Seismocardiography-Based Cardiac Computed Tomography Gating Using Patient-Specific Template Identification and Detection

Abstract

To more accurately trigger cardiac computed tomography angiography (CTA) than electrocardiography (ECG) alone, a sub-system is proposed as an intermediate step toward fusing ECG with seismocardiography (SCG). Accurate prediction of quiescent phases is crucial to prospectively gating CTA, which is susceptible to cardiac motion and, thus, can affect the diagnostic quality of images. The key innovation of this sub-system is that it identifies the SCG waveform corresponding to heart sounds and determines their phases within the cardiac cycles. Furthermore, this relationship is modeled as a linear function with respect to heart rate. For this paper, B-mode echocardiography is used as the gold standard for identifying the quiescent phases. We analyzed synchronous ECG, SCG, and echocardiography data acquired from seven healthy subjects (mean age: 31; age range: 22-48; males: 4) and 11 cardiac patients (mean age: 56; age range: 31-78; males: 6). On average, the proposed algorithm was able to successfully identify 79% of the SCG waveforms in systole and 68% in diastole. The simulated results show that SCG-based prediction produced less average phase error than that of ECG. It was found that the accuracy of ECG-based gating is more susceptible to increases in heart rate variability, while SCG-based gating is susceptible to high cycle to cycle variability in morphology. This pilot work of prediction using SCG waveforms enriches the framework of a comprehensive system with multiple modalities that could potentially, in real time, improve the image quality of CTA.

Keywords: Cardiac gating; cardiac quiescence; computed tomography angiography (CTA); coronary angiography; echocardiography; electrocardiography (ECG); seismocardiography (SCG).

FIGURE 1.

Echocardiography: B-mode echocardiography frame from an apical four chamber view of the heart, with contour shown around the inter-ventricular septum (IVS); Echo Deviation: Motion signals calculated from B-mode sequences by applying the phase-to-phase deviation measure elaborated in ; ECG: Time-series de-noised ECG signal; SCG: Time-series de-noised SCG signal.

FIGURE 2.

Plot of de-noised ECG and SCG wave packets associated with heart sounds and , denoted as and , respectively, from clean SCG signal. Heart sounds are discrete bursts of auditory vibrations that vary in intensity, frequency, quality and duration . By placing an accelerometer on the sternum, the first and second heart sounds can be detected . The first heart sound () is caused by the closure of atrio-ventricular valves; the second heart sound () is from the closure of semilunar valves . leads the onset of systolic period while occurs after the systolic period and precedes the diastolic period.

FIGURE 3.

Schematic system diagram outlining SCG and ECG-based prediction. The upper part of the diagram uses training data to generate the phase delay function (Section II-D) supported by the heart sound-associated waveform detection of SCG (Section II-C) and voting mechanism of echocardiography (Section II-B). The lower part uses testing data to recognize the phase of current heart sound-associated waveform and predict the current heart rate (Section II-F). Eventually, the quiescent phase is predicted with the joint information of the training and testing data (Section II-G).

FIGURE 4.

Example of the composite envelope and the difference envelope which is the difference between the upper and lower envelope formed using Hilbert transform. The peak of the difference envelope is considered the optimal peak of the composite waveform.

FIGURE 6.

Example of absolute phase error versus heart rate of the two heart sound waveforms of SCG and ECG. The left figure is from a healthy subject (subject No. 4), and the right figure is from a cardiac patient (subject No. 14). The absolute phase errors were calculated based on the phase delay function within each cohort.

FIGURE 10.

Average absolute phase error against average heart rate variation of all the 18 subjects. The absolute phase errors on the left are calculated based on the patient-specific phase delay function, and on the right are from each cohort.

FIGURE 5.

Waveform identification rate. The first seven scatter points are from healthy subjects and the rest are from cardiac subjects. H represents healthy subjects and P represents congenital heart disease patients. The subject number is reconciled with those in TABLE 1.

FIGURE 7.

Average absolute phase error against average heart rate of all the 18 subjects. The absolute phase errors in the left figure were calculated based on the patient-specific phase delay function, and in the right figure was from each cohort.

FIGURE 8.

Example of absolute phase error versus heart rate error for ECG and the two heart sound waveforms from SCG. The figure on the left is from the a healthy subject (subject No. 7), and the figure on the right is from a cardiac patient (subject No. 9). The absolute phase errors are calculated based on the patient-specific phase delay function.

FIGURE 9.

Average absolute phase error against average heart rate error of all the 18 subjects. The absolute phase errors on the left are calculated based on the patient-specific phase delay function, and on the right are from each cohort.