Abstract—The ongoing monkeypox outbreak poses a significant global health challenge, underscoring the need for rapid and accurate identification of symptoms and rash stages. Deep learning algorithms have become increasingly valuable for disease diagnosis using medical imaging. This study addresses the classification of monkeypox skin lesions and rash stages using deep learning techniques, employing a combined and augmented dataset from “Monkeypox2022” and “Mpox Skin Lesion Dataset Version 2.0 (MSLD v2.0)”, and a dataset of “RashStageMpox”, which covers various rash stages. Three convolutional neural network architectures were evaluated—MobileNetV2, EfficientNetB0, and EfficientNetB2. EfficientNetB2 demonstrated strong performance, achieving 94.61% accuracy for skin lesion classification (Adam optimizer, batch size 32, learning rate 0.0001) and 90.90% accuracy for rash stage classification (Adam optimizer, batch size 32, learning rate 0.001). Building on this, the study further enhanced EfficientNetB2 by integrating the Squeeze-and-Excitation (SE) module. The SE-augmented EfficientNetB2 achieved even higher accuracy: 96.15% for skin lesions and 96.96% for rash stage classification. The SE module’s dynamic channel-wise feature recalibration improved the model’s focus on critical diagnostic features, resulting in more accurate classification, particularly in correctly identifying scabs and reducing misclassification of vesicles as pustules, as reflected in the confusion matrices. Despite these promising results, limitations remain regarding dataset diversity, image quality, and generalizability to broader patient populations. Future work should include validation with multi-institutional datasets, optimization for mobile deployment, and real-world clinical evaluation, focusing on Explainable Artificial Intelligence (XAI), robust data security, and user-friendly interfaces to ensure ethical and practical implementation.
 
Keywords—monkeypox, deep learning, skin lesion, rash stages, EfficientNetB2, Squeeze-and-Excitation (SE) module


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).