BT30 Artificial intelligence triage is of greater benefit for screening suspected melanoma referrals than squamous cell carcinoma referrals

Abstract There has been a rapid expansion in the use of artificial intelligence (AI) models to aid in diagnosis and triage of skin cancer in the UK in the last few years. This has been driven by a need to manage an increasing number of urgent referrals for suspected skin cancer. Our service has been using one such platform to screen out benign lesions, thus reducing the number of patients subsequently requiring a face-to-face appointment with a trust dermatologist. The AI platform assesses the features of a single dermoscopic image to categorize the lesion into benign or malignant. Information from patient history and macroscopic images are not involved. As dermatologists, we recognize the importance of history (pain, rapid growth) and clinical examination (induration, tenderness) as well as dermoscopic assessment in reaching a diagnosis. We suspected that the use of dermoscopy in clinical practice is of greater value for distinguishing benign from malignant pigmented lesions compared with keratotic lesions. When a suspected skin cancer is referred into our service, the general practitioner is required to state whether they suspect melanoma or squamous cell carcinoma (SCC). (Basal cell carcinomas are referred on a separate, routine pathway). We sought to use this information to assess for any differences in outcomes. A retrospective review of patients assessed by the AI platform was conducted between August 2023 and November 2023. In total 800 patients were included in the study (396 melanoma pathway, 404 squamous cell carcinoma pathway). In total 116 referrals were discharged by the AI platform on the melanoma pathway (29.2%) and 32 referrals were discharged by DERM on the SCC pathway (7.9%). These results support our hypothesis that AI is better at recognizing benign pigmented lesions, as 92% of patients referred on the SCC pathway required further assessment by a trust dermatologist compared with 70% of those referred on a melanoma pathway. This information is of importance to commissioners planning to implement skin cancer AI triage models.