Substituting clinical features using synthetic medical phrases: Medical text data augmentation techniques.

Area of Science:

• Biomedical informatics
• Natural Language Processing

Background:

• Biomedical Natural Language Processing (NLP) is crucial for extracting information from clinical notes.
• Classifying unstructured medical documents, particularly short texts like abstracts, presents challenges due to domain-specific language and data scarcity.
• Existing methods struggle with accuracy in medical document classification.

Purpose of the Study:

• To address the challenges in clinical note classification.
• To propose and evaluate novel data augmentation techniques for medical documents.
• To improve the accuracy of deep learning models in classifying clinical notes.

Main Methods:

• Developed two data augmentation approaches: an ontology-guided method and a combined ontology- and dictionary-based method.
• Utilized three distinct deep learning models to assess the effectiveness of the proposed augmentation techniques.
• Applied methods to enrich training data for medical document classification tasks.

Main Results:

• The proposed data augmentation methods significantly improved classification accuracy for clinical notes.
• Enhanced performance was observed particularly in scenarios with limited available clinical data.
• Both the ontology-guided and combined approaches demonstrated effectiveness in enriching training datasets.

Conclusions:

• Novel data augmentation strategies can effectively enhance the performance of biomedical NLP tasks.
• These methods offer a viable solution to improve clinical note classification accuracy, especially when data is scarce.
• The developed approaches show promise for advancing the analysis of unstructured medical text.