Named Entity Recognition in Chinese Clinical Text Using Deep Neural Network

Area of Science:

• Computational linguistics
• Medical informatics
• Artificial intelligence

Background:

• Electronic health records (EHRs) contain vast amounts of unstructured clinical narrative data.
• Extracting critical information from these narratives is challenging but essential for healthcare.
• Natural Language Processing (NLP), specifically Named Entity Recognition (NER), is key to unlocking this data.

Purpose of the Study:

• To develop and evaluate a novel deep learning method for clinical entity recognition in Chinese EHRs.
• To investigate the impact of unsupervised learning for word embeddings on NER performance.
• To achieve high accuracy in identifying clinical entities within free-text medical documents.

Main Methods:

• A deep neural network (DNN) was developed for unsupervised learning of word embeddings from a large unlabeled corpus.
• A separate DNN was employed for the Named Entity Recognition (NER) task.
• Minimal feature engineering was utilized to focus on the effectiveness of learned embeddings.

Main Results:

• The proposed DNN model with unsupervised word embeddings achieved a superior F1-score of 0.9280.
• This performance surpassed the state-of-the-art Conditional Random Fields (CRF) model in a minimal feature setting.
• Unsupervised learning of word embeddings significantly enhanced DNN performance compared to randomized embeddings.

Conclusions:

• Deep learning models incorporating unsupervised word embeddings are highly effective for clinical NER in Chinese EHRs.
• Unsupervised feature learning from large unlabeled corpora is a valuable strategy for improving NLP tasks in medicine.
• This approach offers a robust method for extracting crucial clinical information from narrative text.