Categorization of human monkey-pox from skin lesion images based on transformer and ensemble learning using GRADCAM

Abstract

As the world keeps healing from the worldwide outbreak of COVID-19, the MPOX virus poses a new risk. The Mpox virus is not as deadly or contagious as COVID-19, but new patient cases are recorded every day from a wide variety of nations.Therefore, it will not come as a surprise if another global pandemic occurs due to a lack of pre- cautionary measures. Therefore, it is essential to detect them before they spread throughout the community. The World Health Organisation (WHO) has issued nu- merous precautionary warnings regarding MPOX. If MPOX spreads swiftly, it poses a significant threat to public health. The result is a significant increase in hospital wait times. This means that hospitals require additional supplementary or auxiliary systems.Recent advances in machine learning have demonstrated immense promise for diagnosis based on image data, including detection of cancer, identification of tu- mour cell, and identification of COVID-19 patients. Therefore, identical technology could possibly be used to detect the human skin infection known as MPOX. After acquiring an image it can be utilised for further diagnosis and early identification of MPOX.In this research, we leverage 13 recently developed deep learning (DL) mod- els to suggest a new strategy for enhancing the accuracy of MPOX image detection and classification. Eight of the suggested models are based on transformer, while the remaining five are CNN-based models that have been pre-trained.The publicly avail- able Monkeypox Skin Images Dataset (MSID) is utilized to evaluate the suggested method. Four standard metrics—precision, accuracy, F1-score and recall —were applied to the outcomes after the models were fine-tuned. We ultimately utilised an ensemble approach to enhance overall performance and obtain more precise results. We achieved the best results possible with our approach, which included a 99% F1 score, 99% accuracy, 100% precision, and 99% recall. Based on these promising out- comes, which surpass those of existing methods, we propose applying the suggested method for widespread testing by health practitioners. This model can be used as a supplementary diagnostic system for the early detection of MPOX skin lesions.