Related Experiment Video
Updated: Oct 13, 2025
07:50
A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts
Published on: September 20, 2018
16.1K
Automatic RadLex coding of Chinese structured radiology reports based on text similarity ensemble.
Yani Chen1,2, Shan Nan3, Qi Tian1,2
1College of Biomedical Engineering and Instrument Science, Zhejiang University, Zheda Road, Hangzhou, 310027, China.
BMC Medical Informatics and Decision Making
|November 18, 2021
Summary
This study introduces a novel hybrid translation and ensemble algorithm for automatic RadLex coding of Chinese radiology reports. The approach significantly improves accuracy for cross-language standardization, achieving a 90.15% F1-score.
Area of Science:
- Radiology Informatics
- Natural Language Processing
- Medical Terminology Standardization
Background:
- Standardized coding of radiology reports is crucial for secondary uses like data analytics and personalized medicine.
- RadLex lexicon improves clarity and reduces variability in radiology reports.
- Establishing RadLex coding and localization in China is underdeveloped, hindering cross-language applications.
Purpose of the Study:
- To develop an effective automatic cross-language RadLex coding approach for Chinese structured radiology reports.
- To enhance the accuracy of RadLex coding by addressing limitations in current translation and text similarity algorithms.
- To facilitate the standardization of radiology reports for secondary data utilization in China.
Main Methods:
- A hybrid translation strategy combining Google neural machine translation (GNMT) and dictionary-based translation for Chinese to English radiology phrases.
- Implementation of four text similarity algorithms: Levenshtein distance, Jaccard similarity, Word2vec CBOW, and WordNet Wup.
- A Multilayer Perceptron (MLP) model to ensemble the outputs of the similarity algorithms, synthesizing contextual, lexical, character, and syntactical information.
Main Results:
- The proposed approach achieved a high F1-score of 90.15%, with 91.78% precision and 88.59% recall.
- The hybrid translation method improved F1-score by 21.44% over GNMT and 8.12% over dictionary translation alone.
- The MLP weighting ensemble algorithm outperformed the best single similarity algorithm (WordNet Wup) by 4.48%.
Conclusions:
- The study presents an innovative automatic cross-language RadLex coding approach for Chinese structured radiology reports.
- The developed method effectively addresses the challenge of standardizing radiology reports across languages.
- This approach can serve as a valuable reference for future research in automatic cross-language coding.
