Advancing Early Disease Detection Using Multimodal Machine Learning Models Integrating Imaging, Genomics, and Clinical Data Sources

Abstract

Background: Early and accurate disease detection is critical for improving patient outcomes, yet conventional diagnostic approaches often rely on isolated data sources, which may provide an incomplete clinical picture. Multimodal machine learning (MML), which integrates diverse data types like medical imaging, genomics, and clinical records, holds promise for a more holistic assessment. However, the comparative performance of these integrated models against standard single-source approaches has not been systematically evaluated. Objective: This systematic review aimed to determine whether MML models for early disease detection yield superior accuracy and diagnostic reliability compared to unimodal models. Methods: A systematic search was conducted in PubMed, Scopus, Web of Science, and the Cochrane Library for studies published between 2019 and 2024. The review included comparative studies that directly evaluated MML models (integrating at least two of: imaging, genomics, clinical data) against unimodal models for disease detection in human patients. Study selection, data extraction, and risk of bias assessment using a modified QUADAS-2 tool were performed in duplicate. Results: Eight studies met the inclusion criteria, encompassing diseases in oncology, neurology, and cardiology. All eight studies reported a statistically significant improvement in detection performance for multimodal models. The most common metrics showed MML models achieving absolute increases in the Area Under the Curve (AUC) of 0.04 to 0.10 over the best unimodal comparator. The greatest performance gains were observed in complex diseases like Alzheimer's and lung cancer. The main limitations were heterogeneity in data fusion techniques and a risk of bias from non-independent model tuning. Conclusion: The consistent findings across diverse clinical domains indicate that integrating multimodal data significantly enhances the accuracy of machine learning models for early disease detection. This evidence supports the paradigm of MML as a superior analytical framework. Future work should focus on standardizing validation practices and demonstrating generalizability in real-world clinical settings to facilitate translation into practice.