New AI tool to help reduce skin cancer waiting times

Artificial intelligence is set to revolutionise the way skin cancer can be detected after researchers found the disease can be identified through skin lesion descriptions and a patient’s personal characteristics.

A first of its kind study led by the University of Essex and Check4Cancer has led to the development of a new AI framework which is able to detect 85 per cent of skin cancer cases when combined with some of the existing assessment metrics.

It is hoped the new framework, known as the C4C risk score, could be used to triage patients quicker, as well as reduce waiting lists by identifying non-suspicious lesions at the earliest possible opportunity.

The new AI framework, which was also developed with the help of Anglian Ruskin University and Addenbrookes Hospital, is thought to be one of the first ever based on patient data alone and could prove to be a vital tool for easing pressure on the health system.

Lead researcher Dr Haider Raza, from Essex’s School of Computer Science and Electronic Engineering, said: “We have developed an AI framework solely based on metadata and observed that it can separate suspicious skin lesions from non-suspicious ones with a high sensitivity.

“The C4C risk score can be used as a decision-aid by telemedicine reporters to help with final lesion classification that is equivocal after image classification alone.

“This has the potential to reduce the number of referrals to a special clinic for possible biopsy and help reduce the waiting times for skin cancer diagnosis.”

The project, outlined in the journal, Scientific Reports, saw researchers analyse 53,601 pieces of metadata from 25,105 patients who attended Check4Cancer’s private skin cancer diagnosis clinics between 2015 and 2022.

Using machine learning, artificial intelligence is able to use previous data to predict whether a person has skin cancer based on descriptions of skin lesions, such as size, pinkness and age.

The model also uses characteristics such as a person’s age, gender, hair colour, and family history to conclude whether a lesion is malignant.

The new C4C risk score has a balanced accuracy of 71 per cent, outperforming the current assessment methods 7PCL and Williams.

The sensitivity score, the ability to identify positive cases, rises to 85 per cent when combined with some factors from 7PCL and Williams.

Researchers are also in the final stages of adding an image assessment method to enhance the accuracy of the C4C risk score.

The new assessment method will be added to the C4C smartphone app, making it quicker and easier for people to have suspicious lesions assessed.

Gordon Wishart, Founder and Chief Medical Officer of Check4Cancer, said: “The clinical data input to Check4Cancer’s AI model for skin lesion classification has led to a significant gain in model performance compared to image assessment alone.

“This fused AI model can be used as a decision aid or to partly automate teledermatology triage for patients with a worrying skin lesion, leading to shorter times to skin cancer diagnosis and treatment.”

The study formed part of a Knowledge Transfer Partnership between the University of Essex and Check4Cancer, funded through Innovate UK.