A comparative study of deep learning models for early diagnosis of monkeypox and chickenpox

Image processing enhances digital images by extracting valuable information and improving clarity through edge recognition, color modification, and filtering techniques. These methods facilitate the creation of training images that enhance the robustness and generalization of neural networks, improving diagnostic efficiency and accuracy. The similar symptoms of chickenpox and monkeypox make early clinical diagnosis difficult, necessitating rapid and precise diagnostic techniques. This study uses deep learning models to tackle the problem of correctly diagnosing chickenpox and monkeypox, facilitating early diagnosis and quick medical intervention for improving public health and epidemic management. Models such as DenseNet121, DenseNet169, DenseNet201, InceptionResNetV2, InceptionV3, MobileNet, MobileNetV2, ResNet50, ResNet101, VGG16, VGG19, and Xception have been used to classify and distinguish conditions. Advanced image processing techniques, including object detection and data normalization, improve model generalization and resilience. Among the others, MobileNet and DenseNet121 achieved highest accuracy rates which is 0.9943 and 0.9914 respectively, while MobileNet offered superior speed. The DenseNet169 model has highest precision and recall which is 0.9958 and f1-score is 0.9956 with an accuracy of 0.9857. The study also demonstrates that DL models are critical in developing robust and effective systems for detecting infectious diseases, aiding in precise and efficient healthcare solutions.