Deep Learning-Based Multi-Pox Disease Identification Using a Hybrid Model

An early detection allows an accurate diagnosis of skin diseases which is crucial for treatment and management. Traditional approaches for diagnosis rely on dermatological expertise that is potentially time-consuming and subjected to human error. The proposed study takes an approach of designing an automated skin disease detection system that utilizes deep learning models for enhancing the accuracy and availability of the diagnostic process. A diverse dataset was used that featured images of skin conditions like chickenpox, measles, monkeypox, and healthy skin. Meanwhile, preprocessing methods including resizing, stretch contrast, gamma correction, and augmentation, have improved the quality of images for the dataset ensuring uniformity. Several different deep learning architectures were developed and tested, namely DenseNet-121, ResNet-50, EfficientNet-B0, and a novel hybrid model that merges ResNet-50 and EfficientNet-B0. The models were trained and validated based on performance metrics including accuracy, precision, recall, and the F1 score for each of the classes. Accuracies of 98% along with precision, recall, and f1 scores of 98.66% were obtained with the hybrid model which performed best among the candidate models. Our results emphasize the promising nature of our approach as it has the potential to impact dermatological diagnostics significantly, and, by extension, the medical industry as a whole.