Autonomous Coronary Artery Segmentation Using Multi-Plane CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing systems are unable to efficiently perform autonomous segmentation of contrast-filled coronary artery vessels on computed tomography (CT) images, which hinders effective diagnosis, treatment, and monitoring of coronary artery diseases.
Innovation Solution
A computer-implemented method and system utilizing three individually trained convolutional neural networks (CNNs) for preprocessing and segmenting CT images, employing the Jerman filter to emphasize elongated structures and a combination of preprocessing, normalization, and multilayer CNNs for region of interest extraction and coronary vessel segmentation, with data augmentation and loss function optimization for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image processing methods are used for coronary artery segmentation, then the system is simpler to implement, but the segmentation accuracy and efficiency are insufficient
Solution Approach 1:
The patent divides the coronary artery segmentation task into multiple specialized CNN models, each trained to detect specific features (vessel centers, boundaries, contrast-filled regions). This segmentation of the detection task allows each model to achieve high precision for its specific function while maintaining manageable individual model complexity.
Solution Approach 2:
The patent transitions from 2D image analysis to 3D volumetric analysis by processing CT scan slices through multiple CNN layers that build three-dimensional representations of coronary vessels. This dimensional transformation enables more accurate segmentation by capturing spatial relationships that 2D methods miss.
2Productivity
If manual segmentation methods are used, then the system requires less computational power, but the processing time and labor intensity increase significantly
Solution Approach 1:
The patent implements preliminary preprocessing steps including noise filtering, contrast enhancement, and vesselness detection before the main segmentation process. The Jerman filter is applied beforehand to emphasize elongated vascular structures, reducing the computational burden during the actual segmentation phase and improving overall processing efficiency.
Solution Approach 2:
The patent introduces intermediate processing stages with multiple CNN models that work sequentially to refine the segmentation. Each model produces intermediate results that are combined and refined in subsequent stages, distributing the computational workload and enabling parallel processing of different feature detections.
3Measurement precision
If simple preprocessing is applied to CT images, then the processing speed is faster, but the segmentation accuracy of contrast-filled vessels deteriorates
Solution Approach 1:
The patent applies the Jerman filter as a preliminary processing step specifically designed to enhance elongated structures like blood vessels. This filter emphasizes vascular structures before they are fed into the CNN models, improving detection accuracy without requiring extensive preprocessing time.
Solution Approach 2:
The patent applies different preprocessing operations to different regions of the CT images based on local characteristics. Areas with contrast-filled vessels receive enhanced filtering and processing, while other regions use standard preprocessing, optimizing both accuracy and processing time by applying computational resources where most needed.
Data Source
AI summary
A computer-implemented method for autonomous segmentation of contrast-filled coronary artery vessels, the method comprising the following steps: receiving (101) a CT scan volume representing a 3D volume of a region of anatomy that includes a pericardium; preprocessing (102) the CT scan volume to output a preprocessed scan volume; converting (103) the CT scan volume to three sets of two-dimensional slices, wherein the first set is arranged along the axial plane, the second set is arranged along the sagittal plane and the third set is arranged along the coronal plane; extracting (104) a region of interest (ROI) by by autonomous segmentation of the heart region as outlined by the pericardium, by means of three individually trained ROI extraction convolutional neural networks (CNN), each trained to process a particular one of the three sets of two-dimensional slices to output a mask denoting a heart region as delineated by the pericardium; combining (105) the preprocessed scan volume with the mask to obtain a masked volume; converting (106) the masked volume to three groups of sets of two-dimensional masked slices, wherein the first group is arranged along the axial plane, the second group is arranged along the sagittal plane and the third group is arranged along the coronal plane and each group includes at least three sets, wherein the first set corresponds to the principal plane of the set and at least two other sets are tilted with respect to the principal plane; and performing autonomous coronary vessel segmentation (107) by autonomous segmentation of the sets of the two-dimensional masked slices by means of three individually trained segmentation convolutional neural networks (CNN), each trained to process a particular one of the sets of the two-dimensional masked slices to output a mask denoting the coronary vessels.