AI Vascular Anomaly Detection With Annotated Vessel Segmentations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI-based methods for vascular abnormality detection in medical imaging face challenges such as reduced robustness due to signal dropout, noise, vessel tortuosity, calcification, and proximity to bone or bifurcations, and are unable to accurately identify the exact location of large vessel occlusions (LVO) within vessel distribution models.
Innovation Solution
A training data preparation method that includes identifying and annotating anatomical landmarks in vessel segmentations, inserting abnormalities like occlusions or aneurysms, and using U-Net Segmentation Networks to enhance the training of AI models for precise vessel segmentation and landmark detection, allowing for robust identification of vascular abnormalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional AI-based approaches are used for automatic interpretation of medical images, then decision making is accelerated, but robustness is reduced in the presence of signal dropout, noise, vessel tortuosity, calcification, and proximity to bone or bifurcations
Solution Approach 1:
The patent segments the vascular abnormality detection task into multiple specialized AI models: a vessel segmentation model, a centerline extraction model, and an abnormality detection model. Each model is optimized for its specific function, allowing the system to maintain high robustness in each sub-task while achieving overall rapid decision-making through the coordinated workflow of these specialized components.
2Extent of automation
If conventional AI-based approaches are used for detecting large vessel occlusions, then automated detection is achieved, but the exact location of occlusion within vessel distribution models cannot be identified
Solution Approach 1:
The patent introduces centerlines as an intermediary representation between vessel segmentation and abnormality detection. The centerline extraction model generates precise central paths of vessels, which serve as a reference framework for the abnormality detection model to accurately locate occlusions within the vessel distribution model, thereby achieving both automation and precise localization.
3Manufacturing precision
If deep learning methodologies are used for vessel tree segmentation, then high Dice overlap is achieved, but the ability to generalize to various pathological scenarios is reduced
Solution Approach 1:
The patent employs a multi-stage dynamic pipeline where the output of each stage serves as input to the next, with each stage being adaptable to different pathological conditions. The vessel segmentation model generates initial masks, the centerline extraction model adapts to various vessel configurations, and the abnormality detection model identifies different types of pathologies, allowing the system to maintain high segmentation precision while generalizing across diverse pathological scenarios.
Data Source
AI summary
A computer-implemented training data preparation method comprises: receiving an input medical image of vessels of a patient; determining a vessel segmentation from the input medical image; identifying and annotating anatomical landmarks in the vessel segmentation to produce an annotated vessel segmentation; and storing the annotated vessel segmentation as training data. A training method for training neural networks based on the training data and a medical diagnostic method applying trained AI models are also provided.