Predicting coronary artery occlusion risk from noninvasive images by combining CFD-FSI, cGAN and CNN

Area of Science:

• Cardiovascular fluid mechanics
• Medical imaging analysis
• Artificial intelligence in healthcare

Background:

• Wall Shear Stress (WSS) is crucial for assessing vascular occlusion risk but cannot be directly measured.
• Existing measurement methods suffer from low resolution, uncertainty, and high costs.
• Accurate and rapid risk prediction is essential for effective cardiovascular disease management.

Purpose of the Study:

• To develop a novel, noninvasive method for predicting coronary artery occlusion risk.
• To integrate computational fluid dynamics (CFD), fluid-structure interaction (FSI), and deep learning (cGAN, CNN) for WSS prediction and risk stratification.
• To validate the method using patient-specific MRI data.

Main Methods:

• Developed WSSGAN (a conditional generative adversarial network) trained on CFD-FSI simulations to predict WSS contours.
• Utilized an 11-layer convolutional neural network (CNN) to classify predicted WSS contours into risk grades (low, medium, high).
• Converted patient MRI scans into 3D geometry for WSSGAN input and subsequent CNN risk classification.

Main Results:

• Successfully predicted WSS contours using the WSSGAN model.
• Achieved accurate classification of coronary artery occlusion risk into low, medium, and high grades via the CNN.
• Demonstrated the method's applicability in a real-world case using patient MRI data.

Conclusions:

• The combined CFD-FSI and deep learning approach offers an accurate and rapid noninvasive method for coronary artery occlusion risk prediction.
• This novel technique overcomes the limitations of direct WSS measurement and traditional imaging methods.
• The developed WSSGAN and CNN models show significant potential for clinical application in cardiovascular diagnostics.