Computer-Based Characterization of Specific Cystic Renal Masses of Bosniac Classification with Traditional Machine Learning and Modified Deep Learning Methods

Authors : Mehmet Ezer, Ali Berkan Ural, Canver Önal

Pages : 311-320

View : 37 | Download : 56

Publication Date : 2024-12-31

Article Type : Research Paper

Abstract :Aim: This study aims to develop a novel artificial intelligencebased pre-diagnosis system for classifying cystic renal masses (CRM) according to the Bosniak classification. The objective is to distinguish between the five diagnostic stages of the Bosniak classification using traditional machine learning (ML) and deep learning (DL) techniques. Material and Method: A total of 20 contrast-enhanced CT images were collected for each of the five Bosniak stages (I, II, IIF, III, IV), verified by a uro-oncologist and radiologist. Additional image variations were generated using the Keras image processing library during the data augmentation phase, resulting in 600 images per stage. This process included operations such as brightness and contrast modification, image rotation, noise addition, and flipping. These augmented images were then used to train both ML and DL models. The k-Nearest Neighbors (kNN) algorithm was applied for the ML approach, while modified Convolutional Neural Networks (CNN) and VGG-16 models were used for the DL approach. Model performances were evaluated using receiver operating characteristic (ROC) analysis and area under the curve (AUC) metrics. Results: The kNN algorithm accurately classified the Bosniak stages with an AUC of 0.854. The VGG-16 model demonstrated superior performance, with an AUC of 0.978, achieving higher classification accuracy than the kNN model. Conclusion: The computerized Bosniak classification system based on CT images effectively differentiates between the five Bosniak stages. This system, utilizing both ML and DL models, has the potential to enhance the pre-diagnosis of CRM in clinical settings and can also effectively exclude other types of renal masses.
Keywords : Renal Mass, Bosniak Classification, Machine Learning, Feature Extraction, Deep Learning