AI tool designed and integrated into hospital workflows to classify skin lesions

Skin cancer claims nearly 1.5 million new cases annually, according to the World Health Organization. The University of Alicante and Sant Joan d'Alacant University Hospital have teamed up to create MEL-IA, an AI system aimed at automating skin lesion classification. Early detection of skin cancer is crucial for better clinical outcomes, and MEL-IA supports healthcare staff in evaluating lesions while working within the hospital's existing IT systems.

MEL-IA consists of a mobile app to capture images of skin lesions and clinical data, an AI model for classification, and complex integration models that securely connect the data with hospital systems. The study, published in the Journal of Medical Systems, was led by Alberto de Ramón, Daniel Ruiz, and Marcelo Saval, along with Pablo Candela. The Sant Joan d'Alacant University Hospital's IT Service contributed through José María Salinas and Diego Guijarro.

The technology encompasses image capture, AI analysis, secure storage, and confidential exchange of clinical data. The system can categorize skin lesions into five major types: melanoma, nevus, basal cell carcinoma, actinic keratosis, and benign keratosis. The dataset for training and validation included over 15,000 dermatoscopic images and clinical patient data, including age, sex, and lesion location, to boost classification accuracy.

Results show an overall accuracy of 86% for MEL-IA, with 88% sensitivity for melanoma detection and 92% for basal cell carcinoma detection. The highest scores corresponded to nevi and basal cell carcinomas. The system transitioned from an experimental phase to live deployment at Sant Joan d'Alacant University Hospital, where it processed 980 dermatological studies with response times under one second. MEL-IA also maintains a longitudinal record of lesions, including images, diagnoses, and medical observations.

The researchers emphasize that MEL-IA is a clinical decision-support tool and not an autonomous diagnostic system. Future steps involve conducting prospective studies with healthcare professionals, testing direct image acquisition via smartphones, and expanding the range of lesion types the system can classify.