International Journal of Scientific Research and Engineering Development

Medical artificial intelligence has achieved some amazing progresses over the last few years, but the systems deployed can only capture a single stream of data, typically radiology data, and do not leverage the more powerful context traditionally provided by clinicians: patient history, clinical notes, lab data, and demographic meta data. The overview examines five significant scholarly publications released in the years of 2019-2022 and delimits the current situation in multi-modal deep learning in medical diagnostics. We contrast the specifications of dataset creation, the encoding of the modality and prototyping of features, in both studies, on regular performance measures. The findings are congruent: multi- modal models are more precise and have higher AUC rates by 6- 10 percentage points compared with the single- modality models, and the largest improvements in the cases with the diagnostically ambiguous cases and those where the visual information alone is not sufficient to provide an accurate diagnosis. We introduce lasting challenges like modality alignment, data unavailable in inference time, interpretability, and demographic bias, and indicate the most promising future directions of practice.

Shifa Bilal Tamboli, Simeen Phiroj Mulani, Arman Tajuddin Shiakh,"Multi-Modal Medical Dataset Creation and Deep Learning for Improved Neural Network Diagnostics: A Comprehensive Literature Review" International Journal of Scientific Research and Engineering Development, V9(2): Page(2075-2080) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.