3D Blood Vessel Extraction Using Neural Network Graph Reconstruction
Solution Overview
Problem
Existing medical imaging techniques require manual, time-consuming segmentation and classification of blood vessels, which are tedious and difficult to utilize for surgical planning.
Innovation Solution
A neural network-based system that automatically segments and classifies blood vessels in volumetric images, using a 3D U-Net style architecture, and generates 3D models by detecting roots and endpoints, determining shortest paths, and combining them into directed graphs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual segmentation and classification of blood vessels is performed, then accuracy of vascular structure identification is improved, but time consumption and operational complexity increase significantly
Solution Approach 1:
The system performs preliminary automated segmentation using a neural network to generate an initial 3D vascular model before manual review. This preliminary action prepares the data in advance, reducing the time clinicians need to spend on manual segmentation while maintaining accuracy through subsequent validation and correction steps.
Solution Approach 2:
An automated neural network segmentation system acts as an intermediary between the raw volumetric imaging data and the clinician's manual segmentation work. This intermediary performs the time-consuming initial segmentation, allowing clinicians to focus on reviewing and correcting results rather than performing complete manual segmentation from scratch.
2Manufacturing precision
If manual pixel-by-pixel segmentation is performed, then precision of blood vessel boundaries is improved, but productivity and efficiency decrease
Solution Approach 1:
The neural network performs preliminary segmentation of blood vessel boundaries at automated speed, generating a precise 3D model that clinicians can then review and correct. This preliminary precise segmentation at high speed replaces the need for manual pixel-by-pixel work while maintaining boundary accuracy.
Solution Approach 2:
The system replaces the mechanical manual pixel-by-pixel segmentation process with an automated neural network-based image processing system. This substitution maintains the precision of boundary identification while dramatically increasing segmentation speed and productivity.
3Reliability
If complete manual segmentation of all blood vessels is performed, then completeness of vascular tree structure is improved, but operational complexity and time requirement increase
Solution Approach 1:
The system performs preliminary automated segmentation to generate a complete 3D vascular tree structure before manual review. This preliminary complete segmentation reduces the complexity of manual operations by providing a ready-made structure that clinicians can validate and correct rather than build from scratch.
Solution Approach 2:
The automated neural network segmentation system serves as an intermediary that handles the complex task of generating a complete vascular tree structure. This intermediary performs the complex automated segmentation, allowing clinicians to focus on simpler validation and correction tasks, thereby reducing overall operational complexity.
Data Source
AI summary
Systems and methods of image processing include a processor in communication with a display and a computer readable recording medium having instructions executed by the processor to read a three-dimensional (3D) image data set from the computer-readable recording medium and automatically generate a tree structure of blood vessels based on patient images of the image data set using a neural network. Manually- and/or semi-automatically-generated 3D models of blood vessels are used to train the neural network. The systems and methods involve segmenting and classifying blood vessels in the 3D image data set using the trained neural network, closing holes, finding roots and endpoints in the segmentation, finding shortest paths between the roots and endpoints, selecting most probable paths, combining most probable paths into directed graphs, solving overlaps between directed graphs, and creating 3D models of blood vessels based on the directed graphs.