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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Chronic Renal Disease (CRD) presents a growing global health challenge, often linked to diabetes and hypertension.
• Accurate, early detection of renal abnormalities in CT images is crucial but challenging due to manual interpretation limitations and conventional AI scalability issues.
• Automated classification of medical images for conditions like CRD requires robust and efficient deep learning models.

Purpose of the Study:

• To develop and evaluate an advanced deep learning model, AL-5-ENeT-B4, for automated classification of Chronic Renal Disease (CRD) from CT images.
• To enhance an EfficientNet-B4 architecture with additional layers for improved accuracy in multiclass renal CT scan classification.
• To compare the performance of the proposed model against established deep learning architectures.

Main Methods:

• A modified EfficientNet-B4 architecture (AL-5-ENeT-B4) was developed, incorporating five additional layers.
• The framework included image preprocessing, data augmentation, and stratified 5-fold cross-validation.
• Deep learning with pre-trained features was employed to classify renal CT scans into four categories: cyst, normal, stone, or tumor.

Main Results:

• The AL-5-ENeT-B4 model achieved high performance metrics, including an average 5-fold cross-validation accuracy of 99.18%, precision of 99.18%, recall of 99.19%, and F1-score of 99.18%.
• Grad-CAM visualization was utilized for enhanced interpretability, highlighting clinically significant regions in the CT scans.
• The proposed model significantly outperformed ResNet50, VGG16, DenseNet121, and ViT-B/16, evidenced by average MCC (0.9912) and Cohen's Kappa (0.9912) values (p < 0.001).

Conclusions:

• The AL-5-ENeT-B4 model demonstrates superior performance and interpretability for automated CRD classification from CT images compared to existing models.
• This AI-driven approach offers a promising tool for early and accurate detection of renal abnormalities, potentially improving patient outcomes.
• The study highlights the effectiveness of modified EfficientNet architectures in complex medical image analysis tasks.