3D CT Coronary Vessel Segmentation with Autonomous ROI Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems are incapable of efficiently providing autonomous segmentation of contrast-filled coronary artery vessels on CT images, hindering effective diagnosis, treatment, and monitoring of coronary artery diseases.
Innovation Solution
A computer-implemented method using machine learning, specifically convolutional neural networks (CNNs), for autonomous segmentation of coronary artery vessels involves preprocessing CT scan volumes, extracting regions of interest using individually trained ROI extraction CNNs, and performing coronary vessel segmentation through multiple trained segmentation CNNs to output masks denoting coronary vessels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If traditional image processing methods are used, then system simplicity is maintained, but autonomous segmentation capability is insufficient
Solution Approach 1:
The system uses trained CNN models to automatically perform ROI extraction and coronary vessel segmentation without requiring manual intervention or complex preprocessing by operators. The models autonomously process CT scan volumes, extract relevant regions, and generate segmentation masks, enabling the system to serve itself in the segmentation task.
Solution Approach 2:
Traditional manual or rule-based image processing methods are replaced with deep learning-based CNN models. The mechanical/manual segmentation process is substituted with automated neural network inference, which learns complex patterns from training data and performs segmentation autonomously, resolving the contradiction between automation extent and system complexity.
2Measurement precision
If 3D volume processing is performed directly, then segmentation accuracy may be improved, but computational complexity and processing time increase
Solution Approach 1:
The 3D CT scan volume is divided into multiple 2D slices along different anatomical planes (axial, sagittal, coronal). Each slice is processed independently by the CNN model, which reduces the computational burden compared to processing the entire 3D volume at once. The individual slice predictions are then integrated to form the complete 3D segmentation, maintaining accuracy while reducing processing time.
Solution Approach 2:
The approach transforms the 3D volume processing problem into multiple 2D slice processing tasks. By converting the volumetric data into two-dimensional slices across different planes, the system leverages the efficiency of 2D CNN processing while still capturing three-dimensional anatomical information through multi-planar reconstruction and integration.
3Reliability
If multiple processing planes are used, then segmentation completeness is improved, but system complexity increases
Solution Approach 1:
The same CNN model architecture and processing pipeline are universally applied across multiple anatomical planes (axial, sagittal, coronal). Rather than developing separate specialized systems for each plane, a single versatile model processes all planes, reducing system complexity while improving segmentation completeness by capturing vascular structures from multiple perspectives.
Data Source
AI summary
A computer-implemented method for autonomous segmentation of contrast-filled coronary artery vessels includes receiving a CT scan volume representing a 3D volume of a region of anatomy that includes a pericardium; preprocessing the CT scan volume to output a preprocessed scan volume; dividing the CT scan volume into a first set of subvolumes; extracting a region of interest by autonomous segmentation of the heart region as outlined by the pericardium, by means of a neural network trained on 3D subvolumes and combining the results of the individual subvolume predictions for the first set to output a mask denoting a heart region as delineated by the pericardium; combining the preprocessed scan volume with the mask to obtain a masked volume; converting the masked volume to a second set of 3D subvolumes; and performing autonomous coronary vessel segmentation to output a mask denoting the coronary vessels.