Medical Imaging Workflow Support via LLM Inference Graphs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Clinical information is fragmented and often unstructured, leading to a loss of vital data, increased workload, and delayed or disrupted patient care due to the complexity of medical imaging workflows, which are hindered by data silos, vendor-specific solutions, and the ambiguity of natural language in medical documentation, resulting in inefficient workflow automation and increased risk of errors.

Innovation Solution

A computer-implemented method using a large language model to transform unstructured clinical information into structured format through an inference graph, enabling rule-driven derivation of medical imaging workflows, allowing clinicians and technologists to adjust protocols via natural language interactions, and providing intelligent support through an AI-based assistance function.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If manual processing of unstructured clinical reports is used, then data completeness is maintained through human interpretation, but workload increases and processing time is extended

Engineering Contradiction:
Improvedata completenessVSAvoidworkflow efficiency
Core Design Contradiction:
Loss of informationVSProductivity

Solution Approach 1:

The patent replaces manual mechanical processing of clinical reports with an AI-based natural language processing system. The AI assistant automatically extracts structured information from unstructured reports, transforming the mechanical task of data extraction from human clinicians to an automated computational system, thereby maintaining data completeness while significantly improving workflow efficiency

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

Solution Approach 2:

The AI assistant enables self-service by automatically performing data extraction and structuring tasks without requiring manual intervention. The system autonomously processes clinical reports, generates structured data, and integrates it into the imaging workflow, allowing the system to serve itself rather than requiring continuous human operational input

Inventive Principle:
Principle #25Self-service

2Stability of the object's composition

If standardized vocabularies and protocols are implemented across diverse clinical IT systems, then data consistency is improved, but implementation difficulty increases due to data silos and vendor-specific solutions

Engineering Contradiction:
Improvedata consistencyVSAvoidimplementation complexity
Core Design Contradiction:
Stability of the object's compositionVSDevice complexity

Solution Approach 1:

The AI assistant acts as an intermediary layer between diverse clinical IT systems and standardized vocabularies. Instead of directly integrating multiple vendor-specific systems, the AI processes unstructured data from various sources and transforms it into standardized structured formats, mediating the complexity of inter-system integration while ensuring data consistency

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the complex task of cross-system data integration into smaller, manageable components. The AI assistant handles specific extraction tasks for different data types (clinical indications, patient history, findings) separately, processing each segment independently and then integrating them into a unified structured output, thereby reducing overall implementation complexity

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If scan protocols are manually configured and adapted for each clinic, then protocol specificity to local requirements is achieved, but time and expertise requirements increase significantly

Engineering Contradiction:
Improveprotocol adaptabilityVSAvoidconfiguration time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The AI assistant performs preliminary action by automatically generating draft scan protocols based on extracted clinical information before human review. The system pre-configures protocol parameters, selects appropriate imaging sequences, and prepares protocol recommendations in advance, allowing clinicians to review and adjust rather than create protocols from scratch, thereby reducing configuration time while maintaining adaptability

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback mechanisms where AI-generated protocol recommendations are reviewed and adjusted by clinicians, and this feedback is used to refine future protocol generation. The system learns from manual adjustments and feedback to improve its protocol adaptation accuracy over time, maintaining high adaptability while reducing the time required for configuration

Inventive Principle:
Principle #23Feedback

4Measurement precision

If advanced image acquisition techniques are used, then diagnostic capability is enhanced, but cognitive burden on clinicians increases and focus on patient care is reduced

Engineering Contradiction:
Improvediagnostic capabilityVSAvoidcognitive burden
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The AI assistant extracts and processes complex information from clinical reports and imaging data, separating the cognitive burden of data interpretation from clinicians. The system automatically extracts relevant clinical indications, patient history, and imaging findings, presenting only the essential information and recommendations to clinicians, thereby enhancing diagnostic capability through advanced techniques while reducing cognitive burden

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4668285A1Method for providing medical imaging workflow support data
Publication Date: 2025.12.24 SIEMENS HEALTHINEERS AG
  • EP4668285A1 patent drawingFigure 1~3
  • EP4668285A1 patent drawingFigure 4~5
  • EP4668285A1 patent drawing

AI summary

In one aspect the invention relates to a computer-implemented method for providing medical imaging workflow support data, the method comprising: - receiving (S1) natural language data (U), the natural language data (U) comprising clinical information, - generating (S2) structured information (N) by applying a large language model (L) to the natural language data (U), the structured information (N) comprising the clinical information in a structured format of at least one input node (31, 32) of an inference graph (G), - calculating (S3) the medical imaging workflow support data by applying at least one rule of the inference graph (G) to the structured information (N), and - providing (S4) the medical imaging workflow support data.