Vascular Branch Classification via CNN Profile Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting coronary artery centerlines in CTA images often detect false branches, which are time-consuming to manually correct, and previous automated techniques struggle to accurately distinguish true from false branches due to the thin, curvy geometry of vessels and variations in branch location.
Innovation Solution
A deep learning-based approach using a convolutional neural network (CNN) to classify vascular branches by sampling multi-channel one-dimensional profiles along vessel branches, learning to differentiate between true and false branches through a trained classifier, and pruning false branches based on probability scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated centerline extraction methods are used to detect coronary artery branches, then detection sensitivity is improved, but false positive rate increases due to leakage into nearby structures
Solution Approach 1:
The detected centerline tree is divided into multiple branch segments that are sampled and classified individually. Each branch segment is processed through a CNN classifier to determine whether it represents a true coronary artery branch or a false positive detection, enabling localized correction without reprocessing the entire centerline structure.
Solution Approach 2:
Manual interactive editing tools are replaced with an automated machine learning system. The CNN-based classifier automatically distinguishes true branches from false positives by learning from training data, eliminating the need for manual removal of false centerlines while maintaining high detection sensitivity.
2Reliability
If manual interactive editing tools are used to remove false centerlines, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The system performs self-correction by automatically classifying and pruning false positive branches without human intervention. The CNN classifier processes branch segments autonomously, identifying and removing false centerlines based on learned patterns from training data, thereby maintaining detection accuracy while eliminating manual processing time.
Solution Approach 2:
A trained CNN classifier is prepared in advance through offline training on annotated data. This pre-trained model enables rapid automated classification during inference, allowing the system to quickly distinguish true branches from false positives without requiring manual editing during the actual processing workflow.
3Device complexity
If traditional classification methods are used, then computational simplicity is maintained, but ability to handle thin curvy vessel geometry and location variations deteriorates
Solution Approach 1:
The approach transitions from traditional 1D or 2D feature analysis to 3D volumetric processing by sampling branch segments within the 3D context of the centerline tree. The CNN classifier processes spatial relationships and geometric patterns in three dimensions, enabling robust handling of thin curvy vessel geometry and location variations that cannot be captured by simpler methods.
Data Source
AI summary
A method and apparatus for learning based classification of vascular branches to distinguish falsely detected branches from true branches is disclosed. A plurality of overlapping fixed size branch segments are sampled from branches of a detected centerline tree of a target vessel extracted from a medical image of a patient. A plurality of 1D profiles are extracted along each of the overlapping fixed size branch segments. A probability score for each of the overlapping fixed size branch segments is calculated based on the plurality of 1D profiles extracted for each branch segment using a trained deep neural network classifier. The trained deep neural network classifier may be a convolutional neural network (CNN) trained to predict a probability of a branch segment being fully part of a target vessel based on multi-channel 1D input. A final probability score is assigned to each centerline point in the branches of the detected centerline tree based on the probability scores of the overlapping branch segments containing that centerline point. The branches of the detected centerline tree of the target vessel are pruned based on the final probability scores of the centerline points.