Prediction Methodologies to Detect Kidney Stones using Deep Learning

Abstract

Under the condition known as nephrolithiasis, unwanted sediments accumulate in the kidneys, interfering with normal urinary system function and, in certain circumstances, obstructing urine flow, causing excruciating pain. Thus, the capacity to identify kidney stones via medical imaging is essential for administering treatment in a timely and efficient manner. Detecting objects in photos may be done with accuracy using Deep Learning techniques. The percentage of errors that are currently caused by human error will be reduced with the use of deep learning techniques in kidney stone identification. In this work, the presence or absence of kidney stones in CT scan pictures has been determined using four distinct deep learning techniques. The four algorithms that are employed are InceptionNetV3, ResNet50V2, MobileNetV2, and VGG16. These kidney stone detection algorithms were constructed using a dataset including 1799 kidney CT scan pictures. A combination of accuracy, precision, and recall metrics were used to evaluate each of the four models' categorization performance. For classification, the InceptionNet neural network yielded the highest results in terms of recall, accuracy, and precision. It yielded results of 0.8331 for recall, 0.866 for precision, and 0.862 for accuracy. These measurements are higher by 11%, 10.5%, and 2.9%, respectively, than the comparable values for the other three models. Therefore, this study validates that, among the four methods under evaluation, InceptionNet should be used for kidney stone diagnosis that occurs automatically.