Deep Learning and Explainable AI for Monkeypox Diagnosis

Monkeypox can be transmitted to people from animals by direct contact and the initial symptoms can be similar to those of chickenpox. The infection can cause life-threatening complications such as pneumonia and sepsis if it becomes severe. Vigilance and sensitization are essential to put an end to further transmission. For the control of this newly emerged viral disease, knowledge on the transmission mode and an early diagnosis is indispensable. To address these limitations, we apply CNN-based models to identify other types of skin diseases and to train a deep learning model for monkeypox classification. The dataset contains normal samples as well as multiple types of skin disorders including monkeypox. To get better classification accuracy, different structures are employed, such as the ResNet, VGG19, VGG16, MobileNet, and Xception. Also, we use Grad-CAM to generate the important regions that influence the model decisions, which is in favor of interpretability. Finally, standard classification performance metrics (accuracy, precision, recall, F1-score, and ROC-AUC) are applied to assess the model performance. The approach in this work could be useful for clinical decision-making to assist the clinician for correct diagnosis of skin diseases.