A Comprehensive Study of Deep Learning Methods for Kidney Tumor, Cyst, and Stone Diagnostics and Detection Using CT Images

Abstract

Kidney disease affects millions worldwide which emphasizes the need for early detection. Recent advancements in deep learning have transformed medical diagnostics and provide promising solutions to detect various kidney diseases. This paper aims to develop a reliable AI based learning system for effective prediction and classification of kidney diseases. The research involves a dataset of 12,446 kidney images which include cysts, tumor, stones, and healthy samples. The data undergoes thorough preprocessing to eliminate noise and enhance the quality of image. Segmentation techniques like Otsu’s binarization, Distance transform, and watershed transformation are applied to accurately delineate and identify distinct regions of interest followed by contour feature extraction which includes parameters like area, intensity, width, height, etc. Subsequently, different deep learning models such as DenseNet201, EfficientNetB0, InceptionResNetV2, MobileNetv2, ResNet50V2, and Xception are trained on incorporating with three optimizers—RMSprop, SGD, as well as Adam and are examined for the metrics such as accuracy, loss, precision, recall, RMSE, and F1 score. Notably, the Xception model outperformed others by achieving an accuracy of 99.89% with RMSprop. Similarly, ResNet50V2 and DenseNet201 demonstrated impressive accuracy of 99.68% with SGD and Adam optimizers respectively. These findings highlight the effectiveness of AI and deep transfer learning in accurate and effective kidney disease detection as well as classification.