Machine Learning Blood Flow Estimation from Vessel Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Invasive methods for assessing blood flow characteristics in patients with arterial disease are risky and costly, while noninvasive simulations using computational fluid dynamics are computationally intensive and challenging to execute in real-time clinical environments.
Innovation Solution
The use of machine learning techniques to predict blood flow characteristics by training algorithms on patient-specific geometric models and physiological data, allowing for rapid and computationally inexpensive estimation of clinically relevant quantities like Fractional Flow Reserve (FFR).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computational fluid dynamics (CFD) simulations are used to estimate blood flow characteristics noninvasively, then measurement precision is improved, but device complexity and computational burden increase
Solution Approach 1:
The patent replaces the complex computational fluid dynamics (CFD) mechanical simulation system with a machine learning-based predictive system. The ML model is trained on CFD simulation data and patient imaging data to predict blood flow characteristics directly, substituting the computationally intensive CFD calculations with a faster ML inference process while maintaining estimation accuracy
Solution Approach 2:
The patent performs preliminary CFD simulations during an offline training phase to generate labeled training data for the machine learning model. This preliminary action allows the ML model to learn from pre-computed CFD results, so that during clinical application, only the faster ML inference is needed without requiring real-time CFD computations
2Measurement precision
If invasive catheterization is used to measure blood flow characteristics, then measurement precision is improved, but object-affected harmful factors increase
Solution Approach 1:
The patent creates a virtual copy of the patient's vascular system through patient-specific geometric models derived from medical imaging (CT or MRI). This digital twin is then used for noninvasive blood flow characterization, eliminating the need for physical invasive catheterization while maintaining the ability to assess hemodynamic properties
Solution Approach 2:
The patent introduces machine learning models as an intermediary between patient imaging data and blood flow characteristic estimation. The ML model processes imaging data and predicts hemodynamic properties without requiring direct physical measurement in the patient, thereby eliminating invasive procedures while preserving measurement capability
3Productivity
If computational fluid dynamics simulations are performed in real-time clinical environment, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent substitutes the computationally intensive CFD simulation engine with a pre-trained machine learning model for real-time clinical deployment. The ML model performs rapid inference on patient imaging data to estimate blood flow characteristics within minutes, achieving real-time productivity without the computational burden of CFD
Solution Approach 2:
The patent performs comprehensive CFD simulations and model training in advance during an offline phase. This preliminary computation creates a ready-to-use ML model that can be deployed in the clinical setting, enabling fast real-time assessments without requiring high-performance computing resources during patient evaluation
Data Source
AI summary
Systems and methods are disclosed for estimating patient-specific blood flow characteristics. One method includes acquiring, for each of a plurality of individuals, a geometric model and estimated blood flow characteristics of at least part of the individual's vascular system; executing a machine learning algorithm on the geometric model and estimated blood flow characteristics for each of the plurality of individuals; identifying, using the machine learning algorithm, features predictive of blood flow characteristics corresponding to a plurality of points in the geometric models; acquiring, for a patient, a geometric model of at least part of the patient's vascular system; and using the identified features to produce estimates of the patient's blood flow characteristic for each of a plurality of points in the patient's geometric model.