Reconstruction of 3-Axis Seismocardiogram from Right-to-left and Head-to-foot Components Using A Long Short-Term Memory Network

This pilot study aims to develop a deep learning model for predicting seismocardiogram (SCG) signals in the dorsoventral direction from the SCG signals in the right-to-left and head-to-foot directions (SCGx and SCGy). The dataset used for the training and validation of the model was obtained from 15 healthy adult subjects. The SCG signals were recorded using tri-axial accelerometers placed on the chest of each subject. The signals were then segmented using electrocardiogram R waves, and the segments were downsampled, normalized, and centered around zero. The resulting dataset was used to train and validate a long short-term memory (LSTM) network with two layers and a dropout layer to prevent overfitting. The network took as input 100-time steps of SCGx and SCGy, representing one cardiac cycle, and outputted a vector that mapped to the target variable being predicted. The results showed that the LSTM model had a mean square error of 0.09 between the predicted and actual SCG segments in the dorsoventral direction. The study demonstrates the potential of deep learning models for reconstructing 3-axis SCG signals using the data obtained from dual-axis accelerometers.

I. Introduction

Cardiovascular diseases (CVDs) are the leading cause of death in the United States, claiming the life of one person every 34 seconds, resulting in a staggering 2,544 deaths per day based on 2020 data [1]. Beyond the devastating human toll, this places an immense strain on healthcare systems and society, with significant economic costs associated with treatment and lost productivity. The early detection of cardiac abnormalities is crucial for achieving better outcomes for patients with CVD, and improving diagnostic methods and accessibility is a key step in this direction. Current diagnostic methods, including non-invasive techniques such as electrocardiography (ECG), medical imaging, and cardiac catheterization, can aid in identifying CVDs. Advancements in technology have also led to the development of new diagnostic options, such as wearable and remote monitoring systems, which can provide continuous monitoring of patients' cardiovascular health outside of traditional healthcare settings. Seis-mocardiography (SCG) is another technique that noninvasively monitors cardiovascular activity by measuring cardiovascular-induced vibrations on the chest [2]–[4]. These vibrations result from a range of cardiac activities, including valve opening and closing, isovolumetric contraction, blood ejection, and rapid left ventricle filling [3], [5]–[7]. Unlike other non-invasive techniques such as ECG and pulse oximetry, which focus on the electrical activity of the heart and the blood oxygen level, respectively, SCG provides complementary insights into the mechanical activity of the heart [5], [8], [9]–. With its ability to evaluate these mechanical activities, SCG has the potential to offer valuable diagnostic information for cardiac conditions such as heart failure, myocardial infarction, ischemia, and hemorrhage, as changes in the mechanical function of the heart can be an early indication of these diseases [10]–[15]. As a result, SCG can enhance our understanding of the cardiac function and contribute to the development of more accurate diagnostic tools for patients with CVD. SCG signals are commonly measured using accelerometers that are placed on the chest surface. These signals are typically measured in three directions of right-to-left, head-to-foot, and dorsoventral. In that regard, while single or dual-axis accelerometers may be used for SCG measurement, three-axis accelerometers are more informative as they offer a more comprehensive understanding of the motion of the heart and chest wall [8].