DiffusionDxNet: Improving Monkeypox Classification Using Diffusion-Based Data Augmentation and DenseNet121

Monkeypox is an emerging infectious disease whose skin manifestations closely resemble other dermatological conditions such as chickenpox and measles, making early visual diagnosis challenging. This diagnostic similarity may delay early detection, highlighting the need for automated and reliable computer-aided diagnostic systems. In our research, we develop a deep learning model using diffusion-based data augmentation and the DenseNet121 model to classify monkeypox into various categories. Because monkeypox image datasets are most often unbalanced, we create synthetic images of minor categories to develop a balanced dataset. This is followed by fine-tuned four-class classification tasks using a pretrained DenseNet121 model. To make our model more robust for better predictive performance, we investigate the combination of multiple learned models through both the uniform and greedy model soup methods. The experiments were conducted using the Monkeypox Skin Images Dataset (MSID), a publicly available four-class dermatological dataset containing 770 images collected from internet-based sources and curated by the Department of Computer Science and Engineering, Islamic University, Kushtia, Bangladesh. The proposed approach was evaluated on an independent test set using standard metrics including accuracy, precision, recall, F1-score, Matthews correlation coefficient, and ROC–AUC. The model achieved an accuracy of 97.06%, macro precision of 96.05%, macro recall of 95.35%, macro F1-score of 95.65%, and ROC–AUC of 0.9969. These results indicate consistent performance across all classes and demonstrate the effectiveness of diffusion-based augmentation combined with robust feature learning. In conclusion, the proposed framework demonstrates strong potential as an efficient and scalable solution for automated monkeypox diagnosis and highlights the usefulness of diffusion based augmentation in medical image analysis.