Medical Report Summarization with Adaptive Entity Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically summarizing electronic medical records (EMRs) struggle to accurately identify rare but informative medical observations amidst a large volume of less informative data, leading to potential omission of crucial information during diagnosis and treatment of chronic diseases.
Innovation Solution
A system that classifies medical reports into predetermined categories and uses a library of entity recognition models trained with adaptive loss strategies to prioritize and accurately identify named entities, accounting for class imbalances and frequency imbalances in the training dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning systems are used to automatically extract information from EMRs, then extraction speed is improved, but detection accuracy of rare but informative medical observations deteriorates when co-occurring with frequent but less informative observations
Solution Approach 1:
The patent segments the EMR processing task by dividing the document into multiple sections (e.g., problem, examination, operation, pathology) and applies different entity recognition models to each section. This segmentation allows the system to handle dense annotations more effectively by focusing on relevant sections, thereby improving both processing efficiency and detection accuracy of rare observations without being overwhelmed by frequent but less informative observations in other sections.
Solution Approach 2:
The patent applies local quality by using different entity recognition models for different sections of the EMR. Each section is assigned a specialized model trained on section-specific characteristics, allowing the system to optimize detection accuracy for rare observations in each local context while maintaining overall processing efficiency. This localized approach prevents the model from being biased toward frequent observations that may dominate in certain sections.
2Device complexity
If a single entity recognition model is used for all medical report categories, then device complexity is reduced, but detection accuracy deteriorates due to category-specific variations in observation patterns
Solution Approach 1:
The patent segments the entity recognition task by creating multiple specialized models for different medical report categories (e.g., radiology, pathology, surgery). Each model is trained on category-specific data and patterns, allowing the system to achieve high detection accuracy for rare observations in each category while maintaining a manageable model library through systematic organization and selection mechanisms.
Solution Approach 2:
The patent introduces a category classification intermediary that first identifies the type of medical report and then routes it to the appropriate specialized entity recognition model. This intermediary layer manages the complexity of having multiple models by providing a simple interface for model selection, thereby maintaining detection accuracy without proportionally increasing system complexity.
3Productivity
If machine learning systems prioritize common classes of medical observations, then identification efficiency of frequent observations is improved, but detection of rare but informative observations deteriorates
Solution Approach 1:
The patent applies local quality by training entity recognition models with section-specific loss functions that adjust the importance weighting of different observation types based on the section context. This allows the model to efficiently identify frequent observations while simultaneously maintaining high detection reliability for rare but informative observations within each specific section, rather than applying a uniform prioritization across the entire EMR.
Data Source
AI summary
Various methods and systems are provided for generating and displaying summaries of patient information extracted from one or more medical reports stored in an electronic medical record (EMR) of a patient. In one embodiment, a method for summarizing medical reports includes, receiving a medical report for a patient, classifying the medical report into a category of a plurality of pre-determined categories, matching the medical report with an entity recognition model from a library of entity recognition models based on the category, identifying a plurality of named entities in the medical report using the entity recognition model, refining the plurality of named entities to produce a summary of the medical report, and displaying the summary of the medical report via a display device.