Monkeypox Classification Using Convolutional Neural Networks (CNN) Pruned Residual Network-50 (ResNet-50) Architecture on Flutter Framework

The monkeypox outbreak, which was previously only found in Africa, has now spread to other continents, including Asia, causing public concern as it occurred shortly after the COVID-19 pandemic was declared over. This disease has symptoms similar to cowpox, chickenpox, and measles, making early detection based on visual observation difficult. To address this issue, various studies have developed Deep Learning (DL)-based classification models using datasets such as WSI, MSID, MCSI, and MSLD v2, which are also utilized in this research. This study proposes a pruned ResNet-50 model using the Global MP method for pruning and QAT for quantization. These modifications not only maintain the model's performance with an accuracy of 94.44%, precision of 94.12%, recall of 94.71%, and F1-score of 94.16%, but also significantly reduce the model size to just 20.993 MB. As a result, the model can be implemented on Android devices with limited resources, enabling rapid and practical early detection of monkeypox in the field without requiring large-scale servers. Blackbox testing results show that the Flutter-based application utilizing this model performs well, potentially providing tangible support for medical personnel and the public in monitoring the spread of monkeypox in a more efficient and accessible manner.