Deep learning for transesophageal echocardiography view classification.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Transesophageal echocardiography (TEE) is crucial for assessing cardiac conditions and guiding cardiac surgery.
• A significant challenge in utilizing TEE data for deep learning is the inherent complexity and unstructured nature of the images.
• Standardized classification of TEE views is essential for consistent data analysis and interpretation.

Purpose of the Study:

• To develop and validate a deep learning-based model for multi-category classification of standardized TEE views.
• To introduce structure into intraoperative and intraprocedural TEE imaging data.
• To facilitate downstream deep learning applications on TEE data.

Main Methods:

• A convolutional neural network (CNN) was trained to predict standardized TEE views.
• The model was trained using labeled intraoperative and intraprocedural TEE videos from Cedars-Sinai Medical Center (CSMC).
• External validation was performed on intraoperative TEE videos from Stanford University Medical Center (SUMC).

Main Results:

• The deep learning model demonstrated high accuracy in classifying all labeled TEE views.
• Top performance was observed for Trans-Gastric Left Ventricular Short Axis View (AUC 0.971 at CSMC, 0.957 at SUMC).
• Other highly accurate classifications included Mid-Esophageal Long Axis View, Mid-Esophageal Aortic Valve Short Axis View, and Mid-Esophageal 4-Chamber View.

Conclusions:

• The developed deep learning model accurately classifies standardized TEE views.
• This classification capability enhances the utility of TEE imaging for deep learning analyses.
• The model provides a structured approach to intraoperative and intraprocedural TEE data, paving the way for advanced AI-driven insights.