Artificial intelligence-based classification of echocardiographic views.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiography analysis can be enhanced by artificial intelligence for automated parameter assessment and disease pattern recognition.
• Accurate cardiac view classification is a prerequisite for applying deep learning algorithms to echocardiographic data.

Purpose of the Study:

• To develop and evaluate convolutional neural networks (CNNs) for automated classification of echocardiographic views.
• To assess the performance of 2D and 3D CNNs on transthoracic echocardiography (TTE) and point-of-care ultrasound (POCUS) datasets.

Main Methods:

• Trained 2D and 3D CNNs on 10,269 TTE videos from 909 patients to classify nine cardiac view categories.
• Validated CNNs internally on 2,582 TTE videos from 229 patients.
• Tested CNNs on comprehensive TTE studies (100 patients) and POCUS videos (408 patients).

Main Results:

• The 2D CNN achieved 96.8% accuracy and 0.997 AUC on comprehensive TTE, and 98.4% accuracy and 0.998 AUC on POCUS.
• The 3D CNN achieved 96.3% accuracy and 0.998 AUC on TTE, and 95.0% accuracy and 0.996 AUC on POCUS.
• Positive predictive values exceeded 93% for specific views using 2D CNNs.

Conclusions:

• An automated cardiac view classifier using CNNs demonstrates high accuracy for TTE and POCUS.
• This AI-driven view classification tool facilitates the integration of deep learning in echocardiography.
• The developed classifier can improve the efficiency and accuracy of echocardiographic analysis.