Computer-aided cholelithiasis diagnosis using explainable convolutional neural network.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Diagnostic Technologies

Background:

• Accurate cholelithiasis diagnosis is critical for global health.
• Existing computer-aided diagnosis (CAD) systems using Convolutional Neural Network (CNN) models are limited by their black-box nature, hindering clinical trust.
• There is a need for interpretable AI models in medical diagnostics.

Purpose of the Study:

• To propose a novel, interpretable CNN-based approach for cholelithiasis classification using ultrasound images.
• To enhance model generalization through synthetic data generation.
• To improve the transparency and trustworthiness of AI in medical diagnosis.

Main Methods:

• Development of a custom CNN architecture for cholelithiasis classification.
• Utilizing a modified deep convolutional generative adversarial network (DCGAN) to create synthetic ultrasound images.
• Implementing a hybrid visual explanation method combining Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME).
• Performance evaluation on ultrasound images from three Indian hospitals, including validation by radiologists.

Main Results:

• The proposed custom CNN approach demonstrated superior performance compared to state-of-the-art pre-trained CNN and Vision Transformer models.
• The hybrid explanation method generated heatmaps providing detailed visual insights into the model's predictions.
• Radiologist validation confirmed the efficacy and trustworthiness of the model's predictions and explanations.

Conclusions:

• The novel CNN approach with hybrid visual explanations offers an effective and transparent solution for computer-aided cholelithiasis diagnosis.
• The method enhances trust in AI-driven medical diagnostics by providing interpretable results.
• This work contributes to advancing AI applications in medical imaging for improved patient outcomes.