Unified Medical Entity Recognition via Bidirectional Attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical text processing technologies face challenges in extracting structured data from natural language texts, particularly in recognizing both continuous and discontinuous medical entities, which are essential for clinical decision systems but are often missed due to the limitations in existing recognition methods.
Innovation Solution
A method and apparatus that segment medical texts into words, determine local and global annotation features, and calculate probabilities using a maximum entropy model to identify labels and combination relationships between words, enabling the recognition of both continuous and discontinuous medical entities by considering the relationships and morphemes within the text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If pipeline-based methods are used for medical entity recognition, then the processing flow is simple and structured, but the accuracy of recognizing discontinuous medical entities deteriorates significantly
Solution Approach 1:
The patent merges the medical entity recognition task with the named entity recognition task into a unified framework. Both continuous and discontinuous medical entities are recognized simultaneously through a single model that processes the input text once, eliminating the need for separate pipeline stages and improving discontinuous entity recognition accuracy while maintaining processing efficiency.
Solution Approach 2:
The patent introduces a new dimension to the recognition framework by explicitly modeling discontinuous entity spans alongside continuous entities. The bidirectional attention mechanism operates across the entire sequence to identify entity boundaries and relationships, adding a dimensional layer that captures discontinuous patterns without complicating the overall processing flow.
2Measurement precision
If bidirectional attention mechanisms are introduced to improve entity recognition accuracy, then the accuracy of recognizing continuous and discontinuous entities is improved, but the computational complexity increases
Solution Approach 1:
The bidirectional attention mechanism serves multiple functions simultaneously: it captures long-range dependencies for discontinuous entities, identifies entity boundaries for continuous entities, and models relationships between entities. This multi-functionality justifies the computational overhead by delivering comprehensive entity recognition accuracy improvements across different entity types.
3Measurement precision
If a unified framework is used to recognize both continuous and discontinuous entities, then the recognition accuracy for both types is improved, but the model complexity increases compared to separate processing methods
Solution Approach 1:
The patent combines continuous and discontinuous medical entity recognition into a single unified model that processes the input text once. The model uses a shared architecture with bidirectional attention mechanisms to handle both entity types simultaneously, improving overall recognition accuracy while avoiding the complexity of maintaining separate processing pipelines for different entity types.
Data Source
AI summary
Embodiments of the present disclosure provide a method and an apparatus for recognizing a medical entity in a medical text. In this method, the medical text is segmented into a plurality of words. Each of the words is used as a target word. A local and a global annotation feature of the target word are determined. The local annotation feature includes the target word. The global annotation feature includes a relationship between the target word and the recognized medical entity. Next, a label for the target word is determined from candidate labels based on the local and global annotation feature. Next, a combination relationship between the target word and a previous word of the target word is obtained based on the label for the target word. The combination relationship includes combination and non-combination. Next, the combined words are recognized as the medical entity based on the combination relationship.