An explainable adaptive channel weighting-based deep convolutional neural network for classifying renal disorders in...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Renal disorders pose significant public health risks, often leading to renal failure.
• Manual diagnosis of renal conditions from CT scans is subjective and requires specialized expertise.
• Current diagnostic methods can be labor-intensive, impacting workflow efficiency.

Purpose of the Study:

• To develop an automated deep learning model for accurate and efficient renal disorder diagnosis from CT images.
• To enhance diagnostic accuracy and workflow efficiency in nephrology.
• To improve the interpretability of AI-driven diagnostic models.

Main Methods:

• Proposed EACWNet model: an adaptive channel weighting-based deep convolutional neural network integrated with explainable AI.
• Utilized VGG-19 backbone with a scale-adaptive channel attention module for feature refinement.
• Trained and evaluated on a public renal CT image dataset.

Main Results:

• EACWNet achieved 98.87% accuracy in classifying renal CT images (cyst, normal, tumor, stone).
• Demonstrated a 1.75% performance improvement over the baseline VGG-19 model.
• Exhibited class-wise precision variations, with lower precision for the heterogeneous stone class.

Conclusions:

• The EACWNet model offers a promising automated approach for renal disorder diagnosis, enhancing accuracy and efficiency.
• Explainable AI methods provide insights into model predictions, aiding clinical understanding.
• Further refinement is needed to address class-specific performance variations, particularly for renal stones.