Fast Model Adaptation for Automated Section Classification in Electronic Medical Records

Area of Science:

• Medical Informatics
• Machine Learning
• Natural Language Processing

Background:

• Medical information extraction automates data retrieval from electronic medical records for improved healthcare.
• Section classification is a key task, identifying and categorizing sections within medical documents.
• High accuracy in section classification models requires extensive human-labeled data, which is costly and time-consuming to acquire.

Purpose of the Study:

• To reduce the annotation cost and time for training section classification models.
• To enable faster adaptation of section classification models to new medical institutions.
• To explore the effectiveness of active learning and distant supervision for automated medical information extraction.

Main Methods:

• Applied active learning to intelligently select data for human labeling, minimizing annotation effort.
• Utilized distant supervision to leverage existing, weakly labeled data for model training.
• Evaluated the performance of these techniques on the section classification task in electronic medical records.

Main Results:

• Active learning reduced annotation cost and time by over 50%.
• Distant supervision achieved good model accuracy using only weakly labeled data.
• Demonstrated the feasibility of adapting section classification models efficiently to new institutions.

Conclusions:

• Active learning and distant supervision are effective strategies for reducing annotation burdens in medical information extraction.
• These methods facilitate rapid and cost-effective deployment of section classification models in diverse healthcare settings.
• The study highlights a path towards more scalable and accessible automated medical record analysis.