IVUS Plaque Tissue Quantification Using Polar Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing coronary plaque using intravascular ultrasound (IVUS) images are limited by high costs, side effects from contrast media, resolution limitations, and manual quantification time, making accurate tissue component diagnosis challenging, especially for atherosclerotic plaques with lipid necrosis and calcification.
Innovation Solution
A deep learning-based method that extracts IVUS cross-sectional images, applies label indices for plaque components, converts to polar coordinates, and uses an AI model to quantify tissue components, providing a color map for easy identification of attenuation and calcification regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical coherence tomography technique is used for tissue analysis, then measurement precision is improved, but cost increases and side effects occur due to contrast media injection
Solution Approach 1:
The patent creates a virtual copy of the optical coherence tomography analysis capability through deep learning models trained on IVUS images. The AI system learns to identify and quantify tissue components (lipid necrosis, calcification, fibrous tissue) by processing IVUS images, effectively copying the diagnostic functionality of OCT without requiring actual OCT hardware or contrast media injection.
Solution Approach 2:
The patent replaces the mechanical/optical imaging system (OCT) with an information processing system (deep learning). Instead of using physical light waves and optical components to image tissue, the system uses trained neural networks to analyze IVUS images and extract tissue composition information, substituting a computational approach for a physical imaging approach.
2Measurement precision
If manual quantification is used for tissue analysis, then measurement precision is improved, but productivity decreases due to excessive time consumption
Solution Approach 1:
The patent implements self-service by enabling the IVUS imaging system to automatically perform tissue component quantification through integrated deep learning algorithms. The system processes the IVUS images and generates tissue composition analysis without requiring manual intervention, allowing the examination equipment to serve its own analysis needs autonomously.
Solution Approach 2:
The patent replaces manual visual analysis with automated computational analysis. Deep learning models process IVUS images to identify and quantify tissue components, substituting human expert analysis with machine-based image processing that maintains diagnostic accuracy while dramatically reducing analysis time.
3Productivity
If deep learning model is used for tissue analysis, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent applies universality by designing a deep learning framework that handles multiple tissue component types (lipid necrosis, calcification, fibrous tissue) and various IVUS image formats through a single unified model architecture. The system performs multiple diagnostic functions including tissue identification, quantification, and risk assessment using one multi-functional AI model rather than separate specialized models.
Data Source
AI summary
A method of analyzing a plaque tissue component based on deep learning, the method including: extracting a plurality of first intravascular ultrasound (IVUS) cross-sectional images into which a first IVUS image that is a preprocedural IVUS image of a patient is divided at predetermined intervals; labeling each of the plurality of first IVUS cross-sectional images by using label indices corresponding to plaque tissue components to form labeled images, performing image conversion to obtain a polar coordinate image through which a distribution of tissue components for each angle is identifiable by performing a coordinate transformation based on the labeled images, extracting a label vector for each angle based on the polar coordinate image, and outputting output data obtained by quantifying the tissue components for each angle by using an artificial intelligence model that is trained by using, as training data, the label vector for each angle.