Dermalyze: uma aplicação para auxílio à triagem de lesões de pele baseado em aprendizado profundo
Year: 2025
Type: conference-paper
Abstract: According to the Brazilian Institute of Cancer, skin cancer is the most common displasya in the country. To mitigate this situation and ensure quality of life for those affected by the disease, early diagnosis is fundamental. When treated early, 5-year survival rates can exceed 96%. In this context, we developed a skin lesion screening application to assist in the identification of skin cancer. This software, designed for use by healthcare professionals, leverages artificial intelligence to assess the cancerous potential of a lesion based on a clinical image captured by a mobile device. The developed model achieved a balanced accuracy of 80%, and the application in which it was embedded was tested in a clinical setting with a total of 82 users. Based on the predictions made during the case study, the model was able to correctly identify the severity of the lesions in approximately 70% of the cases. Additionally, about 92% of users reported the model to be useful during the screening process, describing their experience primarily with words such as comfort and safety.
Institution: Universidade Federal do Espírito Santo
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4416744467