Award

Deep learning for effective triaging of skin disease in the NHS

Description: The problem to be addressed: Skin conditions are extremely common with 1 in 2 of the population affected each year. The health burden on primary care is large with 1 in 5 GP consultations being for skin disease. In addition to GP costs, around £100M per annum in NHS Scotland is spent on secondary care services treating skin disease. Increasing demand for dermatological care outstrips current capacity, exacerbated by vacancies across Primary and Secondary Care. Skin cancer is the commonest human cancer with melanoma the commonest cancer killer in young adults in Scotland. Diagnosis and management of suspected skin cancer represents over 50% of a specialist dermatology service workload with primary care practitioners, often untrained in skin cancer recognition, lacking the necessary experience to distinguish innocent from cancerous lesions. Skin cancer in the UK is increasing in frequency by 5% per year and there is a pressing need to concentrate resources in this area of disease. Why this research is important: Skin disease naturally lends itself to automated image analysis. Lesions can be photographed easily and saved in digital form; NHS Tayside has 20 years of experience of large-scale digital capture of images in primary care, with assessment by secondary care dermatologists providing fast remote diagnosis, resulting in fewer appointments and improved patient satisfaction. Patients are delighted to be reassured about benign lesions, GPs enjoy immediate education from such a system, and fewer patients need to attend hospital appointments. What we aim to achieve: Using our long experience in this area, we now wish to deploy the rich potential of artificial intelligence (AI) systems to assist skin specialists. In a pilot project, funded by Detect Cancer Early, we have developed an AI (deep learning) system able to distinguish many common benign lesions from common skin cancers with state-of-the-art accuracy. However, our preliminary work has used selected and pre-processed image datasets from international research challenges that are not representative of routine NHS data. How we will achieve this: We will develop our technology to achieve diagnostic accuracy with image data from NHS clinics. This next phase will start the integration of a learning AI system into everyday NHS clinical care to refine clinical capability in a real-world setting. Our interdisciplinary group is uniquely positioned to exploit and develop this technology for NHS benefit. Success in this area will be gradual, starting with simple goals such as clinical decision support for the most common benign lesions, but with the potential to become a diagnostic tool for individuals taking their own photos of skin, uploading to a central deep learning tool, and receiving immediate, reliable diagnosis for one of the many thousands of types of skin disease.

Funder: National Institute for Health and Care Research

Amount: £143,300

Funding type: grant

Funder scheme: Artificial Intelligence

Start date: 2021-03-01

End date: 2022-02-28

Investigator: Stephen McKenna (lead)

Institution: University of Dundee

Topic: Cutaneous Melanoma Detection and Management

Subfield: Oncology

Field: Medicine

Domain: Health Sciences