Deep Learning Natural Language Processing Successfully Predicts the Cerebrovascular Cause of Transient Ischemic...

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Risk stratification aids in triaging transient ischemic attack (TIA) clinic referrals.
• Natural Language Processing (NLP) using deep learning may predict cerebrovascular causes of TIA-like presentations from free-text clinical notes.

Purpose of the Study:

• To evaluate the efficacy of deep learning-based NLP models in predicting the cerebrovascular cause of TIA-like presentations using free-text clinical information.
• To compare the predictive capabilities of various classifier models based on the history of presenting complaint.

Main Methods:

• Retrieved consecutive TIA clinic notes from existing databases.
• Preprocessed free-text descriptions of the history of presenting complaint for classification experiments.
• Utilized various classifier models, including convolutional neural networks (CNNs), to predict the cause of TIA-like presentations.
• Assessed model performance using the area under the curve (AUC) of the receiver operator curve.

Main Results:

• The convolutional neural network (CNN) demonstrated the highest predictive capability (AUC ± SD; 81.9 ± 2.0) when trained solely on the history of presenting complaint.
• The integration of additional clinical information yielded variable effects on AUC.
• The highest AUC (88.3 ± 3.6) was achieved when the CNN utilized both the history of presenting complaint and magnetic resonance imaging (MRI) reports.

Conclusions:

• Deep learning-based NLP, particularly CNNs, shows potential for predicting the cause of TIA-like presentations from medical free-text.
• Further research is warranted to explore the application of deep learning-based NLP for automated triaging of TIA clinic referrals and other specialties.