Utility of Deep Learning to Address Missing Modalities from Multi-Modal Medical Imaging Studies: A Systematic Review

Missing modalities pose a significant challenge on multi-modal studies by disrupting the comprehensive analysis of diverse data sources. Deep learning addresses this issue by employing algorithms that can effectively infer and integrate the absent information, thereby ensuring robustness and accuracy of the models while increasing the study's statistical power. This study aims to provide a systematic literature review on deep learning solutions for missing imaging modalities in multi-modal medical data analysis. Articles on PubMed, IEEE explore digital library, and Scopus were searched in the range from January 2013 to May 2025. This systematic search and review identified 234 articles. Adhering to the specified search criteria, 61 published studies were eligible. Among these, 47% employed image synthesis methods, 20% applied knowledge transfer methods, and 33% used latent feature space-based methods. The paper explores the research gaps and challenges associated within each of these categories. Additionally, this review paper illuminates the popular public datasets for multi-modal studies with missing modalities. Furthermore, it presents evaluation metrics and their key attributes. The review concludes with its limitations and a detailed discussion of current challenges and future directions in this domain.