Triple Transformer Ensemble Fusion Method for Pox Virus Classification

Background and objectives: pox viruses are infectious agents that affect both humans and animals, often presenting similar skin lesions, making accurate diagnosis a medical challenge. Early detection and classification are crucial for outbreak control and timely clinical intervention. Automated diagnosis is essential, particularly for accurate multi-class classification. Methods: the novel ensemble method was developed to address the multi-class-wise prediction by using the Triple Transformer Ensemble Fusion Method (TTEFM). The TTEFM method was compared with existing pre-trained transformer methods, including the Vision Transformer (ViT), Mobile_ViT, and Data-Efficient Image Transformer (DEiT). The model was trained and tested using Monkeypox Skin Lesion Dataset (MSLD), which includes four classes: chickenpox, measles, monkeypox and normal. Results: the TTEFM methods outperform other state-of-the-art works. Based on the evaluation metrics, the methods are compared with other pre-trained transformers. The TTEFM method attains 99% accuracy for all the classes. The ensemble techniques were proven using the one-way Analysis of Variance (ANOVA) technique. Conclusion: the automated identification of skin lesions is crucial for clinical diagnosis, enabling dermatologists to identify and treat pox virus infections effectively. The presented TTEFM model provides a highly accurate and reliable solution for medical image classification