Multi-head Attention Fusion Network for Mpox Screening Using Skin Lesion Image Analysis

Monkeypox also known as Mpox is an infectious disease where screening is a vital task for early detection and effective management, particularly during recent outbreaks. Screening using skin lesions is a quick and convenient method. However, it can be challenging to differentiate Mpox from other diseases like chickenpox and measles in the early stages because their skin lesions can appear similar. This study introduces a multi-head Attention Based Fusion Network (MAFNet) to analyze skin lesion images for Mpox detection. The model achieves an accuracy of 89.81%, surpassing traditional architectures like DenseNet, VGG16, and ResNet. Using the multiple head attention mechanisms the proposed model enhances its focus on critical features within skin lesions and more accurately differentiates among various dermatological conditions. Besides that the model is trained not only on Mpox and healthy skin but also along with four other classes of similar diseases, enhancing its ability to generalize effectively. By the identification of Mpox without explicit clinical diagnosis and easy to use web app, this model helps early detection of Mpox patients. Such early identification is particularly valuable in high-traffic areas such as airports, where rapid decision-making is crucial. This approach aims to help prevent future pandemics similar to Covid-19, by promoting timely public health interventions.