Benchmarking Pre-trained CNNs and Vision Transformers for Mpox-related Dermatological Image Classification on MSLD v2.0

The early detection and accurate classification of monkeypox are critical for controlling outbreaks and ensuring timely treatment, especially due to its clinical similarities with other dermatological diseases such as chickenpox, measles, and smallpox. Traditional diagnostic methods can be time-consuming and prone to human error, thus highlighting the need for efficient and reliable automated systems. This study evaluates the performance of five state-of-the-art deep learning models, Xception, ConvNeXt, DenseNet121, MobileNetV2, and Swin Transformer, in the detection and classification of monkeypox and other skin conditions. The primary objective of this research was to assess and compare the models' abilities to classify a diverse dataset of dermatological images based on accuracy, precision, recall, F1-score, and AUC. The models were trained using transfer learning on a publicly available dataset of skin lesions, augmented to increase model robustness. The results show that the Xception model outperformed all others, achieving an exceptional 99.92 % validation accuracy, alongside high precision and recall values. In comparison, other models like Swin Transformer and MobileNetV2 demonstrated lower performance, particularly in generalization to unseen data. The findings indicate that Xception provides the most reliable and efficient solution for real-time detection of dermatological diseases, making it an ideal candidate for clinical use, telemedicine applications, and automated diagnostic systems. This study establishes a strong, standardized baseline for mpox-related dermatological image classification using CNN and transformer architectures, following good-practice principles inspired by ISO 14,971 and IEC 62,304, without claiming formal regulatory compliance.