Kidney Stone Classification using Machine Learning Models on CT Images

Crystalline masses, better known as kidney stones, develop inside the kidneys and lead to severe pains and further complications if not detected timely. Robust and precise kidney stones classification from CT images is critical for the diagnosis and treatment process. This work explores the usage of diverse machine learning models for kidney stone type detection (KNN, Naive Bayes and Neural Network). The dataset employed consists of 4 categories CT images including Cyst, Normal, Stone and Tumor. Preprocessing steps also included resizing images to $128 \times 128$ pixels and converting them into black and white. Histogram of Oriented Gradients (HOG) features were extracted to obtain structural information from the images. The models were measured on 80-20% of training and test sets, with the performance metrics considering accuracy precision recall F1-score. The Neural Network model has obtained the best accuracy of 0.99 and also outperformed than KNN as well Naive Bayes. The KNN model also had significant results and an accuracy of 0.99 whereas the Naive Bayes classifier struggled with lower accuracy, performance metrics etc.