Multimodal Machine Learning Approach for Diagnosing Atopic Dermatitis

This research is part of our multi phase research for doctoral study program. This study presents a novel multimodal deep learning model that combines clinical image analysis with structured anamnesis using a late-fusion architecture. ResNet50 was employed for image-based feature extraction, while MPNet was used to process textual clinical data. Trained and validated across multiple centers, the model achieved an internal validation accuracy of 98.28%, significantly outperforming single-modality models and standard diagnostic baselines. Our approach not only enhances diagnostic accuracy for atopic dermatitis but also highlights the translational potential of multimodal AI to support clinical decision-making in dermatology. It would be an honor for this research to be published in a Scopus-indexed journal, and we offer it as our contribution to promoting aviation safety.