Multimodal CCTA Plaque Identification for Vulnerable Coronary Lesions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current coronary computed tomography angiography (CCTA) methods struggle to accurately assess the risk of vulnerable plaques due to subjective interpretation and labor-intensive diagnostic processes, lacking sufficient capability to identify features like positive remodeling, low density, napkin ring, and spotty calcification.
Innovation Solution
A method utilizing a multimodal large language model (MM-LLM) integrates 3D CCTA images, straightened curved planar reformation (SCPR) images, plaque mask information, fluid dynamics data, and clinical data through preprocessing and feature fusion to enhance vulnerable plaque identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If doctors subjectively interpret the characteristics of vulnerable plaques, then diagnostic flexibility is maintained, but diagnostic accuracy and reliability are insufficient
Solution Approach 1:
The patent segments the vulnerable plaque assessment into four distinct feature dimensions: positive remodeling, low density, napkin ring, and spotty calcification. Each feature is independently quantified and scored, transforming the subjective interpretation process into structured, objective measurements that improve identification accuracy while maintaining manageable complexity through systematic categorization.
Solution Approach 2:
The patent introduces a multimodal large language model (MM-LLM) as an intermediary between the raw CCTA images and the final diagnostic conclusion. This intermediary integrates multiple imaging modalities and clinical data, automatically extracts and scores the four vulnerable plaque features, and generates standardized diagnostic reports, thereby eliminating subjective interpretation while preserving diagnostic flexibility through configurable assessment parameters.
2Measurement precision
If comprehensive feature assessment is performed, then vulnerable plaque identification accuracy improves, but diagnostic time and labor intensity increase
Solution Approach 1:
The patent performs preliminary processing of CCTA images into standardized formats (3D reconstructions, SCPR images, and minimum intensity projection images) before the actual vulnerable plaque assessment. This preliminary action pre-organizes the imaging data and pre-identifies potential plaque regions, so that when the MM-LLM conducts the comprehensive four-feature assessment, it operates on pre-processed data rather than raw images, significantly reducing the time required for the detailed analysis phase.
Solution Approach 2:
The multimodal large language model operates autonomously to perform the comprehensive vulnerable plaque assessment. After receiving the pre-processed imaging data and clinical information, the MM-LLM automatically extracts features, scores each of the four vulnerable plaque characteristics, integrates multiple data sources, and generates the diagnostic conclusion without requiring step-by-step manual intervention, thereby maintaining high assessment accuracy while minimizing additional diagnostic time.
3Adaptability or versatility
If traditional CCTA methods are used, then diagnostic simplicity is maintained, but capability to assess vulnerable plaque risk is insufficient
Solution Approach 1:
The patent merges multiple imaging modalities (3D CCTA images, SCPR images, and minimum intensity projection images) with clinical data into a unified analysis framework. The multimodal large language model integrates these diverse data sources and simultaneously assesses all four vulnerable plaque features (positive remodeling, low density, napkin ring, and spotty calcification), creating a comprehensive risk assessment capability that transcends the limitations of traditional single-modality CCTA analysis.
Solution Approach 2:
The multimodal large language model serves multiple functions within a single system: it processes various imaging formats (3D, SCPR, MIP), extracts multiple feature types (morphological, density, calcification characteristics), integrates clinical data, and generates standardized diagnostic reports. This multi-functional approach enables the system to assess all four vulnerable plaque features and provide comprehensive risk stratification while maintaining a unified, versatile platform that adapts to different clinical scenarios.
Data Source
AI summary
A method and apparatus for identifying a vulnerable coronary plaque based on a multimodal large language model (MM-LLM) are provided. The method includes: acquiring a target 3D coronary computed tomography angiography (CCTA) image and a target straightened curved planar reformation (SCPR) image; acquiring plaque mask information of a blood vessel based on the target SCPR image; acquiring basic plaque data based on the plaque mask information; acquiring fluid dynamics data of each plaque area based on the SCPR image; acquiring basic clinical testing information; standardizing above image data, structural data, and textual data separately, and performing, by a regression mapping network, feature fusion to acquire a first fusion feature; acquiring a trained vulnerable plaque identification LLM; and inputting the first fusion feature into the vulnerable plaque identification LLM to acquire an identification result. The method solve problems of poor single-image feature expression ability and low identification performance in traditional technologies.