Using AI safely in skin cancer diagnosis

In recent weeks Health Secretary Steve Barclay has announced that the NHS should consider using robots to plug workforce shortages, as the risk of introducing new technologies needs to be judged against the negative impact of staff shortages. In my article on UK cancer care published in Lancet Oncology on 14 December 2022, the reasons why the UK has one of the worst cancer survival rates in OECD countries included an outdated IT infrastructure, inadequate investment in artificial intelligence and inefficient cancer pathways for diagnosis and treatment. 


On 6 December 2022, the British Association of Dermatologists wrote to their members to state that, despite long waiting times for skin cancer diagnosis in NHS and private sectors,  “there is currently no published and independently verified evidence to support the safe and effective use of an AI tool in the skin cancer diagnostic pathway in the NHS” and that current tele-dermatology models, based on lesion assessment by consultant skin cancer specialists, are more likely to triage patients to the correct pathway or discharge them back to the GP. So, how can we reconcile these opposing views, and, is there a way to safely assess newly developed skin cancer AI models within current tele-dermatology frameworks?


Check4Cancer, a UK private early cancer detection company, runs a very successful nurse led tele-dermatology pathway, with consultant skin lesion reporting, for patients with a suspicious skin lesion. With an extremely low onward referral rate (15%) for face-to-face consultation, and a biopsy rate close to 10%, this pathway has significantly reduced skin pathway costs as well as the number of unnecessary biopsies for the patients. With two grants from Innovate UK in 2022, and a Knowledge Transfer Partnership with the University of Essex, Check4Cancer plans to build an AI model using more than 70,000 skin lesion images and associated clinical data to partly automate the reporting process, by clearly identifying the lesions that are not malignant. 


The success of the AI model can be tested prospectively within the current tele-dermatology pathway in pilot studies in the private sector as well as the NHS before wider utilisation. If after rigorous testing, the AI model can accurately classify skin lesions that are clearly benign (not malignant), then the reporting process could be partly automated as a solution to reduce the current long waiting times for skin cancer diagnosis. This approach, where a healthcare provider develops and tests a new technology with academic partners against an existing clinical pathway, has a much greater chance of success than technology companies dabbling in healthcare and can be safely tested across many different cancer pathways before implementation is considered. In this way, we can safely introduce new technologies to speed up cancer diagnosis to improve early cancer detection and cancer survival in the UK.


Professor Gordon Wishart is Chief Medical Officer at Check4Cancer and Visiting Professor of Cancer Surgery at Anglia Ruskin University