Deep Learning Explainability with Local Interpretable Model-Agnostic Explanations for Monkeypox Prediction
Abstract
The accurate categorization of medical images is critical in facilitating clinical treatment and care. In recent studies, deep learning (DL) models have demonstrated tremendous possibility in image-based diagnosis, specifically in the areas of pneumonia detection, tumor cell identification, and COVID-19 diagnosis. It is challenging to identify Mpox (formerly known as monkeypox) on human skin due to the unavailability of a publicly available dataset of Mpox. In order to address this issue, authors have collected a dataset called “Monkeypox2022” and is made accessible to the public. The research community can find it on GitHub at this link: https://github.com/Angmo21720/data-set . The dataset was compiled using photographs sourced from various open-source and internet platforms, without any restrictions on their usage, including commercial purposes. In addition, the authors have made enhancements to the VGG16 model and conducted two experiments, Experiment 1 and Experiment 2, to evaluate its performance. Based on our research, it was discovered that the AUC score achieved by the model in Experiments 1 and 2 were 0.972 and 0.748, respectively, in identifying monkeypox patients. Local interpretable model-agnostic explanations (LIMEs) are utilized for analyzing the results of the model employed and uncovering the specific features on which various predictions were made by the model. This allows for a more comprehensive understanding of the factors that contribute to identifying a monkeypox infection.