Automated Identification of Patent Ductus Arteriosus Using a Computer Vision Model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neonatal Cardiology

Background:

• Patent ductus arteriosus (PDA) is a frequent vascular defect in preterm infants, often necessitating intervention to prevent long-term complications.
• Echocardiography is the standard diagnostic and monitoring tool for PDA.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated identification of PDA in echocardiographic images of preterm infants.

Main Methods:

• A convolutional neural network was trained on 1145 color Doppler echocardiography clips from 66 preterm infants.
• A novel attention mechanism was employed to aggregate frame-level predictions for clip-level classification.
• Model performance was assessed using sensitivity, specificity, and Area Under the Curve (AUC).

Main Results:

• The best performing model achieved a clip-level sensitivity of 0.80 and specificity of 0.77 (AUC 0.86).
• Study-level performance reached a sensitivity of 0.83 and specificity of 0.89 (AUC 0.93).
• Color Doppler echocardiography clips yielded superior classifier performance.

Conclusions:

• The developed deep learning model demonstrates robust performance in classifying PDA in echocardiography clips.
• Further development and external validation are recommended for clinical integration.
• This AI tool represents a significant step towards semi-automated, bedside PDA detection in neonates.