Optimization of FFR prediction algorithm for gray zone by hemodynamic features with synthetic model and biometric...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Computational Fluid Dynamics

Background:

• Previous AI coronary diagnosis lacked sufficient data and quality.
• Studies often overlooked crucial flow and biometric features, focusing only on vessel images.
• Optimizing accuracy in the gray zone is vital for stent insertion decisions based on estimated fractional flow reserve (FFR).

Purpose of the Study:

• To develop an AI-driven coronary vascular diagnosis system with enhanced gray zone performance.
• To automate image processing and integrate flow and biometric features for FFR estimation.
• To introduce a comprehensive FFR pre-screening system from CT image extraction to value estimation.

Main Methods:

• Utilized automatic image extraction, lattice Boltzmann method-based CFD analysis, and AI algorithm optimization.
• Developed a two-step algorithm: flow feature calculation from geometrical data, followed by FFR estimation using flow and biometric data.
• Implemented algorithm selection, outlier elimination, and k-fold cross-validation for optimization.

Main Results:

• Tested eight algorithms, including neural networks and machine learning models.
• The random forest model initially showed superior performance.
• The multilayer perceptron regressor demonstrated the highest accuracy in the gray zone post-optimization.

Conclusions:

• The optimized multilayer perceptron regressor significantly improved gray zone accuracy for coronary diagnosis.
• The developed system offers a potential solution for pre-screening and FFR estimation.
• Integration of diverse features and optimized algorithms enhances AI diagnostic capabilities.