A Transfer Learning and Explainable Solution to Detect mpox from Smartphones images

Monkeypox (mpox) virus has become a "public health emergency of international concern"in the last few months, as declared by the World Health Organization, especially for low-income countries. A symptom of mpox infection is the appearance of rashes and skin eruptions, which can lead people to seek medical advice. A technology that might help perform a preliminary screening based on the aspect of skin lesions is the use of Machine Learning for image classification. However, to make this technology suitable on a large scale, it should be usable directly on people mobile devices, with a possible notification to a remote medical expert. In this work, we investigate the adoption of Deep Learning to detect mpox from skin lesion images derived from smartphone cameras. The proposal leverages Transfer Learning to cope with the scarce availability of mpox image datasets. As a first step, a homogeneous, unpolluted, dataset was produced by manual selection and preprocessing of available image data, publicly released for research purposes. Subsequently, we compared multiple Convolutional Neural Networks (CNNs) using a rigorous 10 -fold stratified cross -validation approach and we conducted an analysis to evaluate the models' fairness towards different skin tones. The best models have been then optimized through quantization for use on mobile devices; measures of classification quality, memory footprint, and processing times validated the feasibility of our proposal. The most favorable outcomes have been achieved by MobileNetV3Large, attaining an F-1 score of 0.928 in the binary task and 0.879 in the multi -class task. Furthermore, the application of quantization led to a reduction in the model size to less than one-third, while simultaneously decreasing the inference time from 0.016 to 0.014 s, with only a marginal loss of 0.004 in F-1 score. Additionally, the use of eXplainable AI has been investigated as a suitable instrument to both technically and clinically validate classification outcomes.