Medical Entity Knowledge Graph Denoising for Clinical Relationship Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The construction of healthcare knowledge graphs from electronic health records is challenging due to the lack of a definitive mapping between clinical medical entities, leading to inaccurate relationship quantification and high costs when manually built by medical professionals.
Innovation Solution
A computer-implemented method using a Natural Language Processing (NLP) model to extract medical entities, determine relationship strengths, generate uncommonality and relevance scores, and create a knowledge graph data structure representing pairs that satisfy a relevance threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual construction by medical professionals is used, then accuracy of knowledge graph relationships is improved, but cost and time consumption increase significantly
Solution Approach 1:
The patent uses NLP models to automatically extract and copy medical entity relationships from electronic health records, creating knowledge graphs without manual intervention. The system copies structured relationship data directly from unstructured clinical text through automated entity recognition and relationship extraction, eliminating the need for manual knowledge graph construction while maintaining accuracy through sophisticated language processing algorithms.
Solution Approach 2:
The patent replaces the mechanical manual process of medical professionals constructing knowledge graphs with an automated computational system. The NLP-based system substitutes human cognitive and manual labor with machine learning models that automatically identify entities, extract relationships, and construct knowledge graphs from clinical data, dramatically reducing time and resource requirements.
2Quantity of substance
If all medical entity pairs from electronic health records are included, then completeness of knowledge graph is improved, but noise and irrelevant relationships increase
Solution Approach 1:
The patent applies parameter changes by introducing scoring mechanisms that evaluate the strength and relevance of each extracted relationship. The system transforms raw relationship data into scored relationships, where only those meeting threshold criteria are included in the final knowledge graph. This parameter-based filtering converts the complete but noisy set of all possible relationships into a refined subset of high-confidence relationships.
Solution Approach 2:
The patent extracts only the relevant and high-confidence relationships from the complete set of medical entity pairs identified in electronic health records. The system takes out noise and irrelevant relationships through filtering mechanisms that retain only those relationships meeting predetermined relevance thresholds, producing a clean and reliable knowledge graph subset.
3Adaptability or versatility
If comprehensive relationship extraction is performed without filtering, then coverage of medical entities is improved, but data quality and signal-to-noise ratio deteriorate
Solution Approach 1:
The patent segments the relationship extraction process into distinct stages: initial comprehensive extraction of all potential relationships, followed by separate filtering and scoring stages. This segmentation allows the system to first capture maximum coverage of medical entity relationships and then systematically refine the results through multiple processing steps, maintaining both breadth and precision.
Solution Approach 2:
The patent implements feedback mechanisms where extracted relationships are evaluated, scored, and filtered based on their confidence levels and relevance metrics. The system uses feedback from the extraction process to adjust filtering thresholds and improve subsequent extractions, continuously refining the balance between coverage and precision through iterative optimization.
Data Source
AI summary
A method includes receiving, a plurality of records containing clinical information associated with one or more patients; extracting, using a Natural Language Processing (NLP) model, a plurality of medical entities from the clinical information to generate a first data set that contains the plurality of medical entities; denoising, the first dataset to generate a second data set by: determining relationship strengths between pairs of respective ones of the medical entities; identifying a subset of the pairs of the respective ones of the plurality of medical entities that satisfy a relationship strength threshold; generating uncommonality scores for one or both of a first and a second medical entity in each of the subset of pairs, the uncommonality score for the first medical entity being indicative of a frequency that the first medical entity occurs with the second medical entity across an entire set of instances of the second medical entity in the clinical information, the uncommonality score for the second medical entity being indicative of a frequency that the second medical entity occurs with the first medical entity across an entire set of instances of the first medical entity in the clinical information; and generating a relevance score for each of the subset of pairs based on one or both of the uncommonality scores for the first and second ones of the medical entities included in the respective pair and a frequency of occurrence of the respective pair in the clinical information; and generating a knowledge graph data structure representing ones of the subset of pairs having relevance scores, respectively, that satisfy a relevance threshold.