AI-Driven Hyperspectral Imaging in Clinical Medicine: A Critical Appraisal of Translational Readiness

The instruments used for intraoperative guidance have not improved over the years. Traditional margin evaluations fail in 10–20% of cancer surgeries, and histopathology confirmation can take 48–72 h. A promising alternative is hyperspectral imaging (HSI), which allows for real-time, label-free, non-contact, and detailed biochemical characterization of tissues. This review synthesizes 130 studies from 2015 to 2026, identified through a systematic search across multiple databases following PRISMA-ScR guidelines. The studies cover various clinical areas, including surgical oncology (53%), neurosurgery (10.8%), perfusion assessment (6.2%), gastroenterology (5.4%), dermatology (4.6%), and emerging domains (20%). Each clinical area is evaluated using a four-part Translational Readiness Appraisal framework. This framework shifts the focus from technical proficiency to deployment evidence by examining hardware compatibility with operating rooms, real-time workflow integration, cross-site validation, and regulatory readiness. Since 2015, the combination of clinically approved HSI hardware and artificial intelligence has led to a tenfold increase in publications. However, a significant gap remains: none of the reviewed studies meet the evidence criteria necessary for regulatory approval. Of the studies, 48.5% show strong single-site validation, but 50.8% remain in the proof-of-concept stage. The heavy reliance on intra-patient validation tends to inflate performance estimates, leaving issues like calibration across devices and regulatory pathways largely unaddressed. HSI has demonstrated technical maturity in several areas, particularly in head and neck margin assessment and brain tumor delineation. The primary reason for the lack of clinical adoption is not imaging performance but rather the absence of multicenter validation and a formal regulatory evidence framework.