A deep-learning based system for accurate extraction of blood pressure data in clinical narratives.

Researchers

Journal

Modalities

Models

Abstract

This study presents a novel workflow for identifying and analyzing blood pressure readings in clinical narratives using a Convolution Neural Network. The network performs three tasks: identifying blood pressure readings, determining the exactness of the readings, and then classifying the readings into three classes: general, treatment, and suggestion. The system can be easily set up and deployed by people who are not experts in clinical Natural Language Processing. The validation results on an independent test set show the first two of the three tasks achieve a precision, recall, and F-measure over or close to 95%, and the third task achieves an overall accuracy of 85.4%. The study demonstrates that the proposed workflow is effective for extracting blood pressure data in clinical notes. The workflow is general and can be easily adapted to analyze other clinical concepts for phenotyping tasks.
©2020 AMIA – All rights reserved.