Abstract
This study aimed to classify pediatric ultrasound images as normal or abnormal by identifying the optimal number of image texture features for analysis and developing an effective classification system using selected features. The experiment identified a successful feature selection and classification algorithm with a good performance. This study introduced a new approach for computer-assisted ultrasound image classification. Initially, a Gaussian median filter enhances the image quality and removes noise. For feature extraction, various features, including first-order derivatives, Gray Level Co-Occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM), Gray Level Dependence Matrix (GLDM), Gray Level Size Matrix (GLSZM), and Neighbouring gray tone difference matrix (NGTDM), were extracted using the Pyrandiomics Python package. The Coati optimization algorithm (COA) was employed as a feature selection technique. The Classification was performed using Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), K-nearest Neighbor (KNN), Naï ve Bayes (NB), and Extreme Gradient Boosting (XG-Boost) algorithms. Therefore, this study proposed a new machine learning classifier, the Extreme Gradient Neighborhood classifier (XGNC), using NB, KNN, and XG-Boost, with a classification accuracy of 97.91%, which outperformed the other classifiers mentioned in the study. The results indicated that the optimal feature selection and classifier choice yielded the most accurate computer-aided diagnosis of kidney abnormalities.
Keywords: Coati Optimization Algorithm (Coa), Extreme Gradient Neighborhood (Xgnc), Feature Selection, Kidney, Machine Learning (Ml), Ultrasound (Us).