Automatic ICD code assignment of Chinese clinical notes based on multilayer attention BiRNN | JoVE Visualize

Area of Science:

Natural Language Processing
Medical Informatics
Machine Learning

Background:

Electronic health records rely on International Classification of Diseases (ICD) codes for data organization and analysis.
Automatic ICD code assignment from clinical notes is crucial but challenging, especially for languages like Chinese due to segmentation and character representation issues.
Existing methods struggle with the length and complexity of clinical narratives.

Purpose of the Study:

To develop and evaluate a novel deep learning model for accurate automatic ICD code assignment from Chinese clinical notes.
To address challenges in Chinese word segmentation and character representation within clinical text.
To improve the performance of ICD coding in electronic health records using advanced neural network architectures.

Main Methods:

A multilayer attention bidirectional recurrent neural network (MA-BiRNN) model was proposed.
A hierarchical approach was used for feature representation of discharge summaries, combining character and word level embeddings.
Attention mechanisms were integrated into bidirectional long short-term memory networks to manage long text sequences.

Main Results:

The MA-BiRNN model achieved F1-scores of 0.639 (full-level) and 0.766 (block-level) on a real-world dataset of 7732 Chinese admission records and 1177 ICD-10 labels.
The proposed model outperformed baseline neural network models and demonstrated the lowest Hamming loss.
Ablation studies confirmed the critical role of the multilevel attention mechanism in processing Chinese clinical notes.

Conclusions:

The MA-BiRNN model offers a robust and effective solution for automatic ICD code assignment in Chinese clinical documents.
The combination of character/word embeddings and multilevel attention significantly enhances performance on long, complex clinical narratives.
This approach has the potential to improve the efficiency and accuracy of medical coding in healthcare systems utilizing Chinese electronic health records.