Key result
Deep learning model using ultrasound video achieves ~85% accuracy for detecting pediatric ASDs.
Why the study?
Congenital heart defect is the most common birth defect, and this paper evaluated a deep learning method based on cardiac ultrasound video to assist in atrial septal defect diagnosis.
Does a deep learning model based on multiple instances learning improve the accuracy of atrial septal defect detection in pediatric echocardiography videos compared to clinical doctors?
Observational (n=330)
Double-blind
No
Does a deep learning model based on multiple instances learning improve the accuracy of atrial septal defect detection in pediatric echocardiography videos compared to clinical doctors?
A deep learning model utilizing multiple instances learning on echocardiography videos can accurately detect secundum atrial septal defects in children, outperforming the diagnostic accuracy of human sonographers.
No takes yet. Share an insight, caveat, or question.
May aid pediatric ASD detection on echo; leaves open prospective validation vs clinicians before practice change.
Liu et al. (2023) conducted an observational in Atrial Septal Defect (ASD) (n=330). Multiple instances learning-based deep learning model (resNet18 and r3D) vs. Junior and senior doctors (in manual test subset) was evaluated on Atrial septal defect detection accuracy. A multiple instances learning-based deep learning model using ultrasound video achieved an accuracy of 84.95% and AUC of 89.33% for detecting atrial septal defects in children.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: