Myocardial Segmentation With Multi-Task Learning for Cardiac MRI Diagnosis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automated methods for myocardial segmentation and disease detection in cardiac MRI images are inaccurate, time-consuming, and prone to human error, particularly due to variability in heart anatomy and imaging quality, leading to challenges in differentiating between healthy and diseased tissues.

Innovation Solution

A multi-task deep learning model with an encoder-decoder structure and classification network is used to simultaneously perform segmentation and disease detection, leveraging shared representations and iterative weight updates to enhance accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual interpretation methods are used for myocardial segmentation and disease detection, then diagnostic accuracy can be maintained through expert judgment, but the analysis process becomes time-consuming and prone to human error

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

Solution Approach 1:

The patent replaces the manual mechanical interpretation process with an automated deep learning system. The U-Net architecture with attention gates automatically segments myocardial tissues and detects diseases without human intervention, eliminating time-consuming manual analysis while maintaining diagnostic accuracy through trained neural networks

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

2Productivity

If current automated disease detection techniques are implemented, then analysis speed is improved, but accuracy deteriorates due to inability to differentiate between healthy and diseased tissues

Engineering Contradiction:
Improveanalysis speedVSAvoiddisease detection accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent applies local quality enhancement through attention gates that selectively focus on specific regions of the myocardium. The attention mechanism assigns different weights to different spatial locations, allowing the model to pay special attention to areas with pathological changes while ignoring normal tissues, thereby improving detection accuracy

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent introduces T1 and T2 mapping images as intermediary representations that capture tissue characteristics. These quantitative maps serve as intermediaries between the raw MRI images and disease detection, providing enhanced contrast between healthy and diseased tissues that improves automated detection accuracy

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If separate models are used for segmentation and disease detection, then each task can be optimized independently, but the overall system complexity increases and interrelated aspects are not leveraged

Engineering Contradiction:
Improvetask optimizationVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges segmentation and disease detection into a single integrated U-Net architecture. The model performs both tasks simultaneously within one network, sharing computational resources and features. This reduces system complexity compared to using separate models while maintaining the ability to optimize both tasks through joint training

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a multi-functional model where the U-Net architecture serves dual purposes: segmenting myocardial tissues and detecting cardiac diseases. The same network structure and learned features are used for both segmentation mask generation and disease classification, making the system universal and reducing overall complexity

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

Data Source

PatentEP4708135A1Multi-task learning-based myocardial segmentation and disease detection in cardiac mr tissue mapping images
Publication Date: 2026.03.11 SIEMENS HEALTHINEERS AG
  • EP4708135A1 patent drawingFigure 1
  • EP4708135A1 patent drawingFigure 2
  • EP4708135A1 patent drawingFigure 3

AI summary

Systems and methods for myocardial segmentation and disease detection using a multi-task deep learning model (300). The multi-task deep learning model (300) simultaneously performs segmentation and disease detection using the interrelated aspects to improve both tasks. The multi-task deep learning model (300) includes an encoder-decoder structure (310) where a compressed representation (316) extracted by an encoder of the encoder-decoder structure (310) is used for both reconstructing a segmentation mask (350) in the decoder and as an input for disease detection.