An attention enhanced dilated bottleneck network for kidney disease classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Computer-Aided Design (CAD) aids nephrologists in optimizing workflows and managing large datasets.
• Accurate diagnosis and classification of kidney diseases (cysts, stones, tumors) remain challenging due to complex features.

Purpose of the Study:

• To introduce the Dilated Bottleneck Attention-based Renal Network (DBAR-Net) for automated kidney disease diagnosis and classification.
• To enhance feature representation and model robustness for improved diagnostic accuracy.

Main Methods:

• Developed DBAR-Net incorporating multi-feature fusion, dual bottleneck attention modules, and dilated convolved layer normalization.
• Utilized two-fold convolved layer normalization blocks for fine-grained detail and abstract pattern capture.
• Employed adaptive contextual features from dilated convolution for semantic interpretation.

Main Results:

• DBAR-Net achieved a F1 score of 0.98 and 98.86% classification accuracy on the CT KIDNEY DATASET (8750 images).
• The model demonstrated superior performance compared to VGG16, VGG19, ResNet50, EfficientNetB0, Inception V3, MobileNetV2, and Xception.
• Achieved precise multi-class kidney disease detection with minimal computational complexity.

Conclusions:

• DBAR-Net offers a highly accurate and efficient automated solution for kidney disease diagnosis.
• The proposed architecture effectively handles complex features, leading to enhanced diagnostic capabilities.
• This AI-driven approach shows significant potential for improving clinical workflows in nephrology.