ACC (2022) - Portable Artificial Intelligence Device Detects Heart Murmurs In Real Time In Tandem With A Regular Stethoscope
Cardiac auscultation is performed in virtually every clinical examination to identify heart disease before symptoms are present yet, in practice most heart murmurs are missed. Even significant valvular heart disease is identified by auscultation in only 43% of cases. There are several solutions which leverage digital stethoscope, but they all require a separate computer, and perform processing on the cloud, necessitating high speed internet access. We have designed an innnovative prototype device which can be attached to a regular stethoscope and used to automatically flag heart murmurs.
This device can be simply and discreetly connected between the stethoscope stem and tube, allowing physicians to use the stethoscope as they normally would. Leveraging recent advances in compressed convolutional neural networks has allowed for the development of highly accurate, low latency murmur detection algorithms which are deployable on energy-efficient microcontrollers. Using previously acquired and annotated heart sound recordings, we trained a deep neural network to detect murmurs. This neural network was compressed and deployed to a prototype device designed to filter and process the data in real-time on the edge. To test the algorithm, we enrolled patients with known stenosis and regurgitation lesions in the moderate and severe range as noted on the echocardiograms. These patients were auscultated by a cardiologist on the four cardiac sites.
50 patients were included in the study with known valvular heart disease which included stenosis and regurgitation lesions in the moderate and severe range. The device had a sensitivity of 95% when detecting murmurs accurately.
The study demonstrates the viability of this approach in screening for cardiac murmurs caused by valvular heart disease. Given the unique architecture of this device, which completes all processing locally without the need to transmit data to the cloud, this approach could enable an increase in adoption rates, empowering a plethora of physicians with a discrete, rapid, and accurate screening method for valvular heart disease.