Ensemble of Pre-Learned Deep Learning Model and an Optimized LSTM for Alopecia Areata Classification

The study proposes an Ensemble Pre-Learned Deep Learning and Optimized Long Short-Term Memory (EPL-OLSTM) model to improve the classification of Alopecia Areata (AA) by addressing the limitations of existing Computer Aided Diagnosis (CAD) models. The model utilizes pre-learned CNN structures like AlexNet, ResNet, and InceptionNet to extract features from scalp hair images, which are then optimized using the Battle Royale Optimization algorithm in the LSTM's hyperparameters. The final classification into AA classes is done using fuzzy-softmax. The model was tested on the Figaro1k and DermNet datasets, achieving a 93.1% accuracy, outperforming current state-of-the-art deep learning models.