Deep Learning Network for Cardiac Phase Determination in MRI

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for determining cardiac phases in magnet resonance imaging, especially for retrospective cardiac binning, rely on external ECG signals or complex post-processing techniques like PCA/ICA, which can be cumbersome and less effective, especially in cases with arrhythmia or noisy data.

Innovation Solution

A deep learning network is trained to identify cardiac phases from raw MRI data, eliminating the need for external ECG signals and hand-crafted feature selection, by processing 1D Superior-Inferior projections to predict cardiac phase probabilities or labels, allowing for automatic retrospective binning of cardiac phases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If external ECG signals are used for cardiac phase determination, then measurement precision is improved, but device complexity increases

Engineering Contradiction:
Improvecardiac phase determination accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts the cardiac phase determination function from external ECG devices and implements it within the MRI system using deep learning. The neural network processes MRI data directly to determine cardiac phases, eliminating the need for separate ECG signal acquisition and processing hardware, thus reducing device complexity while maintaining measurement precision

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces a deep learning neural network as an intermediary between raw MRI data and cardiac phase determination. This intermediary processes the data automatically, replacing the need for external ECG signals and manual feature extraction, thereby simplifying the overall system architecture while preserving accuracy

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of operation

If PCA/ICA methods are used for cardiac phase determination, then ease of operation is improved, but measurement precision deteriorates

Engineering Contradiction:
Improveprocessing simplicityVSAvoidcardiac phase determination accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent replaces traditional signal processing methods (PCA/ICA) with a deep learning neural network. This substitution enables the system to automatically learn complex patterns in MRI data without requiring manual feature engineering or mathematical transformations, thereby maintaining ease of operation while significantly improving measurement precision through adaptive pattern recognition

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

Solution Approach 2:

The patent changes the fundamental approach from linear algebra-based methods (PCA/ICA) to non-linear deep learning transformations. The neural network automatically adjusts its internal parameters during training to optimize cardiac phase determination, providing both operational simplicity and high precision without the limitations of fixed mathematical models

Inventive Principle:
Principle #35Parameter changes

3Device complexity

If manual feature selection is used, then device complexity is reduced, but productivity deteriorates

Engineering Contradiction:
Improveprocessing complexityVSAvoidprocessing speed
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent performs feature extraction and selection automatically during the training phase of the neural network. The network learns to identify relevant features from training data, so that during actual cardiac phase determination, no manual feature selection is needed. This preliminary automated feature learning maintains low operational complexity while dramatically improving processing speed through efficient inference

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The neural network is self-sufficient in identifying and extracting relevant features from MRI data without requiring manual intervention. The system automatically adapts to different datasets and conditions, performing feature selection as an inherent part of its operation, thereby eliminating the trade-off between complexity and productivity

Inventive Principle:
Principle #25Self-service

4Productivity

If deep learning network is used for cardiac phase determination, then productivity is improved, but device complexity increases

Engineering Contradiction:
Improveprocessing speedVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent extracts the complex deep learning computations from the real-time processing pipeline and performs them offline during a training phase. The trained network then operates efficiently during actual cardiac phase determination, providing high productivity with minimal real-time computational overhead and reduced perceived system complexity

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The neural network is trained in advance on large datasets to learn optimal feature representations and decision boundaries. This preliminary training action transfers complex computational requirements to the offline phase, enabling fast and accurate cardiac phase determination during actual use with reduced real-time computational demands

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3878361B1Method and device for determining a cardiac phase in magnet resonance imaging
Publication Date: 2024.04.24 SIEMENS HEALTHINEERS AG
  • EP3878361B1 patent drawingFigure 1~2
  • EP3878361B1 patent drawingFigure 3
  • EP3878361B1 patent drawingFigure 4

AI summary

The invention describes a trained deep learning network (22) for determining a cardiac phase in magnet resonance imaging, comprising an input layer (E), an output layer (S) and a number of hidden layers (CR, MP) between input layer (E) and output layer (S), the layers processing input data (ID) entered into the input layer (E), wherein the deep learning network (22) is designed and trained to output a probability or some other label of a certain cardiac phase at a certain time from entered input data (ID). The invention further describes a method determining a cardiac phase in magnet resonance imaging, a related device, a training method for the deep learning network, a control device and a related magnetic resonance imaging system.