Coronary Artery Stenosis Estimation Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the extent of anatomical narrowing in coronary artery disease, such as quantitative coronary angiography and quantitative computed tomography, are invasive or time-consuming and struggle to accurately estimate healthy lumen diameters in diffuse, ostial, and bifurcation lesions where clear distinctions between healthy and diseased lumen diameters are absent.
Innovation Solution
The system and method use a database of healthy vessel sections from individuals to estimate healthy lumen diameters through machine learning algorithms, including robust kernel regression and random forest regressors, to calculate a vessel lumen narrowing score by comparing actual lumen radii to estimated healthy radii, effectively addressing the challenge of non-focal stenoses and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If quantitative coronary angiography (QCA) is used to measure percent stenosis, then measurement precision is improved, but the procedure becomes invasive and time-consuming
Solution Approach 1:
The patent uses cCTA imaging to create a non-invasive copy or representation of the coronary artery anatomy, allowing percent stenosis measurement without requiring invasive QCA procedures. The cCTA provides a virtual model that can be analyzed to determine lumen diameter and stenosis extent, replacing the need for actual catheter-based measurements while maintaining diagnostic accuracy
Solution Approach 2:
The patent replaces the mechanical invasive QCA system with a non-invasive computational imaging approach. Instead of using physical catheters and mechanical measurement tools inside the artery, the system uses cCTA images and automated image processing algorithms to measure lumen diameter and calculate percent stenosis, eliminating the need for invasive mechanical intervention
2Measurement precision
If traditional methods are used to estimate healthy lumen diameter, then simplicity is maintained, but accuracy deteriorates in diffuse, ostial, and bifurcation lesions
Solution Approach 1:
The patent performs preliminary segmentation of the coronary artery lumen from cCTA images before stenosis measurement. By pre-processing the images to create accurate lumen masks and centerlines, the system establishes a foundation for precise diameter measurement. This preliminary segmentation step enables accurate healthy lumen diameter estimation even in complex lesion types where traditional methods fail
Solution Approach 2:
The patent introduces an intermediary reference vessel diameter derived from population-based data and anatomical models. Instead of directly measuring healthy diameter in็ ๅ areas where it's ambiguous, the system uses reference values from healthy vessels of similar size and location as intermediaries to estimate the expected healthy diameter, which then serves as the denominator for percent stenosis calculation
3Measurement precision
If manual core lab analysis is performed for percent stenosis determination, then measurement precision is improved, but productivity decreases due to time consumption
Solution Approach 1:
The patent implements an automated system that performs percent stenosis measurement without requiring manual core lab analysis. The system automatically segments the lumen, identifies the stenotic segment, measures diameters, and calculates percent stenosis from cCTA images. This self-service automation eliminates the need for time-consuming manual measurements by radiologists or core lab personnel while maintaining measurement accuracy
Solution Approach 2:
The patent employs advanced image processing algorithms and computational methods that accelerate the analysis process. By using automated detection algorithms, machine learning models, and efficient computational geometry methods, the system rapidly processes cCTA images to determine percent stenosis, dramatically increasing productivity compared to manual analysis while preserving measurement precision
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A computer-implemented method of identifying a healthy lumen diameter of a vasculature, the method comprising: receiving (301) a data set including one or more lumen segmentations of known healthy vessel segments of a population of individuals; splitting (303) each of the lumen segmentations of the population of individuals into stem-crown-root units; for each stem-crown-root unit, extracting (305) one or more features selected in a list comprising: lumen local diameter using maximum inscribed spheres, and lumen planar area; comparing (307) features from the crown and root units to features of the stem unit, by a machine learning algorithm, to infer healthy lumen diameter at the stem; and using (309) a random forest regressor to determine a population-based healthy lumen diameter for the stem based on the comparison of features.