GAN-Based Image Augmentation and Comparative Analysis of Various CNN Models for Monkeypox Detection

The outbreak of monkeypox across 75 nations has made this epidemic a major threat to the public health. Monkeypox similarity in clinical features to those of smallpox, chickenpox and measles makes it difficult to identify. Effective management and treatment of monkeypox relies on early diagnosis as well as identification. Previous researches have evaluated various deep learning approaches; however, they were conducted on a small dataset available in public domain, which impacts the learning of deep learning models and eventually impacts their precise accuracy and error rates. The goal of this study was to identify and verify the most effective methodology for monkeypox detection by comparing various deep learning techniques and classification models with significantly large dataset. For this reason, a Generative Adversarial Network (GAN) was used in this study to generate 4142 augmented images from 414 images present in original dataset available in public domain which increased the dataset size 10 times with respect to the original dataset. Six popular deep learning models (EfficientNetB3, VGG19, ResNet50, MobileNetV2, VGG16 and InceptionV3) were developed and trained using generated large dataset and evaluated on different parameters to find the best performing one. The results of our experiments show that the EfficientNetB3 model performed the best in terms of classification with an accuracy level of 98.46% whereas other models like VGG16, VGG19, InceptionV3, MobileNetV2 and ResNet50 respectively resulted in 96.76%, 93.78%, 85.68%, 75.25%, 69.01% accuracies. Our findings demonstrate that for classification of images for monkeypox, the EfficientNetB3 approach performs better than other models. In the terms of both test and training sets also, it demonstrated a great level of precision in categorizing monkeypox images, hence it can be a useful asset for prompt and precise diagnosis in clinical situations. These findings seem promising since they highlight how deep learning techniques can be employed for early detection of monkeypox.