3D Medical Image Segmentation via Probability Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for segmenting anatomic entities in digital medical images, such as active shape models, face limitations including the need for initial estimates, potential misalignment between gray-level and shape models, and incorrect landmark location searches, leading to poor segmentation results.
Innovation Solution
A method that constructs both gray value and shape models from training images to segment anatomic entities by using probability distributions of gray values and connection vectors between landmarks, employing Mahalanobis distance for accurate landmark positioning and dynamic programming for optimal contour fitting, ensuring accurate geometric measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If active shape models are used for segmentation, then segmentation automation is achieved, but initial estimates and model misalignment are required, reducing reliability
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing probability distributions of gray values and connection vectors from training images before actual segmentation. This allows the system to automatically locate landmarks without requiring initial estimates or manual intervention, eliminating the reliability issues associated with model misalignment while maintaining automation.
Solution Approach 2:
The patent uses copying by creating a library of probability distributions from training images that replicate the statistical characteristics of anatomical structures. These stored probability models are then applied to new images to automatically locate landmarks, eliminating the need for initial estimates and improving reliability without sacrificing automation.
2Measurement precision
If landmark location search is performed, then segmentation precision is improved, but incorrect landmark positioning occurs, reducing measurement accuracy
Solution Approach 1:
The patent implements feedback by using probability distributions derived from training images to continuously guide and verify landmark location searches. The probability distributions provide statistical feedback on likely landmark positions, allowing the system to automatically correct positioning errors and ensure accurate landmark locations without manual intervention.
Solution Approach 2:
The patent replaces mechanical search methods with probability-based statistical modeling. Instead of relying on manual or heuristic search algorithms that can fail, the system uses probability distributions of gray values and connection vectors to automatically and reliably locate landmarks, improving both precision and correctness.
3Productivity
If geometric measurements are automated, then productivity is improved, but measurement reliability decreases due to segmentation errors
Solution Approach 1:
The patent applies preliminary action by pre-computing probability distributions from training images that encode accurate anatomical geometry information. This allows automated measurements to be performed directly on new images using these pre-learned statistical models, achieving both high productivity through automation and high reliability through the accuracy of the pre-computed probability distributions.
Solution Approach 2:
The patent uses copying by storing accurate geometric information from training images in the form of probability distributions. These copied statistical models are then applied to automatically measure new images with the same accuracy as manual measurements, achieving both productivity improvement through automation and reliability through the use of accurate reference data.
Data Source
AI summary
For each of a number of landmarks in an image an initial position of the landmark is defined. Next a neighborhood around the initial position comprising a number of candidate locations of the landmark, is sampled and a cost is associated with each of the candidate locations. A cost function expressing a weighted sum of overall gray level cost and overall shape cost for all candidate locations is optimized. A segmented anatomic entity is defined as a path through a selected combination of candidate locations for which combination the cost function is optimized. During optimization towards the optimal segmented surface/volume graph traversal methods are exploited.