Medical Image Post-Processing for Organ Segmentation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current model-based organ segmentation methods, such as deep convolutional neural networks (DCNN), often produce suboptimal results due to lack of explicit anatomical information and human-defined criteria, leading to incorrect voxel-by-voxel segmentations in medical images like CT and MR images, particularly in defining start and end slices for organs like the heart.
Innovation Solution
A computer-implemented image post-processing method that analyzes three-dimensional CT or MR images to identify and correct anatomically incorrect segmentations by applying anatomical rules, removing segmentations beyond predefined slices, and using 3D and 2D analysis to ensure spatial connectivity and convex boundaries, thereby improving organ segmentation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If voxel-by-voxel segmentation is used by DCNN, then automation is improved, but anatomical accuracy deteriorates
Solution Approach 1:
The post-processing method segments the organ into connected components and identifies anatomically incorrect regions by analyzing spatial relationships and connectivity patterns, allowing correction of voxel-by-voxel segmentation errors while preserving automation
Solution Approach 2:
A post-processing module acts as an intermediary between the DCNN segmentation output and final results, applying anatomical knowledge and spatial analysis to correct errors without requiring manual intervention, thus maintaining automation while improving accuracy
2Productivity
If DCNN segmentation is applied, then processing speed is improved, but segmentation precision deteriorates
Solution Approach 1:
The post-processing method performs preliminary identification of anatomically incorrect regions before final segmentation output, using efficient spatial analysis and connectivity checks that maintain processing speed while improving precision
Solution Approach 2:
The method replaces manual segmentation review with automated post-processing algorithms that use anatomical rules and spatial relationships to identify and correct errors, maintaining high processing speed while improving segmentation precision
3Manufacturing precision
If anatomical rules are applied in post-processing, then segmentation accuracy is improved, but device complexity increases
Solution Approach 1:
The post-processing method applies anatomical rules locally to specific regions identified as potentially incorrect, rather than processing the entire organ uniformly, thus improving accuracy while minimizing additional complexity
Solution Approach 2:
The method changes the analysis parameters from simple voxel classification to spatial connectivity and anatomical relationship analysis only where needed, improving segmentation accuracy through targeted application of complex rules
Data Source
AI summary
The present disclosure relates to a method and apparatus to improve model-based organ segmentation with image post-processing using one or more processors. The method includes: receiving three-dimensional (3D) CT or MR images and the corresponding voxel-by-voxel segmentation results from the output of an automatic DCNN segmentation model, analyzing the segmentation results on a 3D basis according to organ anatomical information, processing the results to identify regions with anatomically incorrect segmentations, fixing the incorrect segmentations, analyzing the remaining segmentation results on a two-dimensional (2D) and slice-by-slice basis according to predefined organ segmentation criteria, identifying the start and end slices, processing the results to remove segmentations beyond the identified slices to obtain the final results.