Hybrid Neural Network Framework for Multiclass Classification of Kidney Stones from CT scans

In order to diagnose and evaluate illnesses, 3D medical imaging may identify and extract a variety of anatomical traits, one of which is kidney stones. Segmenting renal stones from abdominal CT scans is challenging because of their size, shape, and location variability. Because they include so many vital organs, abdominal CT scans are notoriously difficult for 3D segmentation algorithms to handle. Both the diagnosis and the segmentation processes are enhanced by a pretreatment operation. Abdominal anatomy continues to be a challenge for CT-based kidney stone diagnosis, despite recent advances. The GLCM is employed to extract features from segmented kidney stones in order to classify them in this study. A hybrid deep learning approach called CNXGBoost is created by combining XGBoost with CNNs. XGBoost excels at processing tabular data, and CNN is great at extracting useful features from medical images. Incorporating both approaches improves the accuracy of kidney stone classification. The explainable CT system and hybrid CNXGBoost model were subjected to extensive testing and validation, and the results showed an impressive kidney stone classification accuracy of 98.71%. A reliable and accurate automated kidney stone identification method for difficult abdominal CT images has been developed using XGBoost and CNN.