Analyzing kidney diseases using deep learning techniques

Artificial intelligence, particularly deep learning techniques—a subfield of AI—has found a wide range of applications in the field of healthcare. It plays a crucial role, especially in the fast and more reliable diagnosis of diseases. This study focuses on the artificial intelligence techniques used in the diagnosis of kidney diseases and reviews related research in this domain. In the application part, a dataset composed of kidney images collected from various hospitals in Dhaka, Bangladesh, was combined with a dataset specifically compiled for medullary sponge kidney disease by Adham Mohammed AlDakrany and Mohammed Ahmed Aref. The dataset primarily consists of computed tomography (CT) scans, including both contrast-enhanced and non-contrast studies, featuring coronal and axial slices that cover the entire abdomen and urogram. Each image has been carefully reviewed and verified by radiologists and urologists who contribute to Radiopaedia cases. A comparison was made between the VGG network, known in the literature for achieving the highest accuracy in kidney disease diagnosis, and a simpler 19-layer network. Optimization efforts were conducted on the constructed network to ensure the best possible performance. The dataset was used to train the developed convolutional neural network (CNN), and the performance of this architecture in diagnosing kidney disease was evaluated. Furthermore, fuzzy pooling was integrated into the CNN architecture to achieve improved results.