PoxDetector: A Deep Convolutional Neural Network for Skin Lesion Classification using Android Application

Human monkeypox outbreaks have recently been recorded in several nations throughout the world, with rapidly rising number of cases. Monkeypox may be difficult to diagnose clinically in its early stages because of its similarities to both chickenpox and measles. Since confirmatory Polymerase Chain Reaction (PCR) tests are not readily available and various deep-learning techniques have shown promising results in medical diagnosis, computer-assisted monkeypox lesion detection may be beneficial for monitoring and early identification of suspected cases. This research work presents a precise, computationally fast and reliable alternative for monkeypox diagnosis which facilitates quick processing by integrating deep transfer-learning based methods with deployment in android platform that helps in assisting the situation. Images captured by the camera with live feed or user selected images can be analysed using a deep convolutional neural network running on the same device. Following that, the network categorises images for the identification of either chickenpox, measles, monkeypox or normal skin type. An openly accessible dataset has been utilised for this purpose which results in an accuracy of 88.54 (±2.1%) which outperforms all the other existing models for this task. These positive findings, which exceed the most advanced techniques, imply that the suggested method may be used by the general public for mass screening as well as by the health practitioners to rank the seriousness of a case based on the results provided by this model to provide better attention to them accordingly.