Performance Evaluation of CNN Models for Kidney Abnormality Detection: A Comparative Study

Kidney abnormalities pose serious health risks if not detected early. This study evaluates five Convolutional Neural Network (CNN) models which includes EfficientNet, DenseNet, ResNet, InceptionNet, and VGGNet for kidney abnormality detection using medical imaging. A curated and augmented dataset of annotated kidney images was used to train each model under consistent settings. Performance was measured using accuracy, F1 score, and computational efficiency. EfficientNet outperformed all others, achieving 96.4% accuracy and 95.7% F1 score, due to its compound scaling technique. DenseNet followed closely with high feature reuse and training stability. ResNet performed well but lagged in fine-grained detection. InceptionNet and VGGNet, though effective, were less efficient and accurate. EfficientNet and DenseNet showed strong potential for clinical deployment due to their precision and speed. The findings highlight the importance of choosing advanced, scalable models for automated kidney disease diagnosis, with future work aimed at hybrid models and explainable AI for clinical interpretability.