Artificial Intelligence Matches Dermatologists in Evaluating Skin Cancer Severity

A recent study conducted by researchers at the University of Gothenburg has demonstrated that a straightforward AI model can perform nearly on par with experienced dermatologists in assessing the aggressiveness of squamous cell carcinoma, a common type of skin cancer. This advancement marks a significant step toward integrating artificial intelligence into dermatological diagnostics.

Each year, over 10,000 individuals in Sweden develop squamous cell carcinoma, making it the second most common skin cancer in the country after basal cell carcinoma. This form of cancer typically appears in sun-exposed areas such as the head and neck, often evolving over years due to cumulative UV damage. The disease stems from mutations in the skin's most prevalent cell type, with UV radiation playing a crucial role. Clinically, the cancer presents as rough, scaly patches that may have uneven pigmentation and diminished skin elasticity.

Diagnosing squamous cell carcinoma is generally straightforward; however, accurately assessing how aggressively the tumor is growing preoperatively remains challenging. This assessment influences surgical planning—more aggressive tumors require wider excisions, while less aggressive ones can be treated more conservatively.

In Sweden, routine preoperative punch biopsies are not standard practice. Instead, clinicians rely solely on visual examination, and the entire tumor is excised for histopathological analysis post-surgery. This approach emphasizes the necessity for alternative diagnostic methods that do not depend on tissue samples, such as image analysis powered by artificial intelligence.

In the study, the researchers trained an AI system on 1,829 clinical close-up images of confirmed squamous cell carcinoma cases. The AI was tested on 300 new images to evaluate its ability to recognize three levels of tumor aggressiveness. Its performance was then compared to assessments made independently by seven experienced dermatologists. The findings, published in the Journal of the American Academy of Dermatology International, revealed that the AI's accuracy was nearly identical to that of the dermatologists. Interestingly, the agreement among the dermatologists themselves was only moderate, highlighting the complexity of visual tumor assessment.

The study also identified that tumors with ulcerated or flat surfaces were significantly more likely to be highly aggressive, with these features doubling the risk of higher tumor grades.

While the integration of AI into skin cancer diagnostics has garnered considerable interest, its practical application in healthcare has been limited so far. Dr. Sam Polesie from the University of Gothenburg stresses that AI's potential should be harnessed in well-defined areas where it can add tangible value, such as preoperative tumor assessment. He emphasizes that further refinement and testing are necessary before wider adoption.

This research underscores the promising role of AI in aiding clinical decision-making, potentially leading to more precise and timely treatments for skin cancer patients.