A New Public Corpus for Clinical Section Identification: MedSecId.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Documentation Analysis

Background:

• Section identification in documents aids readers in information retrieval and topic contextualization.
• Clinical medical documentation presents unique challenges for automated section segmentation.
• Effective section identification is crucial for organizing and analyzing large volumes of medical text.

Purpose of the Study:

• To develop and evaluate methods for segmenting sections within clinical medical documentation.
• To introduce MedSecId, a novel, publicly available dataset for medical section identification.
• To analyze relationships between medical concepts across different document sections.

Main Methods:

• Creation of MedSecId, a dataset comprising 2,002 fully annotated medical notes from MIMIC-III.
• Implementation and evaluation of several baseline section identification models.
• Utilizing principal component analysis (PCA) to explore medical concept relationships across sections.

Main Results:

• The study presents MedSecId, a valuable resource for advancing research in clinical document analysis.
• Baseline models demonstrate varying performance in medical section identification tasks.
• Analysis reveals distinct patterns and relationships between medical concepts distributed across different sections of clinical notes.

Conclusions:

• MedSecId provides a foundational resource for developing and benchmarking section identification systems in the medical domain.
• Automated section identification can significantly enhance the usability and analytical potential of clinical documentation.
• Further research can leverage MedSecId to explore deeper insights into the structure and content of medical records.