A multimodal deep learning model integrating patient data and dermoscopic images may improve early skin cancer detection, according to a new study in Information Fusion.
Using the large-scale Society for Imaging Informatics in Medicine (SIIM) and International Skin Imaging Collaboration (ISIC) melanoma dataset, which contains more than 33,000 dermoscopic images paired with clinical metadata, researchers trained an AI model to recognize subtle links between what appears on the skin and who the patient is.
The model achieved 94.5% accuracy and an F1-score of 0.94, outperforming popular image-only models such as ResNet-50 and EfficientNet. (An F1-score is a measure of predictive performance.)
The researchers also performed feature importance analysis to make the system more transparent and robust. Factors like lesion size, patient age, and anatomical site were found to contribute strongly for accurate detection.
“The model is not merely designed for academic purposes. It could be used as a practical tool that could transform real-world melanoma screening,” says Professor Gwangill Jeon from the Department of Embedded Systems Engineering, Incheon National University, South Korea, in a news release. “This research can be directly applied to developing an AI system that analyzes both skin lesion images and basic patient information to enable early detection of melanoma.”
In the future, the model could power smartphone-based skin diagnosis applications, telemedicine systems, or AI-assisted tools in dermatology clinics.
IMAGE CAPTION: A new deep learning system developed by an international research team detects melanoma with 94.5% accuracy by fusing dermoscopic images and patient metadata such as age, gender, and lesion location. The approach enhances diagnostic precision, transparency, and access to early skin cancer detection through smart healthcare technology.
IMAGE CREDIT: Professor Gwangill Jeon from Incheon National University, South Korea
