Multi-Task and Multi-Dimensional Deep Learning for Alopecia Areata Detection and Diagnosis

Abstract

Unhealthy lifestyles and vitamin deficiencies contribute to various scalp-related issues, such as dermatitis and baldness. Alopecia Areata (AA) is a prevalent form of hair loss, commonly diagnosed using medical image processing-based models. However, these models often struggle with overlapping hairs and are highly sensitive to configuration variables, making them less reliable. To overcome these limitations, deep learning has been increasingly applied in medical image analysis for the detection and diagnosis of AA. While several deep learning models have been developed to recognize different scalp conditions, there remains a need for simultaneous identification of AA and other scalp conditions to improve diagnostic accuracy.This study introduces a Multi-Task Deep (MTDeep) learning system, incorporating the MT Faster Residual Convolutional Neural Network with Long Short-Term Memory (MT-FRCNN-LSTM) model. This approach aims to recognize both AA and various scalp conditions in individuals with different baldness patterns. The primary objective of multi-task learning (MTL) is to improve recognition accuracy through the use of a shared encoder. In this model, scalp and AA images are initially processed through an LSTM encoder and an FRCNN encoder, extracting global and local features at different scales. These extracted features are then fused to generate a comprehensive feature representation. Finally, a fully connected layer, followed by a softmax classifier, is applied to categorize different scalp conditions.