1Friedrich-Alexander-Universität Erlangen-Nürnberg Medizinische Klinik 2 - Kardiologie und Angiologie Erlangen, Deutschland; 2Friedrich-Alexander-Universität Erlangen-Nürnberg Machine Learning and Data Analytics Lab, Department Artificial Intelligence in Biomedical Engineering Erlangen, Deutschland
Background
Preliminary studies indicate potential benefits of early intervention in asymptomatic severe aortic stenosis patients. Considering echocardiography's resource limitations, digital auscultation utilizing built-in smartphone microphones presents a viable screening alternative.
Methods
The developed smartphone application incorporates real-time signal analysis to ensure high-quality recordings using the built-in microphone. During the recording, frequencies around 100 Hz and 250 Hz are amplified to align with traditional stethoscope responses and the resulting signal amplitudes are visualized. Audio recordings from smartphones and stethoscopes were collected for 38 severe aortic stenosis patients and 24 controls. To validate signal quality, three experienced physicians visually classified the Mel spectrograms and classification accuracies compared to echocardiography are reported. The noninferiority of built-in smartphone microphones compared to digital stethoscopes was assessed using the Durkalski noninferiority test with a 5% margin. Agreement between raters was evaluated using Fleiss’ kappa.
Results
Both smartphone and stethoscope recordings distinctly revealed the peaks of the first and second heart sounds around 50 Hz, as well as the characteristic crescendo-decrescendo murmur associated with aortic stenosis in the 100-600 Hz range (Figure 1). By visual analysis, classification accuracy using the smartphone was 83% compared to 92% using the digital stethoscope. Usage of smartphone microphones spectrograms was noninferior for the detection of severe aortic stenosis compared to digital stethoscopes (p=0.004). Agreement between raters was almost perfect for both smartphone recordings (κ=0.81) as well as stethoscope recordings (κ=0.85).
Conclusions
Recordings of built-in smartphone microphones are noninferior to digital stethoscopes regarding the identification of severe aortic stenosis. The signals are of sufficient quality and hence this approach holds promise as a reliable and cost-effective screening tool for early detection of valvular heart disease.