An autonomous AI medical device used to triage urgent suspected skin cancer referrals significantly alleviated dermatology service pressures while maintaining a high level of sensitivity for identifying skin cancers. Findings from a real-world U.K. study were presented at the European Academy of Dermatology and Venereology (EADV) Congress 2026 (Abstract P2827).

“Autonomous AI represents a credible and scalable response to increasing referral demand and workforce pressures within dermatology services,” the study authors, led by lead author Lucy Thomas, MPharm, MBChB, MRCP UK, a National Health Service (NHS) Consultant Dermatologist at Chelsea & Westminster Hospital and an honorary clinical lecturer at Imperial College London, wrote in their poster.

“We believe the greatest value of autonomous AI lies not in the technology itself, but in the specialist capacity it unlocks. Every hour saved reviewing low-risk lesions can be reinvested in patients with skin cancer, helping them access timely treatment to improve prognosis, and in patients with severe inflammatory skin disease, where earlier access to specialist care and effective treatments can transform quality of life,” Dr. Thomas stated.

Study and Model Methods

In the U.K., urgent suspected skin cancer referrals have more than doubled since 2009, even though only about 6% result in a true urgent skin cancer diagnosis. Dermatologist roles also remain an unmet need, with about one in four roles in the U.K. unfilled, limiting their capacity and further challenging dermatology services.

Researchers believe that autonomous AI as a medical device could provide a scalable solution, but real-world evaluation is needed. The study sought to evaluate the real-world safety, diagnostic performance, and service impact of using autonomous AI to triage patients within urgent suspected skin cancer pathways.

They used Skin Analytics’ CE-marked Class III autonomous AI as a medical device system, called DERM. In prior studies, DERM showed the ability to identify melanoma in skin lesion images on par with dermatologist accuracy and achieved an area under the receive operator characteristic curve of 95.8%.

After a validation period, researchers implemented the system across two NHS hospital sites. The system implementation was evaluated for 16 months, beginning in December 2024.

In the study workflow, once a patient was referred, they completed their consent, questionnaire, and provided smartphone images of the skin lesions, including dermoscopy, through a photography hub. The patient then went through an assessment by the autonomous AI and was classified as either low risk/benign or high risk plus those with exclusions. From there, low-risk patients were discharged with safety netting advice, and high-risk patients were directed to a teledermatologist to reviewe the images further and either discharged the patient or send them on to a surgical or clinic appointment consult.

Any cases deemed not appropriate for teledermatology or AI-assisted teledermatology, received face-to-face care.

Study Findings

Ninety-four percent of urgent suspected cancer referrals were managed through the AI-assisted teledermatology pathway, while 6% of cases that were not appropriate for teledermatology were handled with face-to-face management. All Fitzpatrick skin types were represented in the cases reviewed.

Eighty-six percent of patients provided consent for autonomous triage.

The AI system discharged 31% of patients at the first hospital site and 25% at the second without clinician review. Teledermatologists discharged another 24% from the first site and 25% from the second.

Autonomous AI reduced the number of patients requiring follow-up care from 27% to 12%. Biopsy rates were also lower with autonomous AI vs standard face-to-face management (43% vs 27%). The study authors suggested that the autonomous AI system improved diagnostic efficiency and reduced unnecessary procedures.

With autonomous AI, the total clinician processing time was approximately 9 minutes per person rather than usual face-to-face appointment time of 20 minutes. The AI-assisted pathway saved approximately 2,851 clinician hours, which amounted to about 8,553 additional face-to-face dermatology appointments during the study period, equaling a capacity gain of 62%. The AI system allowed dermatologists to focus their efforts on higher-risk lesions and more complex cases while low-risk referrals were safely managed through the autonomous pathway.

The researchers noted that six false-negative cases were discharged through the AI-assisted pathway, which were later identified in postmarket surveillance. The six false negatives included five cases of basal cell carcinoma and one melanoma in situ. This resulted in a sensitivity of 98.3% for invasive melanomas (95% confidence interval [CI] = 97.3%–99.0%), 98.4% for squamous cell carcinomas (95% CI = 97.8%–98.8%), and 98.1% for basal cell carcinomas (95% CI = 97.7%–98.5%). The overall specificity was 72.1% (95% CI = 71.7%–72.6%).

“One of the key lessons for us is that deploying an AI system safely isn't a one-off exercise,” said Dr. Thomas. “You need to keep monitoring it, understand when things go wrong, learn from those cases and make sure patients themselves know what to look out for.”

“If these findings are replicated across larger populations and different health-care settings, autonomous AI could become an important part of creating a more sustainable dermatology service—not by replacing dermatologists, but by allowing scarce specialist expertise to be focused where it can make the greatest difference to patients’ lives,” Dr. Thomas concluded.

DISCLOSURES: The study was partly supported by a grant from La Roche-Posay (L’Oreal Dermatological Beauty), and the AI medical device was funded by Chelsea & Westminster Hospital NHS Foundation Trust. For full disclosures of the study authors, visit eadv.org.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.