BT05 A retrospective analysis of the use of Deep Ensemble for Recognition of Malignancy (DERM) for triaging skin lesions

Abstract

In 2025, NICE recommended Deep Ensemble for Recognition of Malignancy (DERM) to assess and triage skin lesions in adults referred to the urgent suspected skin cancer pathway [NICE. Artificial intelligence (AI) technologies for assessing and triaging skin lesions referred to the urgent suspected skin cancer pathway: early value assessment. Available at: https://www.nice.org.uk/guidance/htg746 (last accessed 5 March 2026)]. At our trust, a clinician reviews all triage decisions by DERM. Based on NICE recommendations, a proposal was made to remove the clinician review of DERM triage decisions for benign lesions. This retrospective analysis was conducted to assess the safety of this proposal. In July 2025, DERM AI analysed over 652 cases, of which 193 were reported as benign. Dermatology clinicians overturned 59 (31%) of the benign cases to requiring a face-to-face review at the trust. We analysed the outcomes of the 59 overturned cases. In total, 18% of the overturned cases were recommended for an urgent suspected cancer pathway appointment, and of these seen, 29% proceeded to excision of the lesion. Importantly, two of the overturned cases – assigned a benign diagnosis and recommended discharge by DERM AI – had a histological diagnosis of malignancy: superficial spreading melanoma and micronodular basal cell carcinoma (BCC). Another 6% of patients had clinical diagnoses (without histology) of BCC, actinic keratosis and Bowen disease. As a result of our findings, our trust has chosen not to automate our skin cancer triage pathway and will keep the clinician in place to support assessment and triage of benign and malignancy lesions. Previous analysis of DERM-vB AI tool showed that no skin malignancies were missed (Thomas L, Hyde C, Mullarkey D et al. Real-world post-deployment performance of a novel machine learning-based digital health technology for skin lesion assessment and suggestions for post-market surveillance. Front Med 2023; 10: 1264846).