AI Stroke Risk Prediction from Carotid Artery Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting stroke risk, such as computational fluid dynamics and 4D flow MRI, are inefficient due to long processing times, high computational power requirements, and limited anatomical image quality, making them unsuitable for real-time clinical adoption.
Innovation Solution
An AI-based system that rapidly generates flow information from patient-specific carotid artery geometric parameters, using machine-learned models like recurrent neural networks or transformers, to predict stroke risk by perturbing these parameters and selecting candidate flows that match measured flow data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computational fluid dynamics is used to analyze carotid flow, then flow information can be obtained, but processing time is long and computational power requirements are high
Solution Approach 1:
The patent creates a simplified geometric model that copies the essential features of the patient-specific carotid anatomy. This simplified model serves as a surrogate for the complex original geometry, enabling rapid flow analysis while preserving the key hemodynamic characteristics needed for stroke risk assessment
Solution Approach 2:
The patent transforms the complex 3D geometric parameters into a reduced set of key parameters that capture the essential flow dynamics. By changing the parameter representation from detailed spatial coordinates to simplified geometric descriptors, the system achieves faster processing while maintaining diagnostic accuracy
2Measurement precision
If patient-specific geometry is used for CFD analysis, then accurate flow information is obtained, but preprocessing time for segmentation and meshing is long
Solution Approach 1:
The system creates a simplified geometric copy that retains the essential flow-determining features of the patient-specific anatomy. This copied model eliminates the need for time-consuming segmentation and meshing while preserving the hemodynamic information necessary for accurate stroke risk prediction
Solution Approach 2:
The patent extracts only the critical geometric features from the full patient-specific model that are necessary for flow analysis. By taking out and retaining only the essential parameters that influence flow patterns, the system bypasses the lengthy preprocessing steps while maintaining diagnostic accuracy
3Reliability
If traditional methods are used for stroke risk prediction, then comprehensive analysis is performed, but the complex workflow with significant processing time limits clinical adoption
Solution Approach 1:
The patent changes the parameter set from comprehensive detailed geometric parameters to a reduced set of key geometric descriptors. This parameter transformation simplifies the workflow while maintaining the reliability of stroke risk prediction by focusing on the most diagnostically relevant features
Data Source
Figure 1
Figure 2~3B
Figure 4
AI summary
For predicting (140) stroke risk, an artificial intelligence (450) rapidly generates (140) flow information from input of geometric parameters of a carotid of a patient. An image processor (420) predicts (152) the stroke risk from the flow information. In one approach, the values of the geometric parameters of the carotid of the patient are perturbed based on uncertainty (130). The artificial intelligence (450) generates (140) candidate flow information for each perturbation. The candidate flow information sufficiently matching (150) a measurement of flow for the patient is used as the flow information for stroke risk prediction (152).