Image Segmentation Consistency Checks for Failure Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods in medical imaging, particularly ultrasound, are prone to failure due to poor image quality, anatomical anomalies, and unexpected imaging views, leading to inaccurate and inconsistent segmentation results, which are difficult to detect.
Innovation Solution
A method that compares segmentation results from multiple images of the same anatomical region taken at different time points to identify inconsistencies, using machine learning algorithms to determine the accuracy of the segmentation procedure by measuring consistency between these results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If model-based segmentation is used to automate delineation of anatomical structures, then productivity is improved, but reliability deteriorates due to segmentation failures from poor image quality and anatomical anomalies
Solution Approach 1:
The patent implements a feedback mechanism by comparing segmentation results across multiple images of the same anatomical region. The system computes a consistency measure between segmentations and uses this feedback to determine whether the segmentation is accurate or has failed, allowing automatic correction or rejection of unreliable segmentations
Solution Approach 2:
The patent introduces an intermediary consistency measure that acts as a mediator between the segmentation algorithm and the final output. This consistency measure compares segmentations across multiple images and serves as an intermediate validation step before accepting the segmentation result, preventing unreliable automated delineation from being accepted
2Productivity
If segmentation algorithms are applied to poor quality images, then productivity is maintained, but measurement precision deteriorates due to inaccurate boundary detection
Solution Approach 1:
The system uses feedback from consistency comparison across multiple images to identify when boundary delineation has failed. By computing consistency measures between segmentations of the same anatomical region imaged at different times, the system can detect when poor image quality has led to inaccurate boundary detection and reject those results
3Productivity
If automated segmentation is used to reduce manual assessment time, then productivity is improved, but device complexity increases due to need for multiple images and consistency checking
Solution Approach 1:
The system implements self-service by having the segmentation algorithm validate its own results through consistency checking. The automated segmentation system performs its own quality control by comparing results across multiple images and determining accuracy without requiring external manual verification, thereby managing complexity through self-validation
Data Source
Figure 1~3
Figure 4
AI summary
An image analysis method and device is for detecting failure or error in an image segmentation procedure. The method comprises comparing (14) segmentation outcomes for two or more images, representative of a particular anatomical region at different respective time points, and identifying a degree of consistency or deviation between them. Based on this derived consistency or deviation measure, a measure of accuracy of the segmentation procedure is determined (16).