Semi-supervised learning with natural language processing for right ventricle classification in echocardiography-a...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Assessing right ventricular (RV) size and function is crucial in echocardiography.
• Manual analysis of echocardiographic images and reports is time-consuming and labor-intensive.
• Developing automated methods can improve efficiency and scalability in cardiac diagnostics.

Purpose of the Study:

• To develop and validate deep learning models for automated assessment of RV size and function from echocardiographic images.
• To leverage natural language processing (NLP) for classifying text reports and generating training data for image analysis models.
• To create a pipeline for auto-annotation of echocardiographic images using NLP and medical reports.

Main Methods:

• A deep learning model was trained using NLP to classify echocardiographic reports.
• A view classifier was developed to identify relevant echocardiographic views (4-chamber, RV-focused).
• Image classification models were trained on automatically annotated data to assess RV size and function.

Main Results:

• The NLP text classifier achieved high sensitivity and specificity for identifying impaired RV function and RV enlargement.
• The image classification models demonstrated substantial agreement with written reports for RV function and size assessment.
• The automated pipeline enabled training of image-assessment models without manual image annotation, facilitating dataset expansion.

Conclusions:

• Deep learning models, trained with NLP-assisted auto-annotation, can effectively assess RV size and function from echocardiograms.
• This approach offers a fast, cost-effective method for expanding training datasets for cardiac image analysis.
• The developed pipeline has the potential to streamline the diagnostic process and improve the efficiency of cardiac assessments.