Digital Mammography Preprocessing Using Probabilistic Hough Transform

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for preprocessing digital mammography images are computationally slow and unable to accurately handle mammography cleavage views or represent the curved shape of the pectoral muscle, limiting their effectiveness in detecting abnormalities and optimizing image contrast.

Innovation Solution

The use of probabilistic Hough transform and a vertical Sobel filter for faster and more accurate segmentation of breast tissue and pectoral muscle, along with pseudo-modality normalization, to enhance image preprocessing for CAD servers and mammography workstations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If region growing method or gradient method is used for breast segmentation, then segmentation can be performed, but processing is computationally slow

Engineering Contradiction:

Improvesegmentation accuracyVSAvoidprocessing speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces traditional mechanical image processing algorithms (region growing, gradient methods) with a neural network-based automated segmentation system. The neural network is trained to directly identify breast tissue boundaries and internal structures, substituting iterative computational methods with a trained model that processes images in a single pass, thereby dramatically improving processing speed while maintaining segmentation accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent implements preliminary action by pre-training the neural network on a large dataset of annotated mammography images before deployment. This pre-training phase establishes the segmentation model in advance, so that during actual clinical use, the network can immediately perform accurate segmentation without requiring complex real-time computations, thus resolving the speed-accuracy tradeoff.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If single-line Hough transform is used for pectoral muscle segmentation, then processing is faster, but segmentation result cannot accurately represent the curved shape of the pectoral muscle

Engineering Contradiction:

Improveprocessing speedVSAvoidcurved shape representation accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies curvature by training the neural network to recognize and segment curved boundaries of the pectoral muscle. Unlike the single-line Hough transform that assumes straight lines, the neural network learns to detect curved edges and surfaces, accurately representing the anatomical curvature of the pectoral muscle while maintaining efficient processing through the trained model's direct inference capability.

Inventive Principle:

Principle #14Spheroidality (Curvature)

3Manufacturing precision

If generalized Hough transform is used for pectoral muscle segmentation, then curved shape can be represented, but calculation is more expansive resulting in slower processing time

Engineering Contradiction:

Improvecurved shape representation accuracyVSAvoidprocessing speed

Core Design Contradiction:

Manufacturing precisionVSProductivity

Solution Approach 1:

The patent substitutes the computationally intensive generalized Hough transform with a neural network model that has been trained to directly detect curved pectoral muscle boundaries. The neural network performs curved edge detection through learned features rather than exhaustive line-by-line Hough transform calculations, achieving both accurate curved shape representation and faster processing speeds.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

4Adaptability or versatility

If existing segmentation methods are used, then one connected region can be detected, but the methods cannot handle a mammography cleavage view which includes the medial portions of both right and left breasts

Engineering Contradiction:

Improvehandling of cleavage viewsVSAvoidsegmentation capability

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent implements universality by designing a neural network segmentation model that can handle multiple view types (standard views and cleavage views) and various anatomical configurations within a single unified framework. The network is trained on diverse datasets including cleavage views with multiple breast regions, enabling it to automatically adapt to different image types and correctly segment multiple connected regions when present, without requiring separate specialized algorithms for each view type.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS8340387B2Fast preprocessing algorithms for digital mammography CAD and workstation

Publication Date: 2012.12.25 THREE PALM SOFTWARE

AI summary

A method and apparatus are disclosed for an image preprocessing device that automatically detects chestwall laterality; removes border artifacts; and segments breast tissue and pectoral muscle from digital mammograms. The algorithms in the preprocessing device utilize the computer cache, a vertical Sobel filter and a probabilistic Hough transform to detect curved edges. The preprocessing result, along with a pseudo-modality normalized image, can be used as input to a CAD (computer-aided detection) server or to a mammography image review workstation. In the case of workstation input, the preprocessing results improve the protocol for chestwall-to-chestwall image hanging, and support optimal image contrast display of each segmented region.