Abstract

Fluorescence-guided surgery (FgS) is increasingly used across oncologic specialties to enhance intraoperative visualisation of tumour tissue and lymphatic drainage; however, its clinical impact remains limited by heterogeneous tracer uptake, variable signal intensity, and reliance on subjective visual interpretation, leading to inter-operator variability, uncertainty at tumour margins, residual disease, and inconsistent nodal assessment. This narrative review examines the role of artificial intelligence (AI) in addressing these limitations, synthesising evidence published between January 2000 and December 2025 across neuro-oncology, gynaecological oncology, and thoracic oncology. In neuro-oncology, early clinical and preclinical studies have directly evaluated real-time AI-enhanced interpretation of intraoperative fluorescence, including quantitative analysis of 5-aminolevulinic acid (5-ALA) and hyperspectral imaging, providing proof-of-concept evidence that AI can augment margin detection beyond subjective visual assessment. In contrast, gynaecological and thoracic oncology currently lack validated studies in which AI directly interprets intraoperative fluorescence signals, despite fluorescence imaging being clinically established in both fields; instead, AI development in these specialties has progressed primarily in adjacent domains such as radiomics, digital pathology, risk stratification, surgical planning, and intraoperative computer vision, demonstrating technical maturity but limited integration into fluorescence-guided decision-making. Overall, the available evidence supports proof-of-concept feasibility for real-time AI-enhanced fluorescence interpretation in neuro-oncology, while identifying a clear translational gap in gynaecological and thoracic oncology that warrants targeted research to integrate existing AI capabilities into intraoperative fluorescence-guided surgery.