ML Algorithm Update via NLP Medical Report Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The manual annotation of medical data sets for training machine learning algorithms in healthcare is tedious and prone to errors, leading to inefficient training and potential false negatives or false positives in medical image analysis, which can result in missed treatments or unnecessary psychological distress.

Innovation Solution

A method that utilizes natural language processing algorithms to extract indications of medical findings from medical reports, combined with machine learning algorithms to update and refine the detection of medical findings in medical images, reducing the need for manual annotations and enhancing the efficiency of training data generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation of medical data sets is performed to train machine learning algorithms, then training data quality is improved, but time consumption and error rate increase

Engineering Contradiction:
Improvetraining data qualityVSAvoidannotation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system enables automatic self-annotation by training machine learning algorithms to extract medical findings from unstructured text reports and automatically annotate corresponding medical images, eliminating the need for manual expert annotation while maintaining high training data quality

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Manual annotation processes are replaced with automated natural language processing and machine learning-based extraction systems that can process medical texts and images automatically, substituting human labor with computational methods

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

2Measurement precision

If manual annotation of medical data sets is performed to train machine learning algorithms, then training data quality is improved, but error rate decreases

Engineering Contradiction:
Improvetraining data qualityVSAvoiderror rate
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The system implements feedback mechanisms where machine learning algorithms automatically extract findings from text reports, compare them with existing annotations, and use discrepancies to refine their extraction accuracy, continuously improving reliability through iterative learning

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The automated extraction system serves itself by using extracted findings to train and retrain the underlying algorithms, creating a self-improving system that reduces errors over time without requiring manual correction

Inventive Principle:
Principle #25Self-service

3Productivity

If automated detection systems are implemented to analyze medical images, then productivity is improved, but system complexity increases

Engineering Contradiction:
Improvedetection efficiencyVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system integrates multiple functions into a unified automated detection platform that can process different types of medical images, extract findings from various text report formats, and train multiple machine learning algorithms simultaneously, reducing operational complexity despite increased capabilities

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

4Productivity

If automated detection systems are implemented to analyze medical images, then productivity is improved, but manual annotation requirements decrease

Engineering Contradiction:
Improvedetection efficiencyVSAvoidautomation level
Core Design Contradiction:
ProductivityVSExtent of automation

Solution Approach 1:

The system achieves high-level automation by enabling machine learning algorithms to automatically extract medical findings from unstructured text reports and use these extractions to self-train and improve detection capabilities, minimizing the need for any manual annotation intervention

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4390960A1Systems and methods for providing an updated machine learning algorithm
Publication Date: 2024.06.26 SIEMENS HEALTHINEERS AG
  • EP4390960A1 patent drawingFigure 1
  • EP4390960A1 patent drawingFigure 2
  • EP4390960A1 patent drawingFigure 3

AI summary

Systems and methods for providing updated machine learning algorithms are provided. An exemplary system comprises an interface configured to: obtain, from a model aggregator device, a machine learned algorithm, the machine learned algorithm being configured to extract medical findings from medical data sets; obtain, from the model aggregator device, a natural language processing algorithm, the natural language processing algorithm being configured to extract indications of medical findings from text data; and obtain, form one or more local databases, medical data sets and associated medical reports. Further the exemplary system comprise a computing unit with: - a data parsing module configured to query the one or more local databases for a set of medical data sets and a set of associated medical reports, the medical data sets each comprising at least one medical finding; - a reference data generation module configured to generate indications of at least a part of the medical findings by applying the natural language processing algorithm to the medical reports; - a data gating module configured to determine whether to allow a local data item comprising one of the medical data sets and a corresponding indication to be incorporated into a training process of the machine learning algorithm; and - a training module configured to update the machine learned algorithm.