AI Model Predicts Lung Cancer Risk Using Multimodal Data.

A new research paper introduces LUNG-KGMM, an advanced AI framework designed to predict the risk of lung cancer. This model uniquely combines various data sources, including longitudinal electronic health records, detailed radiology reports, visual representations from chest radiographs, and structured clinical knowledge derived from guidelines. The goal is to identify individuals at risk for lung cancer up to six years before diagnosis, enabling earlier intervention. The framework addresses challenges like data heterogeneity and potential leakage by employing a specialized report processing pipeline and a training objective that accounts for incomplete follow-up data. It also incorporates clinical guidance as an auditable knowledge stream, enhancing transparency. Tested on public MIMIC databases and a real-world Xiamen cohort, LUNG-KGMM consistently outperformed state-of-the-art prediction methods, demonstrating its effectiveness and potential for broader application, though local adaptation may be necessary.