Monkeypox Classification Using Fractal-CNN

The past resurgence of Monkeypox draws attention to the need for urgent fast and sensitive diagnostic methods, especially in low resource settings. This work introduces a new approach on classifying Monkeypox lesions using a Fractal Convolutional Neural Network. The recursive multi-path fractal blocks are incorporated into the proposed fractal-CNN architecture to capture complex, multi-scale features that are essential to distinguish such lesions from similar skin conditions like chickenpox and measles. It developed a dataset of 2,056 skin lesion images, both positive and negative for Monkeypox, and preprocessed those using data augmentation techniques for increasing variability. Transfer learning and fine-tuning of hyperparameters for medical images were applied. Experimental results indicate that Fractal-CNN obtains high classification performance with an accuracy of 95.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, an F1-score of 0.978, and an AUC-ROC of 0.993 better than the baseline models: ResNet50 and MobileNetV2. The model was further optimized to be deployed on a mobile device, achieving an average inference time of 45 ms. This means real-time diagnosis can be made possible in mobile health applications. This project holds promise for the application of Fractal-CNN in public health; scale up, accurate, and accessible screening for Monkeypox could be made available to remote or underserved areas. The future work would include extension of the dataset and addition of techniques for explain ability to enhance interpretability for clinical applications.