Nail alterations have been observed in patients with chronic kidney disease (CKD) due to the buildup of uremic toxins in the blood. Many of these conditions present overlapping visual features, making them difficult to distinguish, especially for non-specialists. Although current research has explored differences between abnormal and healthy nails, classification within a single AI model remains limited. This study aims to develop an artificial intelligence (AI) system capable of classifying some nail conditions associated with CKD, including Lindsay’s nails, splinter hemorrhages, onycholysis, leukonychia, and normal nails, using photographic images. A total of 564 publicly available images were validated for labelling accuracy and divided into training, validation, and test sets, with all images categorised into five classes. Model development was conducted using Python on Google Colab with a batch size of 32. Data augmentation techniques, including rotation, zoom, shear, horizontal flipping, and rescaling, were applied to improve model generalisability. Four convolutional neural network (CNN) architectures (VGG16, Xception, ResNet50, and EfficientNet) were trained for 50 epochs. The highest performance was achieved by VGG16, with a test accuracy of 87%, followed by Xception (65%), and both EfficientNet ResNet50 (20%). Confusion matrix analysis further confirmed the strong performance of the VGG16 model. Overall, this study demonstrates a potentially efficient and low-cost solution for assisting early identification of CKD through nail assessment, representing a promising step toward real-world screening applications. Future development should focus on expanding the dataset and improving the model’s accuracy.

Keywords

Chronic Kidney Disease (CKD), Lindsay’s Nails, Artificial Intelligence (AI), Convolutional Neural Networks (CNN), and Image Classification