Leveraging Deep Pre-trained Networks for Advanced Skin Lesion Classification for Human Monkeypox Detection

In response to recent human monkeypox outbreaks, the imperative of swiftly identifying and isolating infected individuals to curb transmission underscores the significance of innovative solutions. This study introduces an Android mobile application harnessing deep learning capabilities to address this urgent need. Developed using Java within Android Studio and Android SDK 12, the application leverages the device's camera via the Camera 2 API for real-time image capture. The captured video images are processed by a deep convolutional neural network (CNN) embedded within the device. Training the CNN involved utilizing a dataset containing skin lesion images from monkeypox-infected individuals and other skin conditions, employing a deep transfer learning methodology. The training and testing phases were executed using Matlab, with the selected network further trained using TensorFlow and adapted into a TensorFlow Lite model for mobile deployment. Successful testing on various devices yielded average inference times of 197 ms, 91 ms and 138 ms, affirming the application's efficiency. Facilitating swift preliminary diagnosis, the application empowers individuals with skin lesions to seek prompt medical attention, potentially curtailing disease transmission. Notably, the system exhibits a commendable 91.11% accuracy in classifying images, indicating its reliability. Moreover, its adaptable architecture suggests broader utility for training in diagnosing diverse skin diseases, reflecting a promising avenue for future healthcare innovations.