Artificial Intelligence for Precision Medicine: A Review from Medical Imaging and Physiological Signals to Genomics, Drug Discovery and Trustworthy Deployment

DOI

https://www.doi.org/10.59256/ijsreat.20260606020

Published In

Volume 6, Issue 6 (2026)

Published Online

November-December 2026

Pages

139-145

Article Type

Review Article

Check for updates

Precision medicine aims to fit diagnosis and treatment to the individual patient, and artificial intelligence has become one of its main tools. Deep networks now read computed tomography scans, mammograms, skin images and echocardiograms; classifiers interpret electrocardiogram and electroencephalogram signals; and learned models predict protein structures and drug–target interactions. The problem is that these advances are reported in separate literatures, each with its own data types, metrics and failure modes, and the barriers to clinical use, such as privacy, bias and security, are usually discussed apart from the models they constrain. This review brings imaging, physiological signals, clinical risk prediction, genomics and drug discovery, and trustworthy deployment into one framework. It organizes 37 studies into a five-layer taxonomy and compares their data, methods and reported results. Imaging studies report expert-level performance, including an area under the curve of 0.944 for lung cancer screening and fewer false positives and false negatives in breast screening, but also show weaker performance on darker skin tones. Signal studies report a deep network that exceeded the average cardiologist in arrhythmia detection, and automated machine learning improved cardiovascular risk prediction over the Framingham score. Genomic and drug discovery studies show that deep models can predict protein structure with near-experimental accuracy and can find new antibiotics. Deployment studies show that federated learning across 20 institutions improved average accuracy by 16% over single-site models, while privacy, security and explainability remain open. Few studies combine more than one data layer, and fairness is rarely measured. Future work should develop multimodal models trained under privacy guarantees and evaluated across diverse patient groups.