Medical Image Embedding Vector Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for segmenting medical image data lack efficiency and accuracy in identifying medical features and abnormalities, particularly those with gradual transitions between normal and affected tissue.

Innovation Solution

A computer-implemented method using two machine learning-based systems to process medical image data. The first system generates image embedding vectors representing various medical image features, while the second system compares these vectors with a feature vector based on an indication of a specific medical feature, identifying the most similar vector and determining its location within the image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional segmentation methods are used to process medical images, then the processing can be performed with simpler systems, but the accuracy in identifying medical features with gradual transitions is insufficient

Engineering Contradiction:
Improveaccuracy of identifying medical featuresVSAvoidcomplexity of processing system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the medical image processing task into multiple stages: generating embedding vectors for different features, creating feature vectors from indications, and comparing them to identify matches. This segmentation of the processing pipeline enables higher accuracy in identifying medical features with gradual transitions while maintaining manageable system complexity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces embedding vectors as an intermediary representation between the original medical image data and the final segmentation results. These vectors serve as a bridge that captures gradual transitions in tissue characteristics, enabling more accurate identification of medical features without requiring overly complex direct processing methods.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If automated methods are implemented to identify medical features, then the productivity of medical image analysis is improved, but the difficulty of detecting features with gradual boundaries increases

Engineering Contradiction:
Improveefficiency of medical image analysisVSAvoiddifficulty of detecting features with gradual boundaries
Core Design Contradiction:
ProductivityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent transforms the medical image data into embedding vectors that represent features in a different parameter space. This parameter transformation converts the difficult task of detecting gradual boundaries in raw image data into a more manageable vector comparison task, where similarities can be measured using standard distance metrics, thereby maintaining high productivity while overcoming detection difficulties.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional mechanical or manual segmentation approaches with machine learning-based vector comparison methods. This substitution enables automated identification of medical features with gradual transitions, significantly improving productivity while the learning-based approach naturally handles the difficulty of gradual boundaries through pattern recognition in the embedding space.

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

3Measurement precision

If multiple machine learning systems are used to process medical images, then the accuracy of feature identification is improved, but the complexity of the processing system increases

Engineering Contradiction:
Improveaccuracy of feature identificationVSAvoidcomplexity of machine learning system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the machine learning processing into distinct functional components: an embedding generation system that creates feature representations, a feature vector generation system that processes indications, and a comparison system that identifies matches. This segmentation allows each component to be optimized independently while working together to achieve high accuracy, managing overall system complexity through clear functional separation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent designs the embedding vectors to serve multiple functions: they represent medical features for identification, capture gradual transitions in tissue characteristics, and enable comparison with feature vectors from various types of indications. This multi-functionality reduces the need for separate specialized systems, thereby improving accuracy without proportionally increasing system complexity.

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

Data Source

PatentEP4535283A1Medical image processing
Publication Date: 2025.04.09 SIEMENS HEALTHINEERS AG
  • EP4535283A1 patent drawingFigure 1
  • EP4535283A1 patent drawingFigure 2
  • EP4535283A1 patent drawingFigure 3

AI summary

Medical image data (202) representing a medical image is received at a first ML system. The first ML system (204) generates, based on the received medical image data (202), a plurality of image embedding vectors (206) corresponding to a respective plurality of medical image features, each of the plurality of image embedding vectors relating to a different respective medical image feature and comprising medical image feature data indicative of the presence or absence of the respective medical image feature at each of a plurality of locations in the medical image. An indication (208) of a first medical image feature included in the medical image is received at a second ML system (210). The second ML system generates a feature vector (212) based on the indication. A comparison of the feature vector with the plurality of image embedding vectors is performed and a first image embedding vector is identified based on the comparison.