Medical Image Metadata Generation via Landmark Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for detecting and classifying medical images rely heavily on metadata, which is often missing, incorrect, or not standardized across different formats and institutions, limiting their effectiveness, especially for non-DICOM images.

Innovation Solution

The use of machine learning models, such as deep reinforcement learning and random forests, to automatically identify body landmarks in medical images and generate metadata tags directly from the image data, rather than relying on existing metadata, ensuring accurate and consistent tagging across various image formats.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models are used to automatically identify body landmarks and generate metadata tags from image data, then accuracy and consistency of metadata tags are improved, but device complexity and computational resources increase

Engineering Contradiction:
Improveaccuracy of metadata tagsVSAvoidcomplexity of machine learning system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary action by pre-training machine learning models on large datasets of medical images with annotated body landmarks before deployment. This pre-processing of training data and model preparation enables the system to automatically generate accurate metadata tags when processing actual medical images, resolving the contradiction between accuracy and complexity by establishing the computational foundation in advance

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system introduces an intermediary machine learning model that acts as a mediator between raw medical image data and metadata tag generation. This intermediary component processes the image data to identify body landmarks and infer anatomical objects, thereby improving tag accuracy while managing system complexity through modular architecture

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If machine learning models are applied to identify anatomical objects from body landmarks, then reliability of anatomical object identification is improved, but processing time increases

Engineering Contradiction:
Improvereliability of anatomical object identificationVSAvoidprocessing time for image analysis
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system applies segmentation by dividing the image analysis process into distinct stages: first identifying body landmarks, then using those landmarks to identify anatomical objects, and finally generating metadata tags. This segmented approach improves reliability by breaking down the complex task into manageable steps while optimizing processing time through efficient resource allocation at each stage

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary identification of body landmarks before proceeding to anatomical object identification. This preliminary action provides a reliable foundation for subsequent analysis, improving overall reliability while managing processing time by organizing the workflow in a logical sequence where each step builds on previous results

Inventive Principle:
Principle #10Preliminary action

3Productivity

If the system processes medical images to generate metadata tags immediately following image reconstruction, then productivity of metadata generation is improved, but use of computational resources increases

Engineering Contradiction:
Improvespeed of metadata tag generationVSAvoidcomputational resource consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The system merges the metadata tag generation process with the existing medical image reconstruction workflow by integrating the machine learning model into the post-processing pipeline. This combining of functions improves productivity by eliminating separate processing steps while managing computational resources through shared infrastructure and optimized resource allocation during the reconstruction phase

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3444824B1Detecting and classifying medical images based on continuously-learning whole body landmarks detections
Publication Date: 2023.05.03 SIEMENS HEALTHINEERS AG
  • EP3444824B1 patent drawingFigure 1A
  • EP3444824B1 patent drawingFigure 1B
  • EP3444824B1 patent drawingFigure 2

AI summary

A computer-implemented method for automatically generating metadata tags for a medical image includes receiving a medical image and automatically identifying a set of body landmarks in the medical image using one or more machine learning models. A set of rules are applied to the set of body landmarks to identify anatomical objects present in the image. As an alternative to using the set of rules, in some embodiments, one or more machine learning models to the set of body landmarks to identify anatomical objects present in the image. Once the anatomical objects are identified, metadata tags corresponding to the anatomical objects are generated and stored in the medical image. Then, the medical image with the metadata tags is transferred to a data repository.