Deep Learning Image Reconstruction for Mammography

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current image reconstruction techniques in mammography, such as digital breast tomosynthesis (DBT), face challenges in achieving perfect image reconstruction due to noise and streak lines, leading to suboptimal diagnostic quality and requiring multiple radiation exposures, which increases patient dose and analysis time.

Innovation Solution

The use of deep learning-based methods and systems that process and reconstruct images by obtaining multiple 2D tomosynthesis projection images from varying angles, generating synthetic 2D images that minimize similarity metrics with standard mammography images, thereby enhancing image quality and reducing radiation exposure.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If standard mammography is used to obtain images, then image quality for microcalcifications is improved, but radiation dose to the patient increases

Engineering Contradiction:
Improveimage qualityVSAvoidradiation dose
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent segments the imaging process into multiple low-dose projection acquisitions at different angles, then reconstructs a 3D volume from these segments. This allows the system to achieve comprehensive diagnostic information without requiring a single high-dose exposure, resolving the contradiction between image quality and radiation dose.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from 2D projection imaging to 3D volumetric imaging. By adding the temporal/angular dimension and reconstructing a 3D volume from multiple 2D projections, the system achieves superior diagnostic capability without proportionally increasing radiation dose, as each projection uses lower dose than traditional mammography.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If tomosynthesis is used to obtain 3D representation, then localization of lesions is improved, but image reconstruction quality deteriorates due to noise and streak lines

Engineering Contradiction:
Improvelocalization accuracyVSAvoidimage reconstruction quality
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

The patent replaces traditional iterative algebraic reconstruction techniques with a deep learning-based neural network approach. The neural network learns optimal reconstruction patterns from training data, substituting the mechanical iterative computation with a learned model that produces higher quality images with reduced noise and streak artifacts.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the reconstruction parameters by using a trained neural network model with optimized weights and biases. Instead of fixed iterative parameters, the system uses learned parameters that adapt to different imaging conditions, improving reconstruction quality while maintaining localization accuracy.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If multiple imaging techniques are used sequentially, then diagnostic accuracy is improved, but analysis time increases

Engineering Contradiction:
Improvediagnostic accuracyVSAvoidanalysis time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent merges the advantages of both standard mammography and tomosynthesis into a single integrated workflow. The system acquires multiple projections and reconstructs a 3D volume that provides both the localization capability of tomosynthesis and the diagnostic quality of standard mammography, eliminating the need for sequential imaging and reducing analysis time.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal imaging system that performs both 2D projection imaging and 3D volumetric imaging functions within a single acquisition sequence. The same set of projection images is used for both 2D display and 3D reconstruction, making the system multi-functional without requiring separate imaging procedures.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Manufacturing precision

If deep learning-based reconstruction is used, then image quality is improved, but computational complexity increases

Engineering Contradiction:
Improveimage reconstruction qualityVSAvoidcomputational complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary training of the neural network model using large datasets before actual image reconstruction. This preliminary action creates a pre-trained model that can be quickly applied to new images without requiring complex real-time optimization, reducing computational complexity during actual use while maintaining high reconstruction quality.

Inventive Principle:
Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach generates synthetic 2D images of comparable or superior diagnostic quality to standard mammography images, reducing patient radiation exposure, decreasing imaging and analysis time, and facilitating the adoption of full 3D protocols, while improving image reconstruction accuracy and reducing computational complexity.

Implementation Method 1

obtain a plurality of two-dimensional (2D) tomosynthesis projection images of an organ by rotating an x-ray emitter to a plurality of orientations relative to the organ and emitting a first level of x-ray energization from the emitter for each projection image

Methodology Applied
Scientific EffectX-ray emission: X-Ray

Data Source

PatentEP3675043B1Systems and methods for deep learning-based image reconstruction
Publication Date: 2024.01.24 GENERAL ELECTRIC CO
  • EP3675043B1 patent drawingFigure 1
  • EP3675043B1 patent drawingFigure 2
  • EP3675043B1 patent drawingFigure 3

AI summary

Methods, apparatus and systems for deep learning based image reconstruction are disclosed herein. An example at least one computer-readable storage medium includes instructions that, when executed, cause at least one processor to at least: obtain a plurality of two-dimensional (2D) tomosynthesis projection images of an organ by rotating an x-ray emitter to a plurality of orientations relative to the organ and emitting a first level of x-ray energization from the emitter for each projection image of the plurality of 2D tomosynthesis projection images; reconstruct a three-dimensional (3D) volume of the organ from the plurality of 2D tomosynthesis projection images; obtain an x-ray image of the organ with a second level of x-ray energization; generate a synthetic 2D image generation algorithm from the reconstructed 3D volume based on a similarity metric between the synthetic 2D image and the x-ray image; and deploy a model instantiating the synthetic 2D image generation algorithm.