A Comparative Study of Feature Selection Methods for ML-Based Kidney Stone Prediction

In the field of medical diagnosis, the accurate prediction of kidney stone presence plays a major role. Kidney stones are the solid particles formed due to mineral deposits in the urine, it comes in various shapes and sizes, posing a unique challenge for detection. Various research and imaging techniques have enabled the identification of these stones but their interpretation often requires the expertise of medical professionals. Integrating Computer-Aided Diagnosis (CAD) systems into clinical practice has proven invaluable, offering practical tools to aid clinicians in their assessments. However, the presence of noise in medical images has introduced errors in kidney stone classification. To enhance the precision of kidney stone prediction, this study proposes a comprehensive investigation into optimized feature selection methods and their synergy with Machine Learning (ML) techniques. The dataset, comprising 1200 images of healthy CT scans and 1200 images of kidneys affected by stones were carefully prepared through image resizing and rescaling. Feature extraction was accomplished using the Gabor filter, followed by the application of three distinct optimization algorithms: Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), and Bee Colony Optimization (BCO) to identify the most relevant features. The outcomes derived from these optimized feature sets were subsequently evaluated using three different ML models: Support Vector Machines (SVM), Decision Trees (DT), and Artificial Neural Networks (ANN). The validation process encompassed nine unique combinations of optimization techniques and Machine Learning (ML) models. The results demonstrate that the combination of ACO with ANN has achieved the highest accuracy of 98% in kidney stone prediction, signifying its exceptional performance in this critical diagnostic task.