Skin lesion classification from smartphone clinical images is attractive for low-resource settings, but images alone miss important context such as a patient's medical history and lesion symptoms. The new arXiv paper introduces GAMF-Net, a multimodal fusion framework that combines these complementary inputs for smartphone-based classification.
The model's name points to two design principles: grouped fusion and availability awareness. While the abstract does not give full architectural details, the emphasis on availability suggests the method is intended to handle situations where not all data modalities are present for every patient.
The work is positioned as a step toward more practical computer-aided diagnosis in resource-constrained environments, where low-cost imaging and clinical data can be used together rather than relying on images alone.