CKD Detection Using CNN on Ultrasound Images Based on Estimated Glomerular Filtration Rate (EGFR) Values

Chronic kidney disease (CKD) is a progressive and irreversible decline in kidney function that, if left untreated, can lead to serious complications. Ultrasonography (USG) is a widely used imaging modality for detecting CKD, yet its interpretation remains highly dependent on the radiologist’s expertise. This study aims to develop a CKD detection system using a convolutional neural network (CNN) on kidney ultrasound images based on estimated glomerular filtration rate (eGFR), and to evaluate the system’s performance.
This research employed a research and development (R&D) approach with an experimental design. The dataset consisted of kidney ultrasound images from CKD and non-CKD patients with corresponding eGFR values. The methodology included image preprocessing, CNN model training, and accuracy evaluation using classification metrics. The results demonstrated that the developed CNN model achieved a total accuracy of 97% on internal test data and 95.8% on external validation. The model’s sensitivity reached 100% for the normal category, 91.67% for CKD stage 4, and 90% for CKD stage 5. Specificity exceeded 96% across all categories, with high precision and F1-scores above 94% for all classes.
This system has proven to be effective as a diagnostic support tool for automatically detecting CKD through kidney ultrasound imaging. Its advantages lie not only in accurately classifying CKD from USG images but also in correlating the classification results with patients' eGFR values. This provides more precise clinical information and supports appropriate CKD staging and management planning.