SRS319 - The role of AI and imaging in early skin cancer diagnosis: a systematic review

Abstract Aims Skin cancer is one of the most common malignancies, with >17 000 melanoma diagnoses annually in the UK (2016–2019). This review assessed non-invasive diagnostic methods for skin cancer and their potential role as alternatives or adjuncts to clinician face-to-face assessment. Methods A literature search of PubMed, Medline, and Embase identified studies reporting diagnostic accuracy, sensitivity, specificity, and outcomes (N = 208). Results Lesions were classified as malignant or as melanoma, squamous cell carcinoma (SCC), or basal cell carcinoma (BCC). Reflectance Confocal Microscopy (RCM) demonstrated high sensitivity for BCC (0.911) and, when combined with dermatoscopic interpretation, achieved sensitivity of 0.948 for malignancy and 0.931 for melanoma. However, application is limited by small study sizes, high cost, and the need for specialist expertise. Artificial intelligence (AI) analysis of dermatoscopic images outperformed clinicians for diagnosing malignancy (AI versus all clinicians: 0.837 versus 0.777, P < 0.0001; AI versus specialists: 0.837 versus 0.744, P < 0.0001), melanoma (0.869 versus 0.802 and 0.869 versus 0.819, both P < 0.0001), and BCC (0.891 versus 0.847, P < 0.0001). AI using both clinical and dermatoscopic images achieved higher sensitivity for melanoma compared with combined clinician and dermatoscopic assessment (0.869 versus 0.802, P < 0.0001) and experts (0.869 versus 0.819, P < 0.0001), though evidence remains limited for BCC and SCC. Conclusions AI-based diagnostic tools can improve skin lesion triage, referral optimisation, clinical decision making and remote assessment, particularly in primary care settings.