Abstract
On July 23, 2022, the World Health Organization (WHO) officially declared the Mpox outbreak a "Public Health Emergency of International Concern" (PHEIC), highlighting the urgent need for effective prevention and control measures worldwide. To assist healthcare managers and medical professionals in efficiently and accurately identifying Mpox cases from similar conditions, this study proposes a lightweight deep learning model. The model uses EfficientNet as the backbone network and employs transfer learning techniques to transfer the pre-trained EfficientNet parameters, originally trained on the ImageNet dataset, into this model. This approach allows the model to have strong generalization capabilities while controlling the number of parameters and computational complexity. Experimental results show that, compared to existing advanced methods, the proposed method not only has a lower number of parameters (only 4.14 M), but also achieves optimal values in most performance metrics, including precision (95.92%), recall (95.69%), F1 score (95.80%), ROC AUC (0.998), and PR AUC (0.999). Furthermore, statistical analysis shows that the cross-validation results of this model have no significant differences (p > 0.05), which verifies the robustness of the method in Mpox identification task. Additionally, ablation experiments demonstrate that as the version of EfficientNet's expanded network increases, the model complexity rises, with performance showing a trend of initially increasing before decreasing. In conclusion, the model proposed in this study effectively balances model's complexity and inference accuracy. In practical applications, model selection should be based on the specific needs of decision-makers.