Evaluation of Wearable Acoustic Sensors and Machine Learning Algorithms for Automated Measurement of Left Ventricular Ejection Fraction.

Left ventricular ejection fraction (EF) is a predictor of mortality and guides clinical decisions. Although transthoracic echocardiography (TTE) is commonly used for measuring EF, it has limitations, such as subjectivity and requires expert personnel. Advancements in biosensor technology and artificial intelligence are allowing systems capable of determining left ventricular function and providing automated measurement of EF. In this study, we tested new wearable automated real-time biosensors (Cardiac Performance System [CPS]) that compute EF using waveform machine learning on cardiac acoustic signals. The primary aim was to compare the accuracy of CPS EF with TTE EF. Adult patients presenting to cardiology, presurgical, and diagnostic radiology clinical settings in an academic center were enrolled. TTE examination was performed by a sonographer, followed immediately by a 3-minute recording of acoustic signals from CPS biosensors placed on the chest by nonexpert personnel. TTE EF was calculated offline using the Simpson biplane method. A total of 81 patients (aged 19 to 88 years, 27 women, 20% to 80% EF) were included. Deming regression and Bland-Altman analysis were performed to assess the accuracy of CPS EF against TTE EF. Both Deming regression (slope 0.9981; intercept 0.03415%) and Bland-Altman analysis (bias -0.0247%; limits of agreement [-11.65, 11.60]%) demonstrated equivalency between CPS EF and TTE EF. The receiver operating characteristic for measuring sensitivity and specificity of CPS in identifying subjects with abnormal EF showed an area under the curve value of 0.974 for identifying EF <35% and 0.916 for detecting EF <50% CPS EF intraoperator and interoperator assessments demonstrated low variability. In conclusion, this technology measuring cardiac function from noninvasive biosensors and machine learning on acoustic signals provides an accurate EF measurement that is automated, real-time, and acquired rapidly by personnel with minimal training.

Assessment of Left Ventricular Diastolic Function using Phonocardiogram Signals: A Comparison with Echocardiography.

This paper presents a fully-automated end-to-end phonocardiogram(PCG)-based wearable system capable of providing echocardiography-like metrics for left ventricular (LV) diastolic function assessment. Proxy metrics for five echocardiographic parameters were calculated based on physiologically-motivated features extracted from PCG signals using noise-subtraction, heartbeat-segmentation, and quality-assurance algorithms. The clinical value of these proxy metrics was evaluated using the latest American Society of Echocardiography/European Association of Cardiovascular Imaging guidelines for evaluation of LV diastolic function. When tested on a group of n=34 patients, proxy metrics successfully identified LV diastolic dysfunction in a n=29 subset with 87.5% accuracy, and elevated LV filling pressures in a n=17 subset with 75% accuracy.

Fully-Automated Diagnosis of Aortic Stenosis Using Phonocardiogram-Based Features.

The irreversible damage and eventual heart failure caused by untreated aortic stenosis (AS) can be prevented by early detection and timely intervention. Prior work in the field of phonocardiogram (PCG) signal analysis has provided proof of concept for using heart-sound data in AS diagnosis. However, such systems either require operation by trained technicians, fail to address a diverse subject set, or involve unwieldy configuration procedures that challenge real-world application. This paper presents an end-to-end, fully-automated system that uses noise-subtraction, heartbeat-segmentation and quality-assurance algorithms to extract physiologically-motivated features from PCG signals to diagnose AS. When tested on n=96 patients showing a diverse set of cardiac and non-cardiac conditions, the system was able to diagnose AS with 92% sensitivity and 95% specificity.