Chronic renal disease classification using the AL-5-ENeT-B4 model from CT images.

Abstract

Chronic Renal Disease (CRD) is an increasing global health burden, largely driven by the rising prevalence of diabetes and hypertension. Early and accurate identification of renal abnormalities from computed tomography (CT) images is clinically important, yet manual interpretation is time-consuming and conventional machine learning methods often have limited scalability for complex multiclass medical image classification. This study proposes AL-5-ENeT-B4, a modified EfficientNet-B4 architecture enhanc d with five additional layers for automated CRD classification from CT images. The proposed framework is integrated an image preprocessing, data augmentation, stratified 5-fold cross-validation, and deep learning of pre-trained features to classify renal CT scans into one of the four categories: cyst, normal, stone, or tumour. The results of this model were evaluated using 5-fold cross-validation. The proposed model achieved an average K-fold accuracy (99.18%), precision (99.18%), and recall score (99.19%), with an F1-score (99.18%). Furthermore, Grad-CAM visualisation used to provide better interpretability by highlights those renal areas that had clinical significance in relation to the model's decisions. The proposed AL-5-ENeT-B4 model is compared with ResNet50, VGG16, DenseNet121, and ViT-B/16 and demonstrated significantly improved performance achieving an average MCC of 0.9912 and Cohen's Kappa of 0.9912(p < 0.001).