance the
early detection of liver cancer. The pipeline combines structured clinical data such as patient demographics,
laboratory values, and risk indicators—with imaging features extracted from liver ultrasound, CT, or MRI scans.
Multiple machine learning and deep learning models are employed at different stages of the workflow, including
feature engineering, modality-specific prediction, and decision-level fusion. Clinical data are analyzed using
gradient-boosting and ensemble classifiers, while convolutional neural networks are utilized for automated
imaging feature extraction and lesion characterization. Outputs from individual models are fused using a metalearning framework to generate a unified diagnostic prediction. The proposed system is designed to improve
diagnostic accuracy, sensitivity, and robustness compared to single-model or single-modality approaches.
Experimental evaluation demonstrates that the multi-model pipeline achieves superior performance in identifying
early-stage liver cancer, particularly in cases with ambiguous imaging findings or incomplete clinical records. In
addition, the framework supports explainability by highlighting clinically relevant features and imaging regions
contributing to the final decision, facilitating clinician trust and adoption. This integrated AI-driven approach has
the potential to assist radiologists and hepatologists in early diagnosis, reduce inter-observer variability, and
support timely clinical intervention. The proposed pipeline represents a scalable and adaptable solution for
precision diagnostics in liver oncology and can be extended to other multimodal medical decision-support
applications.