Classification of Visually Similar Scalp Diseases using Deep Learning: A Hybrid CNN-VIT Approach with Cross-Attention Fusion

The study introduces ScalpViT, a novel hybrid deep learning model designed to accurately diagnose visually similar scalp diseases, such as Psoriasis, Seborrheic Dermatitis, Tinea Capitis, Alopecia Areata, Folliculitis, and Eczema. These conditions often share overlapping visual features, making diagnosis challenging. ScalpViT combines a Vision Transformer (ViT) and a Convolutional Neural Network (CNN) with a cross-attention fusion module to capture both global spatial patterns and local texture details. Trained on a dataset of approximately 7,000 images, ScalpViT achieved a 94.3% accuracy, outperforming other models like ResNet-50 and EfficientNet-B3. The model also provides visual explainability through GradCAM and Attention Rollout, enhancing its clinical applicability, especially in resource-limited settings.