Nested Graph-Cut Segmentation for Ultrasound Tissue Boundaries

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image segmentation methods for high-frequency ultrasound and quantitative ultrasound images face challenges with accurate intensity distribution and shape preservation, particularly in nested structures, leading to unsatisfactory segmentation results due to acoustic attenuation and missing boundaries, which require manual initialization and extensive datasets for training.

Innovation Solution

An iterative framework comprising a nested graph-cut algorithm for segmentation and spline-based distribution estimation, which exploits nested relationships between objects, eliminates the need for initialization and uses undirected links and padded nodes to avoid incorrect segmentations, allowing for automatic segmentation of nested objects without initial seeds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If clustering algorithms (K-means or EM) are used to categorize intensity values, then segmentation can be performed, but intensity fluctuates significantly with depth due to acoustic attenuation and focusing effects

Engineering Contradiction:

Improveautomatic segmentationVSAvoidintensity distribution accuracy

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

The patent divides the image into multiple depth bands and applies clustering algorithms separately in each band. This local segmentation approach allows the intensity distribution to be modeled independently for different depth regions, compensating for acoustic attenuation and focusing effects that cause global intensity variations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies different clustering parameters and intensity distribution models to different depth bands. By adapting the segmentation parameters locally to each depth region, the method accounts for the varying acoustic properties at different depths, improving measurement precision while maintaining automatic segmentation.

Inventive Principle:

Principle #3Local quality

2Shape

If level-set based method with super-ellipse model is used, then shape preservation is improved, but proper initialization is required

Engineering Contradiction:

Improveshape preservationVSAvoidinitialization requirement

Core Design Contradiction:

ShapeVSEase of operation

Solution Approach 1:

The patent employs an active shape model that automatically adapts to the specific shape characteristics of each target object without requiring manual initialization. The model learns shape variations from the image data itself and performs segmentation autonomously, eliminating the need for user-provided initial contours while maintaining accurate shape preservation.

Inventive Principle:

Principle #25Self-service

3Reliability

If active shape model (ASM) is used to segment objects with missing boundaries, then shape modeling is improved, but a large number of manually segmented datasets are necessary to generate sufficient eigen-shapes

Engineering Contradiction:

Improvesegmentation reliability with missing boundariesVSAvoidmanual segmentation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent pre-trains the active shape model using a small set of manually segmented examples to learn the statistical distribution of shape variations. Once trained, the model can automatically segment new images with missing boundaries without requiring additional manual annotations, thus achieving reliable segmentation while minimizing manual time investment.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a statistical shape model from a small training set that can be copied and applied to multiple new images. This trained model captures the essential shape variations and can be reused across different datasets, eliminating the need to perform manual segmentation for each new image while maintaining segmentation reliability.

Inventive Principle:

Principle #26Copying

4Extent of automation

If graph cuts with layer-cake structure are used, then multiple objects can be segmented, but the regularization cost is not discontinuity-preserving and requires seeds

Engineering Contradiction:

Improvemulti-object segmentation capabilityVSAvoidseed initialization and discontinuity preservation

Core Design Contradiction:

Extent of automationVSEase of operation

Solution Approach 1:

The patent uses a dynamic programming approach that automatically determines the optimal segmentation configuration for multiple objects without requiring fixed seed initialization. The algorithm dynamically adapts to the image content and intensity variations, preserving discontinuities at object boundaries while maintaining automation across multiple nested objects.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS10157466B2Method for automatic tissue segmentation of medical images

Publication Date: 2018.12.18 RIVERSIDE RES INST

AI summary

A method to segment images that contain multiple objects in a nested structure including acquiring an image; defining the multiple objects by layers, each layer corresponding to one region, where a region contains an innermost object and all the objects nested within the innermost object; stacking the layers in an order of the nested structure of the multiple objects, the stack of layers having at least a top layer and a bottom layer; extending each layer with padded nodes; connecting the top layer to a sink and the bottom layer to a source, wherein each intermediate layer between the top layer and the bottom layer are connected only to the adjacent layer by undirected links; and measuring a boundary length for each layer.