Proposal of Semantic Annotation for German Metadata Using Bidirectional Recurrent Neural Networks

Area of Science:

• Digital Health
• Medical Informatics
• Machine Learning

Background:

• Digital healthcare's distributed nature and new data sources hinder comprehensive overviews and data utilization.
• Data integration, enhanced by metadata and semantic annotations, is crucial for addressing these challenges.

Purpose of the Study:

• To present a novel approach for predicting Unified Medical Language System (UMLS) codes from German metadata.
• To evaluate the effectiveness of recurrent neural networks and data augmentation in improving code prediction accuracy.

Main Methods:

• Development of a recurrent neural network model for predicting UMLS codes.
• Utilizing German metadata as input for the prediction task.
• Augmenting the training dataset with Medical Subject Headings (MeSH) and its German translations.

Main Results:

• The developed model achieved a robust performance with 75% accuracy in predicting UMLS codes.
• Augmenting the training data with German MeSH terms significantly improved model accuracy.

Conclusions:

• Sophisticated machine learning tools, like recurrent neural networks, can significantly contribute to healthcare data integration.
• The proposed approach demonstrates a viable method for standardizing and integrating distributed German healthcare data.