Skin cancer is a pervasive global health problem, with melanoma being the most deadly subtype. Responsible for 80% of all skin cancer deaths, melanoma highlights the urgent need for early and accurate detection. Research shows that delays in diagnosis can reduce five-year survival rates by up to 20%.
However, traditional diagnostic methods, such as the widely used 7-point checklist (7PCL), mainly focus on melanoma. This narrow scope has limited their effectiveness in detecting other types of skin cancers such as basal cell carcinoma (BCC) and squamous cell carcinoma (SCC).
A groundbreaking study led by researchers from Anglia Ruskin UniversityTHE University of EssexCheck4Cancer and Addenbrooke’s Hospital introduce a revolutionary approach. Their artificial intelligence (AI)-based framework not only extends detection capabilities to all types of skin cancer, but also significantly outperforms existing methods in terms of sensitivity and specificity. This innovation is expected to redefine the way skin cancer is identified, managed and treated.
In the UK, skin cancer diagnosis follows a two-week pathway system for urgent referrals. This process prioritizes patients with suspicious lesions for specialist evaluation within two weeks. Although effective in theory, the system faces significant challenges in practice.
Referrals for skin lesion testing have increased significantly, from 159,430 in 2009 to more than 506,000 in 2020. Non-urgent cases, such as those involving BCC, often experience wait times of 18 weeks or more. Only 80% of non-urgent referred patients are seen within this target time.
To further complicate matters, the COVID-19 pandemic has disrupted healthcare systems, leading to delays and a significant delay in diagnosis. An estimated 17% of melanoma cases in Europe were diagnosed later after lockdown, highlighting the crucial need for faster and more accurate detection tools. Current methods, such as the 7PCL and the Williams score, are found to be inadequate, with sensitivity levels of only 62% and 60%, respectively.
Recognizing these shortcomings, researchers have turned to AI, leveraging its power to analyze large data sets and identify patterns that traditional methods overlook. The result is a comprehensive framework that uses clinical metadata to detect suspicious lesions with unprecedented accuracy.
At the heart of this new framework are the C4C risk factors, a set of seven clinical characteristics that predict skin cancer risk across all subtypes. These factors include lesion characteristics such as size, color, shape, and inflammation, as well as patient-specific attributes such as natural hair color at age 15 and age of the lesion. Unlike traditional methods, which focus exclusively on melanoma, C4C risk factors consider non-pigmented lesions and other forms of skin cancer.
Using machine learning, researchers distilled 22 clinical characteristics into these seven key factors, applying proportional weighting to create the C4C risk score. This score achieved a sensitivity of 69%, significantly outperforming the 7PCL and Williams scores. Professor Gordon Wishart, Chief Medical Officer of Check4Cancer, highlighted the importance of this breakthrough:
“This study shows the importance of using clinical data in the classification of skin lesions, which should help improve the detection of skin cancer. »
The researchers’ work involved analyzing metadata from more than 53,600 skin lesions collected across the UK. This large dataset enabled the development of an AI framework capable of classifying lesions as suspicious or non-suspicious with remarkable accuracy.
One of the most significant findings of the study is the performance improvement achieved by combining C4C risk factors with traditional methods such as the 7PCL and Williams score. This fusion resulted in the highest overall sensitivity and specificity, making it the most reliable diagnostic tool to date.
The AI model also demonstrated its potential to reduce unnecessary biopsies and streamline diagnostic times. By integrating clinical metadata with image analysis, it offers a scalable solution that can be implemented in teledermatology, particularly beneficial in remote or underserved areas.
Consultant plastic surgeon Per Hall highlighted the wider implications of this work:
“In the past, the focus has been on pigmented lesions and melanomas, but other things develop on the skin and need to be treated, such as basal cell carcinomas and squamous cell carcinomas. This work eliminates potentially serious lesions while identifying those whose skin is more prone to the development of cancer.
The introduction of the C4C framework marks a pivotal moment in skin cancer detection. With regulatory approval expected by 2025, this AI-driven approach promises to ease the burden on healthcare systems while improving patient outcomes.
Skin cancer referrals are expected to increase in the coming years due to an aging population and increased awareness. Innovations such as the C4C Risk Score can help manage this demand by providing accurate and efficient diagnostic tools that prioritize high-risk cases.
Professor Wishart expressed optimism about the future impact of this research:
“Our new AI model could lead to a reduced need for patient referral for biopsies, shorter wait times for skin cancer diagnosis and treatment, and better patient outcomes. »
This study represents a collaboration of experts from multiple institutions, highlighting the value of interdisciplinary research to address complex healthcare challenges. Part-funded by a Knowledge Transfer Partnership (KTP) grant from Innovate UK, the research has been published in Scientific reports.
The C4C framework is more than a technological success; it’s a step toward transforming skin cancer care. By combining clinical expertise with the analytical power of AI, this approach sets a new standard in early detection and comprehensive care.