VirLesDetNet: Pre-Trained Hybrid Deep Learning Approaches for Virus-Based Skin Lesion Detection

Abstract

The monkeypox virus is a DNA virus with a double-stranded structure and belongs to the Orthopoxvirus family. While skin lesions areAa major indicator of monkeypox, they are often indistinguishable from early-stage chickenpox and measles lesions, leading to potential misdiagnoses. To address this issue, aAnew hybrid deep learning model has been developed to classify skin lesions into four categories: normal, monkeypox, chickenpox, and measles, using the publicly available Monkeypox Skin Images Dataset (MSID). AThe dataset was initially expanded through image preprocessing and data augmentation techniques. Seven pre-trained deep learning models were then trained individually. After evaluating their performance, the top three models were selected, and an ensemble model was created to improve overall accuracy through majority voting based on the probabilistic outputs from these models. The model's effectiveness is validated by accuracy, recall, precision, F1 score, and a confusion matrix. The proposed ensemble model, which combines EfficientB3, ResNet152, and MobileNetV3, achieved a detection accuracy rate of 94.82%.