Chronic Kidney Disease Prediction using DL: A Comparative Study of CNN VGG16 and MOBILENET

This study presents an automated Deep Learning (DL) framework for the identification and diagnosis of chronic kidney disease (CKD) using Computed Tomography (CT) scans, aiming to overcome the challenges associated with traditional diagnostic methods. The research utilizes CNNs (CNNs), Visual Geometry Group (VGG16), and MobileNet to improve accuracy, precision, and recall in CKD classification. Three 2D-CNN models—a six-layer CNN, MobileNet, and VGG16 with 16 layers for detection, along with a four-layer CNN for classification—were trained and evaluated on a Kaggle dataset consisting of 12,445 CT images, with an 80:20 train-test split. The study emphasizes on the comparative assessment of DL models (CNN, VGG16, and MobileNet) for CKD detection, the implementation of an automated classification system to enhance diagnostic accuracy, and the validation of model performance using key evaluation metrics, including accuracy, precision, recall, F1-score, and loss values. Among the models, CNN exhibited better performance, obtaining an accuracy of 82%, outperforming MobileNet (70%) and VGG16 (72%), showcasing its advanced feature extraction capabilities. Performance evaluations further confirmed CNN’s effectiveness in CKD classification, exhibiting lower validation loss and greater generalization ability compared to other models. The findings highlight the significant potential of DL-based Computer-Aided Diagnosis (CAD) systems in enhancing early CKD detection, reducing radiologists’ workload, and improving decision-making, thereby leading to better patient care.