Abstract
Monkeypox is a rapidly spreading virus which poses significant diagnostic challenges, due to its overlap with other viral illnesses. The availability of polymerase chain reaction (PCR) assays in resource-constrained environments is often hindered. In this work, deep learning is used to automate monkeypox detection by using skin lesion images. To enhance the quality and volume of the available dataset, advanced deep learning algorithms are combined with image augmentation techniques. By combining Flip, Mirror, rotate (FMR) random image augmentation with Automated White Balance Correction (AWBC), the detection model becomes more robust. With CoAtNet, a synthesis of convolutional neural networks and transformers, lesion images are captured both locally and globally. By using a hybrid architecture, the visual data can be analyzed more comprehensively and diagnostic errors may be reduced. This model is the most accurate, precise, recall, and F1-score with other existing models. An automated monkeypox detection system can be improved significantly by incorporating CNNs and transformers. Data augmentation strategies are also recommended as a way to enhance these models. The training dataset enhancing the model’s ability to generalize to new cases. There is substantial promise in deep learning-based diagnostic tools for monkeypox, especially in areas with limited access to traditional laboratory testing. This work can support healthcare systems in combating the spread of this virus. Diagnostic gaps in such regions can be bridged with such systems, thereby contributing to the global health community’s collective response.
Keywords: CoAtNet, Deep Learning Pipeline, FMR Random Image Augmentation, Monkey Pox Diagnosis, MSLD, Multi-scale Attention-guided Lesion Segmentation, White Balance Correction.