2D Medical Image Analysis Using Multimodal External-Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for analyzing 2D medical images with extracorporeal devices face challenges in differentiating between external objects and internal tissue structures, requiring large training datasets or complex annotations, and are prone to concept drift when using synthetic data.

Innovation Solution

A method utilizing additional image data from a different modality, combined with artificial intelligence, to identify and localize external objects, allowing for a sequential analysis process that reduces training data requirements and improves image interpretation by separating and adapting analysis based on the additional object information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning is used for detecting and identifying equipment with large training datasets, then the system can learn to differentiate between extracorporeal objects and inserted devices, but the requirement for large training datasets increases annotation complexity and data processing requirements

Engineering Contradiction:
Improvedifferentiation accuracyVSAvoidtraining data volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent segments the analysis process into two distinct phases: first analyzing additional image data (such as optical or infrared images) to identify extracorporeal objects, then using this information to adapt the analysis of 2D medical image data. This segmentation allows the system to handle different types of objects with different analysis methods, reducing the need for large unified training datasets while maintaining high differentiation accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary analysis of additional image data before analyzing the 2D medical image data. By first identifying extracorporeal objects in the additional image data and determining their positions, the system can then adaptively adjust the analysis of the medical image data, focusing computational resources on relevant regions and reducing the overall data processing requirements while improving accuracy.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If segmentation of external and inserted objects is performed to improve image analytics, then object identification improves, but annotation complexity increases significantly

Engineering Contradiction:
Improveobject identification accuracyVSAvoidannotation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the object identification task into two segments: identifying extracorporeal objects in additional image data (which has lower annotation complexity) and then using this information to guide identification in 2D medical image data. This segmentation reduces the overall annotation complexity by handling different object types in separate analysis passes rather than requiring comprehensive annotation of all objects in all image types simultaneously.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent uses additional image data as an intermediary to facilitate object identification. By first analyzing additional image data to identify extracorporeal objects and their positions, this intermediate information serves as a guide for the subsequent analysis of 2D medical image data, reducing the direct annotation burden on the medical images themselves while maintaining high identification accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Quantity of substance

If synthetic cases are created to avoid large training datasets, then data requirements are reduced, but concept drift causes mismatch between training and test conditions

Engineering Contradiction:
Improvetraining data volumeVSAvoidtraining-test consistency
Core Design Contradiction:
Quantity of substanceVSReliability

Solution Approach 1:

The patent introduces an additional dimension by using multiple types of image data (additional image data and 2D medical image data) instead of relying solely on synthetic cases within a single modality. By analyzing additional image data (such as optical or infrared images) that capture extracorporeal objects from a different perspective or modality, the system reduces dependence on synthetic training cases while maintaining reliability through multi-modal validation.

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

Data Source

PatentEP4156090B1Automatic analysis of 2d medical image data with an additional object
Publication Date: 2025.07.16 SIEMENS HEALTHINEERS AG
  • EP4156090B1 patent drawingFigure 1~2
  • EP4156090B1 patent drawingFigure 3~4
  • EP4156090B1 patent drawingFigure 5~6

AI summary

A method for automatically analysing 2D medical image data (MID), comprising an additional object (EO), is described. According to the method, medical image data (MID), including the additional object (EO), are acquired from an examination portion (ROI) of a patient (P) by a first modality (2) and additional image data (AID) are acquired from the examination portion (ROI) using a different modality (C). Based on the acquired medical image data (MID) and the acquired additional image data (EID) an automatic image analysis is performed, which is adapted to the additional object (EO). Moreover, an analysis device (40, 50) is described. Further, a medical imaging system (1) is described.