A new study from the University of Gothenburg shows that AI can assess the aggressiveness of squamous cell carcinoma in skin cancer as accurately as experienced dermatologists.
A recent study led by the University of Gothenburg has shown that a simple artificial intelligence model can perform on par with experienced dermatologists in assessing the aggressiveness of squamous cell carcinoma, a common form of skin cancer. This discovery could herald a new era in cancer diagnosis and treatment, potentially streamlining the process and ensuring timely interventions.
More than 10,000 Swedes are diagnosed with squamous cell carcinoma each year, making it the second most common form of skin cancer in the country after basal cell carcinoma. The disease typically develops in sun-exposed areas like the head and neck, often due to prolonged UV radiation exposure.
Lead researcher Sam Polesie, an associate professor of dermatology at the University of Gothenburg, described the nature of squamous cell carcinoma as a result of cumulative UV radiation, affecting fair-skinned adults significantly.
“This type of cancer, which is a result of mutations of the most common cell type in the top layer of the skin, is strongly linked to accumulated UV radiation over time. It develops in sun-exposed areas, often on skin already showing signs of sun damage, with rough scaly patches, uneven pigmentation and decreased elasticity,” Polesie, who is also a practicing dermatologist at Sahlgrenska University Hospital, said in a news release.
Diagnosing squamous cell carcinoma is generally straightforward, but preoperative assessments to determine tumor aggressiveness are more challenging. Aggressive tumors require prompt surgical intervention with wider margins, while less aggressive ones can be managed with simpler procedures.
Traditionally, preoperative punch biopsies for suspected squamous cell carcinoma are not routine in many countries, including Sweden. Instead, surgeries are performed based on clinical suspicion, with the excised specimens later analyzed histopathologically. This practice underscores the need for non-invasive assessment methods, such as AI-based image analysis.
The research team trained their AI system on 1,829 clinical close-up images of confirmed squamous cell carcinoma cases. They then tested the AI model on 300 images, comparing its performance to that of seven experienced dermatologists.
The results, published in the Journal of the American Academy of Dermatology International, indicated that the AI’s performance was almost identical to that of the dermatologists.
Interestingly, while the AI model’s assessments were nearly indistinguishable from the experts, agreement among the individual dermatologists was only moderate. This discrepancy highlights the complexity of the task and the potential for AI to bring consistency to the diagnostic process.
The study also pinpointed two clinical features — ulcerated and flat skin surfaces — as indicators of more aggressive tumor growth. Such tumors were more than twice as likely to be categorized into the higher levels of aggressiveness.
AI has generated significant interest in skin cancer care in recent years. However, its practical impact in health care has been limited thus far, according to Pelosie. He stresses the need for clearly defined application areas where research can deliver real value for Swedish health care.
“We believe that one such application area could be the preoperative assessment of suspected skin cancers, where more nuanced conclusions can influence decisions. The model we’ve developed needs further refinement and testing, but the way forward is clear — AI should be integrated where it actually adds value to decision-making processes within healthcare,” Polesie added.
Source: University of Gothenburg