Structuring electronic dental records through deep learning for a clinical decision support system.

Area of Science:

• Medical Informatics
• Natural Language Processing (NLP)
• Computational Linguistics

Background:

• Extracting information from unstructured clinical text is a significant challenge in medical informatics.
• Electronic dental records (EDRs) contain valuable data that needs efficient structuring for clinical use.

Purpose of the Study:

• To construct a novel NLP workflow for extracting structured information from Chinese EDRs.
• To enhance clinical decision support systems (CDSSs) by providing accurate, structured data from EDRs.

Main Methods:

• Developed a hybrid NLP workflow integrating deep learning (Sentence2vec, Word2vec) with keyword-based methods.
• Utilized unsupervised learning for text vector representation.
• Extracted attributes, attribute values, and tooth positions based on an existing ontology.

Main Results:

• The hybrid method demonstrated superior performance compared to keyword-based and deep learning-only methods.
• Achieved high precision (0.94) and recall (0.74) for attribute recognition.
• Achieved high precision (0.94) and recall (0.82) for attribute value recognition, with F scores of 0.83 and 0.88, respectively.

Conclusions:

• The proposed NLP workflow effectively structures narrative text from EDRs.
• Provides a robust foundation for data-driven CDSSs by delivering accurate input information.
• Highlights the efficacy of hybrid deep learning and keyword approaches in clinical text analysis.