Dependence Of SCG Clustering On SCG Measurement Location

Abstract

<h3>Introduction</h3> Seismocardiography (SCG) refers to audible and subsonic heart sounds acquired on the chest wall, and is potentially useful for HF diagnosis and monitoring. Previous studies typically measured SCG signals at a location which varied among different studies (e.g. xiphoid process, 4th ICS, etc.). One potentially useful SCG feature is the tendency of the signal to cluster into two different respiration dependent forms, and this clustering may in tern depend on SCG measurement location. <h3>Objective</h3> Investigate the dependence of SCG clustering on SCG measurement location. <h3>Methods</h3> SCG, electrocardiography (ECG), and spirometry were measured in 15 healthy males (19-31 yo) at 3 locations (mid-sternum, 4^th ICS at left lower sternal border, and xiphoid process). Respirations were measured using spirometry. SCG events were segmented using the ECG-R wave, then grouped into 2 clusters using the "k-medoid" algorithm. The "decision boundary" between clusters and clustering accuracy were determined using support vector machine (SVM) methods. <h3>Results</h3> Figure 1 shows a sample distribution of the clustered events in the mid-sternum during the respiratory cycle. Table 1 shows the classification parameters in % for the three measurement locations. Although of comparable accuracy, the mid-sternum SCG location had a 76% larger SD. The data also demonstrates a 16-27% average clustering disagreement between the mid-sternum and the other two "listening" positions. <h3>Conclusions</h3> The chest wall location chosen for "listening" to the SCG may be important to maximize SCG's utility for HF diagnosis and monitoring. The mid-sternum had slightly lower average classification accuracy with relatively higher SD compared to the LLSB or the xiphoid. More research is needed to further optimize SCG sensor placement positions for use in HF patients.