Two-center validation of Pilot Tone Based Cardiac Triggering of a Comprehensive Cardiovascular Magnetic Resonance Examination

Background The electrocardiogram (ECG) signal is prone to distortions from gradient and radiofrequency interference and the magnetohydrodynamic effect during cardiovascular magnetic resonance imaging (CMR). Although Pilot Tone Cardiac (PTC) triggering has the potential to overcome these limitations, effectiveness across various CMR techniques has yet to be established. Purpose To evaluate the performance of PTC triggering in a comprehensive CMR exam. Methods Fifteen volunteers and twenty patients were recruited at two centers. ECG triggered images were collected for comparison in a subset of sequences. The PTC trigger accuracy was evaluated against ECG in cine acquisitions. Two experienced readers scored image quality in PTC-triggered cine, late gadolinium enhancement (LGE), and T1- and T2-weighted dark-blood turbo spin echo (DB-TSE) images. Quantitative cardiac function, flow, and parametric mapping values obtained using PTC and ECG triggered sequences were compared. Results Breath-held segmented cine used for trigger timing analysis was collected in 15 volunteers and 14 patients. PTC calibration failed in three volunteers and one patient; ECG trigger recording failed in one patient. Out of 1987 total heartbeats, three mismatched trigger PTC-ECG pairs were found. Image quality scores showed no significant difference between PTC and ECG triggering. There was no significant difference found in quantitative measurements in volunteers. In patients, the only significant difference was found in post-contrast T1 (p = 0.04). ICC showed moderate to excellent agreement in all measurements. Conclusion PTC performance was equivalent to ECG in terms of triggering consistency, image quality, and quantitative image measurements across multiple CMR applications.


Introduction
Synchronization of cardiovascular magnetic resonance (CMR) data acquisition to cardiac motion is typically required to obtain artifact-free images of the heart and owing blood. Conventionally, the R-wave of the electrocardiogram (ECG) is used as a trigger signal to synchronize data acquisition with the cardiac cycle. However, ECG setup can be time consuming; skin preparation is required and electrodes may need to be repositioned several times to obtain an adequate signal, thereby adding to the overall exam time [1]. The ECG signal also is prone to distortion caused by gradient and radiofrequency (RF) interference as well as the magnetohydrodynamic effect [2][3][4]. When the ECG fails, triggering based on peripheral pulse oximetry is often used as a backup; however, this method is susceptible to nger movement [5], frequently fails in patients with poor peripheral circulation, and an unpredictable trigger delay makes it di cult to adjust pulse sequence timing [6]. Cardiac triggering using other external biosensors have been proposed to overcome the limitations of ECG synchronization [6][7][8][9][10][11]. However, these methods typically require additional investigative hardware and operator training, making them less practical in a non-research setting. Self-gating (SG) cardiac synchronization extracts a cardiac motion signal from central k-space lines acquired at a constant time interval, eliminating the need for external gating hardware [12][13][14]. However, SG requires continuous acquisition and consistent contrast, thus its applications are limited primarily to 3D and dynamic imaging. Changes in RF and gradient pulses can also distort the SG signal [15], and the need to acquire additional central k-lines can prolong acquisition time [16].
The Pilot Tone (PT) technique was proposed as a novel method to synchronize CMR acquisition with cardiac and respiratory motion [17]. The PT is a continuous RF signal generated by a small loop antenna positioned close to or inside the scanner bore, or mounted to a receiver coil. The frequency of the PT signal is within the bandwidth of the MR receiver system, but outside of the imaging bandwidth [17][18][19]. Hence, the PT does not disrupt the timing of the CMR pulse sequence, nor does it induce image artifacts. The PT is modulated by physiological motion and can be detected by the MR receiver coils to extract both respiratory and cardiac motion signals [20]. The cardiac component is highly correlated with cardiac contractile motion and can be viewed as a surrogate of ventricular volume [18] and used as a contact-free alternative to the ECG to synchronize data acquisition. The feasibility of retrospective and prospective PT cardiac (PTC) triggering has been previously demonstrated [21], and a fully automatic real-time PTC research framework was subsequently integrated with a variety of CMR applications. The objective of this exploratory study is to evaluate the feasibility of performing a comprehensive CMR exam with PTC triggering, and compare its performance with standard ECG synchronization in healthy volunteers and in patients referred for clinical imaging.

Study population
Fifteen healthy volunteers (36.8 ± 14.7 years, 7 females) and 20 cardiac patients (44.0 ± 15.9 years, 8 females) were recruited from two CMR centers (15 volunteers and 6 patients from Site A; 14 patients from Site B). The study was approved by the respective local institutional review boards and informed written consent was obtained from all subjects. The exclusion criteria for volunteers included: 1) under 18 years of age, 2) pregnant, 3) claustrophobic, and 4) standard contraindications to CMR. All cardiac patients were referred for CMR by local physicians. Demographic information for volunteers and patients is included in Supplementary Material 1.

PTC signal extraction and trigger detection
Raw PT signals were acquired using the imaging receiver coils, simultaneous with image acquisition, and sampled at a rate of 2000 Hz. Detuning of the receiver coils during RF transmission, as is normally performed to avoid damage to the receiver system, induced a spike-like artifact in the PT signal. To eliminate this interference and to extract cardiac motion, a previously described PTC signal-processing algorithm [22] was employed as follows. A calibration scan consisting of four RF pulse trains (SINC shaped envelope, 1 ms duration, 70° ip angle, 40 pulses, 4ms between pulses) was performed, with a two second gap between pulse trains. RF interference was characterized by averaging and debiasing the PT samples over the RF pulse train repetitions. Principal component analysis (PCA) was applied, and the rst two eigenvectors were stored as RF artifact subspace. Next, raw PT signals from a 20s free breathing undisturbed training scan were debiased and bandpass ltered with separate respiratory (0.2-0.6 Hz) and cardiac (0.8-2.5 Hz) frequency bands. PCA was applied to the respiratory ltered signal and the rst two eigenvectors were stored as respiratory subspace. Finally, PCA was applied to the cardiac ltered signal and the rst eigenvector orthogonal to both the RF artifact subspace and the respiratory subspace was calculated. The outcome of this approach yielded a channel combination vector that maximized the cardiac signal while suppressing respiratory contributions and RF artifacts. This vector was used in subsequent scans to combine un ltered PT signals from all receiver coils into a single channel PT cardiac surrogate. A constant velocity Kalman lter was applied to the PT cardiac surrogate, and the rst derivative taken to generate a denoised signal with no direct current offset and a latency of approximately 100 ms. This ltered rst derivative signal enabled prospective triggering and is referred to as the PTC signal. The processing of the PTC did not assume regular rhythm; the only assumption made during the PTC calibration process was that the heartrate fell within the range of 48-150 beats per minute (0.8-2.5 Hz).
The polarity of the PT cardiac surrogate was determined during calibration based on skewness. Under resting conditions, diastole is longer than systole, thus the skewness indicates the direction of diastole and can be used to de ne the polarity of the PTC. The maximum of the PTC signal acquired during the calibration scan was also stored, and the threshold for triggering set at 0.4 times this maximum. This threshold was set empirically as a compromise between reliably avoiding double triggers caused by the mid-diastolic "shoulder" of the signal, while simultaneously avoiding missing triggers due to low signal maxima. A diagram of the PTC calibration process is summarized in Fig. 1.
The delay between the PTC trigger and ECG R-wave can be estimated during calibration by assuming that the PT cardiac surrogate reaches its maximum simultaneous with the R-wave. The delay was calculated as the average of the times between the PT trigger points in the training data and their preceding PT cardiac surrogate local maxima, plus the processing latency induced by Kalman lter. This delay estimation was later used to automatically set the acquisition window for single-shot scans, which are typically timed to diastole. Trigger analysis PTC and ECG triggered, breath-hold segmented cine series were acquired in the majority of subjects, including 15 of the volunteers at Site A, and 14 of the patients at Site B. These data were used to evaluate PTC trigger accuracy. During each PTC triggered image acquisition, the ECG signal and triggers were recorded simultaneously by the scanner. PTC and ECG trigger timing were evaluated using MATLAB (MathWorks, Natick, MA, USA). The rst heartbeat was not used to acquire image data; therefore, the rst trigger was excluded from the analysis. PTC and its corresponding ECG triggers were identi ed. The Rwave based ECG trigger was assumed to always occur prior to the motion-based PTC trigger, as electrical activity of the heart precedes mechanical motion. Therefore, trigger delay was de ned as the time elapsed from the ECG trigger to the PTC trigger within each heartbeat. Trigger jitter was de ned as the standard deviation of the trigger delays across heartbeats. Both mean trigger delay and trigger jitter were calculated in milliseconds and as a fraction of average cardiac cycle measured by ECG (% RR ECG ).

Image qualitative analysis
To qualitatively evaluate the image sharpness and artifact level, matching cine, LGE, T1-and T2-weighted DB-TSE images acquired with both triggering methods were scored by two experienced CMR readers (30 and 8 years of experience); these images came from subsets of volunteers and patients at each center, depending on the availability of PTC and ECG comparative data. Images from all cardiac views from a single subject, single application, and single triggering method were combined into a single display and presented in random order to the readers who were blinded to the triggering method. The scoring criteria ranging from 5: excellent, no apparent artifacts and/or blurring to 1: very poor, image totally obstructed by artifacts and/or blurring. In 6 patients from Site A with both ECG and PTC triggered LGE, the images were randomized and the global presence/absence of LGE was evaluated by a physician.

Image quantitative analysis
Cardiac function and ow measurements were quantitatively evaluated using suiteHEART (NeoSoft, Pewaukee, WI, USA). Biventricular cardiac output (CO), stroke volume (SV), ejection fraction (EF), and end systolic and end diastolic volumes (ESV, EDV) were measured from cine images acquired in both volunteers (Site A) and patients (Site B). Aortic and MPA CO, SV, and peak velocities were measured from phase contrast images. Myocardial T1, T2, and T2* relaxation times were measured in the interventricular septum from parametric maps in volunteers, per SCMR guidelines [23]. In patients scanned at Site A, where both PTC and ECG triggering was performed, pre and post contrast T1 values were measured in the interventricular septum, and extracellular volume (ECV) fractions were calculated.

Statistical analysis
The Wilcoxon signed rank test with an alpha level of 0.05 was performed on the image quality scores to test for differences between the two triggering methods. Pairwise Student's t-test with an alpha level of 0.05 was performed on the quantitative measurements to test for signi cant differences. The p-values of biventricular cardiac function measurements, aortic ow, MPA ow, and parametric mappings were adjusted separately using the Benjamini and Hochberg correction method [24] with a false discovery rate of 0.05. Two-way mixed effects, absolute agreement, single rater intra-class correlation coe cients (ICC) were also computed to test the agreement between the two triggering methods. An ICC of 0.9-1 indicated excellent agreement; 0.75-0.89 indicated good agreement; 0.5-0.74 indicated moderate agreement; and 0-0.49 indicated poor agreement. Figure 2 shows images for each CMR sequence acquired using both triggering methods in a healthy volunteer scanned at Site A, and Fig. 3 shows PTC triggered images in a patient scanned at Site B.

Results
Movies of cine and ow image series in the healthy volunteer can be found in Supplementary Material 4 and 5. The number of subjects included in each analysis is listed in Table 1. Volunteer and patient data were analyzed separately; as a result, data from the two centers were not combined in any of the subsequent analyses.

Trigger analysis
An example of PTC vs. ECG signal and triggers are shown in Fig. 4a. PTC signal extraction failed in one volunteer at Site A who had dextrocardia with situs inversus. The recording of ECG trigger timing failed in one patient at Site B. The PTC signal was inverted in two volunteers and one patient, i.e., the skewness criteria to determine PT signal polarity failed, causing the trigger timing to be shifted to the time of diastolic lling. Although PTC triggered images were collected and scored for quality in these subjects, PTC triggering was considered to be a failure and these data were excluded from the trigger timing analysis. Out of 1987 PTC triggers from the remaining subjects (12 volunteers at Site A and 12 patients at Site B), three mismatches with ECG were found (Fig. 4b). The average trigger delay and trigger jitter in each subject are listed in Table 2. Figure 5 shows the trigger delay in milliseconds vs. RR interval for both healthy volunteers (5a) and patients (5b), as well as one example of an outlier (5c). LGE was present in three of the six patients and the ndings from both triggering methods agreed.   A delay was observed in the PTC trigger with respect to ECG based R-wave triggering. This was expected as the PTC trigger is based on the mechanical contraction of the heart, which is preceded by the R-wave.
The length of delay varied somewhat between subjects, potentially due to subtle differences in cardiac anatomy and function; this variation was also observed in other PTC studies [20,25]. The PTC processing induced an additional systematic delay which was consistent among subjects. This delay could vary from that reported in other studies due to the use of different processing algorithms (i.e., prospective vs. retrospective, lter length, and peak triggering vs. threshold triggering). Some beat-to-beat variation in PTC trigger delay, or "jitter", measured relative to the ECG trigger, was observed. Taking the ECG trigger time as the gold standard, jitter was assumed to be purely caused by the instability of PTC. The worstcase jitter measured in any subject was less than the temporal resolution of the image acquisition. Jitter was worse in patients than in healthy volunteers. This may be due to alterations of excitation-contraction coupling [26], or dysfunction of electrical conduction pathways [27,28] that can accompany cardiac disease. Since the PTC is dependent on the mechanical motion of the heart, which is ultimately what impacts CMR image quality, timing of data acquisition and image quality may actually be improved with PTC in cases where there is a mismatch between electrical activity and mechanical contraction [29].
Among the 24 subjects with successful PTC calibration, only three trigger discrepancies were found between PTC and ECG, and two of these appeared to be due to premature ventricular contraction (PVC).
Arrhythmias can be problematic for cardiac synchronization regardless of the triggering technique [30]; however, given that the purpose of cardiac triggering is to synchronize data acquisition with cardiac motion, it is logical to assert that trigger detection based on a motion signal may be more effective than the electrical ECG signal. In a previous study [29], when both triggering methods were compared with cardiac motion extracted from real-time cine images, PTC was shown to have fewer mis-triggers than ECG, especially in patients with arrhythmias. Our designation of any discrepancy between PTC and ECG as an error in the PTC trigger was a conservative approach that did not consider this.
Although there was no difference in cine, LGE, and DB-TSE image quality scores between the two triggering methods in terms of blurring and artifacts, it is worth noting that the PTC triggered cine and ow series started at mid-systole, rather than end diastole as is the case for ECG triggering. This timing difference did not affect quanti cation of global cardiac function or ow, as standard commercial software can typically choose the end-systolic and end-diastolic frames [31], regardless of the timing of these frames within the image series.
In static imaging applications including parametric mapping, the trigger jitter was negligible compared to the duration of diastole and caused no signi cant difference in myocardial mapping values in volunteers. However, in patients the post contrast T1 values measured using PTC were found to be higher than those measured using ECG. Given that PTC images in this patient cohort were consistently obtained after conventional ECG images, the signi cance of this nding could largely be attributed to the difference in post-contrast injection timing.
The PTC delay time was estimated during calibration and the trigger timing automatically adjusted accordingly. However, this process may not be effective for the dark-blood preparation pulse which is typically applied at end-diastole (the ECG R-wave) with data acquisition timed to late diastole of the same cardiac cycle [32]; a mismatch in this timing could lead to myocardial signal loss. While the trigger timing of DB-TSE was set automatically in this manner, minimal cardiac motion artifact was observed. The majority of artifacts were related to respiratory motion, and these were seen in both PTC and ECG triggered images.
The PTC requires a small loop antenna known as the PT generator; such a device has been embedded in the anterior body array coil of several current Siemens MRI systems operating at various eld strengths. No hardware or software modi cations are required to obtain PTC signals on these scanners. No additional patient preparation is required beyond positioning the coil so that the PT generator is placed directly above the heart. At the time of the study, PTC signal and triggering were detected using a research algorithm, which is now commercially available as Beat Sensor®. The elimination of the time associated with ECG lead placement and the challenges of ECG interference, especially at higher eld, make the PTC a promising cardiac synchronization technique.
While the potential advantages of PTC are substantial, this study has several limitations. The small sample size limits the strength of conclusions that can be drawn from a number of comparisons.
LGE and parametric maps with both triggering methods were available in only six patients from one center; no de nite conclusions should be drawn based on such a limited sample. It must also be noted that PTC calibration failed in several subjects. PTC signal polarity was inverted in three subjects with high heart rates, causing the PTC trigger time to be shifted to peak diastolic lling rather than peak systolic contraction, introducing additional delay relative to the R-wave. This difference in trigger timing did not impact cine, ow, and parametric mapping image analyses, since the delay was consistent from beat-tobeat. However, a more robust determination of PT signal polarity is desirable. Poor PTC performance in the subject with situs inversus may have been due to the position and orientation of the heart relative to the PT generator, and perhaps could have been corrected by repositioning the coil. Although the PTC algorithm has been improved since the time of this study, adjusting the coil position and recalibrating may still be required. If these steps do not improve PTC triggering, then falling back to conventional triggering methods should be considered.
The current study included three patients with BMI over 40, and two patients with EF less than 35%, including one patient in whom both factors coexisted. No issues were observed in PTC triggered sequences in these patients. Although the patients in this study were referred for CMR with a variety of indications, further evaluation is needed in patients with a range of body habitus, cardiac orientations, and heart rates, and in various conditions that can alter cardiac motion and blood ow. The motionderived PT signal could be attenuated in severely obese or small pediatric patients, or in patients with heart failure with reduced EF, or ventricular dyssynchrony. On the other hand, ECG triggering may be challenging in patients with conduction system disorders. Further investigation is required to compare the performance of PTC to ECG under the above circumstances.

Conclusion
PTC triggering was successfully evaluated across a wide range of CMR applications in healthy volunteers and in cardiac patients at two una liated imaging centers. PTC performance was compared with standard ECG triggering and found to provide accurate triggering as well as comparable image quality and quantitative results. PTC may offer a more e cient and effective method than ECG for CMR cardiac synchronization. Authors' contributions YP acquired a subset of volunteer and patient data at Site A, performed the analysis, and was the primary contributor in writing the manuscript. JV, NJ assisted in the scan protocol setup and data acquisition at Site A. MT assisted in the image quality and diagnostic assessment. VY provided assistance with statistical analysis. AA, PG, RW, SNV, and DP assisted in the scan protocol setup and data acquisition at Site B. MB, CH, and PS developed the technique used in this study. OPS assisted in the image quality assessment and was a major contributor in writing the manuscript. All authors read and approved the nal manuscript.

Ethics approval and consent to participate
The study was approved by the local institutional review boards of both institutions, and informed written consent was obtained from all subjects.

Consent to publish
The informed written consent included consent for publication.

Figure 1
Diagram of PTC calibration including PT cardiac surrogate extraction (a) and PTC trigger detection (b).
For PT cardiac surrogate extraction (a), raw PT calibration data without RF interference is band pass ltered to enhance cardiac and respiratory signals, and raw PT calibration data with RF interference is averaged to enhance RF pulse artifacts to stabilize identi cation of the corresponding subspaces by PCA. Subspaces of unwanted signal contributions for RF artifacts and respiration are identi ed by their rst two eigenvectors. The cardiac subspace is identi ed by its rst eigenvector and unwanted contributions to the combined signal are minimized by orthogonalizing the vector with respect to both subspaces, resulting in the nal cardiac channel combination vector.
For trigger detection (b) the raw PT calibration data without RF interference is processed the same way as the continuous raw PT data stream during triggered measurements: raw PT data is combined using the cardiac channel combination vector to form the PT cardiac surrogate, then a constant velocity Kalman lter and the rst derivative are applied to generate a PTC signal with low latency which served as the output waveform seen on the scanner user interface and on which the trigger is detected. During calibration only, the polarity of the PT cardiac surrogate is determined by identifying the skewness of the Kalman outputs, since diastole has a longer duration than systole. The average over the local maxima of the PTC signal is stored for signal scaling, and the triggering threshold was set to be 0.4 times of the stored value.

Figure 2
Comparison between PTC and ECG triggered images in a healthy volunteer.
PTC triggered images were acquired including morphology, segmented cine covering the left and right ventricles, segmented ow of the ascending aorta and the main pulmonary artery (MPA), quantitative myocardial relaxation time maps. Subject has a heartrate around 90 beats per minute. Corresponding ECG triggered images were also collected using identical scan parameters and slice positions for comparison.

Figure 3
An example of PTC triggered images of a cardiac patient.
PTC triggered images were acquired including morphology, segmented cine covering the left and right ventricles, quantitative myocardial relaxation time maps, and LGE images. Subject was referred to an CMR exam for myocardial sarcoidosis. Subject was severely obese with EF less than 35%, and has a heartrate around 80 beats per minute. ECG vs. PTC signal recorded simultaneously during segmented cine acquisitions.
Triangle represents the triggering of each heartbeat. A delay was observed in the PTC trigger with respect to ECG based R-wave triggering. This was expected as the R-wave precedes the mechanical contraction of the heart, upon which the PTC trigger was based. 4a) shows an example of successful PTC acquisition where no mismatching between ECG and PTC was found. 4b) shows all cases of mismatched PTC-ECG trigger pairs. Two out of three mismatching observed were due to arrhythmias in patient. Although ECG was used as reference standard, one mismatched trigger case was apparently due to ECG failure.
Mismatched triggers are indicated by the red boxes. The average trigger delay in volunteers was 178.9 ± 9.4 ms, 20.4 ± 1.5 %RR ECG . The average trigger delay in patients was 187.6 ± 12.5 ms, 21.5 ± 1.8 %RR ECG . The trigger delay was consistent regardless of the RR interval, especially within each individual. Trigger jitter was higher in patients. An abnormally long PTC trigger delay was pointed out as outlier in (b), which was due to incorrect early detection of ECG shown in (c).