Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attention-based Multi-fidelity Machine Learning Model for Computational Fractional Flow Reserve Assessment

Haizhou Yang, C. Alberto Figueroa, Krishna Garikipati

TL;DR

An attention-based multi-fidelity machine learning model (AttMulFid) is proposed for computationally efficient and accurate FFR assessment with uncertainty measurement and shows that the geometric features are able to represent the entirety of the geometric information and intelligently allocate attention based on crucial properties of geometry.

Abstract

Coronary Artery Disease (CAD) is one of the most common forms of heart disease, which is caused by a buildup of atherosclerotic plaque (known as stenosis) in the coronary arteries, leading to insufficient supplement of blood, oxygen, and nutrients to the heart. Fractional Flow Reserve (FFR), measuring the pressure ratio between the aorta and distal coronary artery, is an invasive physiologic gold standard for assessing the severity of coronary artery stenosis. Despite its benefits, invasive FFR assessment is still underutilized due to its high cost, time-consuming, experimental variability, and increased risk to patients. In this study, an attention-based multi-fidelity machine learning model (AttMulFid) is proposed for computationally efficient and accurate FFR assessment with uncertainty measurement. Within AttMulFid, an autoencoder is utilized to intelligently select geometric features from coronary arteries, with additional attention on the key area. Results show that the geometric features are able to represent the entirety of the geometric information and intelligently allocate attention based on crucial properties of geometry. Furthermore, the AttMulFid is a feasible approach for non-invasive, rapid, and accurate FFR assessment (with 0.002s/simulation).

Attention-based Multi-fidelity Machine Learning Model for Computational Fractional Flow Reserve Assessment

TL;DR

An attention-based multi-fidelity machine learning model (AttMulFid) is proposed for computationally efficient and accurate FFR assessment with uncertainty measurement and shows that the geometric features are able to represent the entirety of the geometric information and intelligently allocate attention based on crucial properties of geometry.

Abstract

Coronary Artery Disease (CAD) is one of the most common forms of heart disease, which is caused by a buildup of atherosclerotic plaque (known as stenosis) in the coronary arteries, leading to insufficient supplement of blood, oxygen, and nutrients to the heart. Fractional Flow Reserve (FFR), measuring the pressure ratio between the aorta and distal coronary artery, is an invasive physiologic gold standard for assessing the severity of coronary artery stenosis. Despite its benefits, invasive FFR assessment is still underutilized due to its high cost, time-consuming, experimental variability, and increased risk to patients. In this study, an attention-based multi-fidelity machine learning model (AttMulFid) is proposed for computationally efficient and accurate FFR assessment with uncertainty measurement. Within AttMulFid, an autoencoder is utilized to intelligently select geometric features from coronary arteries, with additional attention on the key area. Results show that the geometric features are able to represent the entirety of the geometric information and intelligently allocate attention based on crucial properties of geometry. Furthermore, the AttMulFid is a feasible approach for non-invasive, rapid, and accurate FFR assessment (with 0.002s/simulation).
Paper Structure (20 sections, 10 equations, 12 figures, 5 tables)

This paper contains 20 sections, 10 equations, 12 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Data Generation
  4. Low-fidelity Physics-based Model
  5. High-fidelity Computational Fluid Dynamic Model
  6. Feature Extraction
  7. U-Net-based Multi-fidelity Model with Feature Fusion
  8. Uncertainty Quantification
  9. Gradient-based Attention
  10. Results and Discussion
  11. Feature Extraction and Geometry Reconstruction
  12. Synthetic Vessels
  13. Patient Vessels
  14. Hemodynamic Assessment
  15. Synthetic Vessels
  16. ...and 5 more sections

Figures (12)

  • Figure 1: The framework of AttMulFid generation and utilization
  • Figure 2: Comparison between pressure profiles generated by low-fidelity and high-fidelity with the same vessel geometry and boundary conditions
  • Figure 3: The architecture of autoencoder
  • Figure 4: Archetechture of FFU-Net
  • Figure 5: Gradient-based attention generation
  • ...and 7 more figures