Researchers in the laboratory of Xinming Huang who are developing a tool to analyze skin images for signs of skin cancer have finished third in an international artificial intelligence (AI) competition aimed at improving early diagnosis of melanoma, a potentially lethal type of skin cancer.
The team’s model combined two types of AI to finish third in the competition, which was organized by the International Skin Imaging Collaboration (ISIC). The challenge drew 181 teams of academic and corporate researchers.
“Melanoma can be difficult for primary care doctors, especially those who are not skin experts, to identify in its early stages,” says Huang, a professor in the Department of Electrical and Computer Engineering . “AI has the potential to help those physicians spot patients who need to be referred to specialists.”
TwinSight, the model under development in Huang’s lab, builds on research that PhD student Antony Garcia Gonzalez and Huang reported in 2025 in the peer-reviewed journal Cancers . TwinSight analyzes the visual characteristics of a lesion using both a convolutional neural network and a vision transformer, two different deep-learning computer programs that examine complex patterns in data.
The TwinSight model seeks to address a critical problem: Melanoma, especially in its early stages, can be difficult to visually distinguish from benign skin moles, and delays in diagnosis and treatment raise the risk that the cancer will spread.
Primary care physicians will often refer patients with suspicious lesions to dermatologists, who will typically examine the lesions using a dermoscope, a magnifying tool, and perform a skin biopsy to confirm the diagnosis.
In 2024, U.S. regulators approved an AI-powered tool for use in evaluating skin lesions and providing information to primary care physicians. Yet developing an AI tool that can match the skill of a dermatologist remains elusive. The WPI team says that, in fact, they do not expect AI to replace dermatologists.
“We are aiming to create a handheld tool that primary care physicians could use during a routine visit to image a suspicious lesion and get an immediate AI assessment,” says Garcia, a Fulbright scholar who is also a research associate on the electrical engineering faculty at the Technological University of Panama. “That could help them decide which patients to refer to dermatologists and potentially lead to earlier diagnosis of skin cancer.”
AI models need large amounts of data to identify patterns in images. Research teams participating in the ISIC challenge each received a dataset of 5,294 anonymous images to train their models and a second dataset of 479 anonymous images to analyze.
The WPI team also used a dataset of 187 images, obtained from a hospital in Florida, to validate their model. In the ISIC competition, TwinSight earned a score of 0.662 on the Dice coefficient, the measure the challenge used to rank how closely each model’s diagnoses matched those of experts.
The WPI team is continuing to develop TwinSight for potential use in clinical settings. Huang says the team envisions a handheld device tethered to a laptop or computer tablet. A clinician would place the device on a patient’s skin to collect a finely detailed photograph that would be analyzed with AI. The team is collaborating with dermatologists in Florida to develop and test the tool.
“The ultimate goal is to help doctors make faster and more confident decisions when they are examining patients,” says Huang. “To do that, AI tools need to evolve.”