Automated identification of innocent Still's murmur using a convolutional neural network

Area of Science:

• Pediatric Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Still's murmur is the most common innocent heart murmur in children.
• Primary care providers often misidentify Still's murmur, leading to unnecessary referrals.
• Accurate identification of innocent murmurs is crucial to avoid over-referral.

Purpose of the Study:

• To develop a computer algorithm for identifying Still's murmur at the point of care.
• To assist primary care providers in distinguishing innocent murmurs from pathological ones.
• To reduce the rate of over-referral to pediatric cardiology specialists.

Main Methods:

• A convolutional neural network-based algorithm was trained and tested.
• 1,473 pediatric patient heart sound recordings were analyzed.
• Recordings included Still's murmurs, pathological murmurs, other innocent murmurs, and normal heart sounds.

Main Results:

• The algorithm achieved 90.0% sensitivity and 98.3% specificity in identifying Still's murmur.
• Murmur sounds from the lower left sternal border yielded the highest accuracy.
• The area under the receiver operating characteristic curve was 0.943.

Conclusions:

• The developed algorithm accurately identifies Still's murmur.
• This AI tool can decrease unnecessary referrals to pediatric cardiologists.
• The algorithm may reduce the need for echocardiography in cases of benign findings.