Explainable artificial intelligence and feature based technique for the classification of kidney ultrasound images

Abstract

Millions of people worldwide are affected by chronic kidney disease (CKD), which is one of the main causes of death. Using machine learning (ML) models, this study attempts to create a computer-aided diagnostic (CAD) system that can autonomously detect chronic kidney disease (CKD) with improved interpretability. An online medical database provided 340 ultrasound images used in this study, which included both normal and abnormal instances. 94 texture and intensity attributes were obtained from these images using Pyrandiomics. Six machine learning methods were used for classification: According to the evaluation results, support vector machine (SVM), decision tree (DT), random forest (RF), k-nearest neighbors (k-NN), XG-Boost, and naïve Bayes (NB) models were considered. Among these models, the random forest model demonstrated the highest accuracy. Explainable artificial intelligence (XAI) methods, namely Shapley additive explanation (SHAP), were utilized to improve model transparency. Clinicians could be assisted in comprehending the reasoning behind the predictions using SHAP analysis, which identifies the most important features impacting the ML model and visualizes the ranking of each individual feature.