UCLA

A critical need has emerged for constant, vigilant heart function monitoring of patients in the clinic who are at risk due to heart failure (HF). A major indicator of HF is reduced left ventricular ejection fraction, more commonly known as ejection fraction (EF). Current clinical methods of measuring EF include echocardiography, magnetic resonance imaging (MRI), radionuclide ventriculography, and radionuclide angiography. These methods are expensive, require the presence and support of expert technicians, and are not capable of continuous monitoring.

Our mission has been to develop the first wearable heart function monitoring system with the following characteristics: it must be capable of real-time, continuous monitoring; it must provide accurate measurement of EF; it must be low-cost and easy to use. To achieve this objective, we have developed the Integrated CardioRespiratory (ICR) Heart Function Monitoring System, which utilizes the electrical signal of the heart, the electrocardiogram (ECG), and the acoustic signal of the heart, the phonocardiogram (PCG), to compute EF. It consists of ICR acoustic sensors, ECG sensor electrodes, a sensor application system, and the ICR patient monitor system. The system acquires ECG and PCG signals, upon which signal processing is performed. This is accompanied by advanced machine learning methods which provide accurate computation of EF.

We present several signal processing and machine learning methods in order to accomplish this. Noise suppression algorithms for both ECG and PCG signals are described. Novel methods are presented for identifying key events in the PCG signals, and extracting from them critical features regarding cardiac dynamics and function. We have also developed a neural network architecture to utilize these features in order to compute an EF value.

The deployment of the ICR system in a clinical trial is discussed, in which the accuracy of its EF computation ability is demonstrated.