Deep Neural Network Multi-Label Segmentation for Cardiac CT Lesion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual assessment of coronary lesions in cardiac computed tomography and angiography (CCTA) images is labor-intensive, time-consuming, and often produces inconsistent diagnoses due to the complex, three-dimensional nature of coronary arteries, which can vary when compared to other imaging modalities.
Innovation Solution
A method using deep neural networks to detect and assess coronary lesions by mapping 3D CCTA images to multi-label segmentation maps, generating one-dimensional parametric curves for the coronary tree, and determining lesion severity based on these curves, enabling automatic and robust detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual assessment of coronary lesions is performed by clinicians, then diagnostic accuracy can be maintained through expert judgment, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The patent replaces the manual mechanical assessment process with an automated deep neural network system. The CNN-based algorithm automatically segments coronary arteries, generates centerline curves, and quantifies lesion severity from CCTA images, eliminating the need for manual clinician measurement while maintaining diagnostic accuracy through learned patterns from training data.
Solution Approach 2:
The system enables self-service automated assessment where the deep neural network performs lesion detection and severity quantification without requiring manual expert intervention. The algorithm independently processes CCTA images, generates segmentation maps, extracts parametric curves, and computes stenosis percentages, making the diagnostic process autonomous and highly efficient.
2Reliability
If manual assessment of coronary lesions is performed by clinicians, then diagnostic judgment can be applied, but inconsistency in diagnoses occurs across different clinicians or imaging modalities
Solution Approach 1:
The patent transforms subjective diagnostic judgment into objective quantitative parameters. By computing precise stenosis percentages, plaque volumes, and luminal area reductions from automated segmentation and centerline analysis, the system replaces variable human assessment with consistent mathematical measurements that eliminate inter-clinician variability and improve diagnostic reliability.
3Productivity
If deep neural networks are used for automatic lesion detection, then assessment time is significantly reduced, but the complexity of the system increases
Solution Approach 1:
The patent segments the complex diagnostic task into distinct modular processing stages: (1) CNN-based multi-label segmentation of coronary arteries and plaques, (2) centerline extraction and parametric curve generation, (3) lesion detection and severity quantification. This segmentation allows each module to be independently optimized and trained, managing overall system complexity while achieving high-speed automated assessment.
4Ease of operation
If manual visualization tools are used to view 3D coronary structures, then clinicians can assess lesions from multiple perspectives, but the process becomes labor-intensive
Solution Approach 1:
The patent replaces manual manipulation of 3D visualization tools with automated computational processing. The deep neural network system automatically generates segmentation maps, extracts centerline curves, and calculates lesion metrics from 3D CCTA data without requiring clinicians to manually rotate, slice, or measure structures, thereby maintaining comprehensive 3D analysis capability while dramatically improving assessment efficiency.
Data Source
AI summary
Methods and systems are provided for detecting coronary lesions in 3D cardiac computed tomography and angiography (CCTA) images using deep neural networks. In an exemplary embodiment, a method for detecting coronary lesions in 3D CCTA images comprises, acquiring a 3D CCTA image of a coronary tree, mapping the 3D CCTA image to a multi-label segmentation map with a trained deep neural network, generating a plurality of 1D parametric curves for a branch of the coronary tree using the multi-label segmentation map, determining a location of a lesion in the branch of the coronary tree using the plurality of 1D parametric curves, and determining a severity score for the lesion based on the plurality of 1D parametric curves.