Fast and accurate view classification of echocardiograms using deep learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiography is crucial in cardiology but limited by human interpretation for precision medicine.
• Deep learning (DL) shows promise for image analysis but faces challenges with complex echocardiogram formats.
• Automated view recognition is a critical step for AI-assisted echocardiography.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying standard echocardiographic views.
• To assess the model's accuracy compared to expert human interpretation.

Main Methods:

• A convolutional neural network was trained on 267 transthoracic echocardiograms (still images and videos).
• The model simultaneously classified 15 standard views (12 video, 3 still) from real-world clinical data.
• Model performance was evaluated on test data, including low-resolution images.

Main Results:

• The model achieved 97.8% accuracy in classifying 12 video views without overfitting.
• On single low-resolution images, the model attained 91.7% accuracy across 15 views, surpassing board-certified echocardiographers (70.2-84.0%).
• Data visualization confirmed the model uses clinically relevant features and recognizes view similarities.

Conclusions:

• Deep learning models can effectively learn to recognize echocardiographic views.
• AI-assisted interpretation holds potential for enhancing echocardiography and precision cardiology.
• This study provides a foundation for developing advanced AI tools in echocardiography.