Appendix
Data acquisition
We developed software capable of reading, decoding, analyzing, modifying and encoding Inveon list-mode files, as described previously(1). We used this software to decode all gating and time packets included in the list-mode file for export to a CSV file, in order to remove contained respiratory trigger events and to encode the new ECG-based respiratory trigger events. ECG- and respiratory-signals (RRS) were registered using a dedicated physiological monitoring system (BioVet; m2m Imag. Corp., Cleveland, OH, USA). The BioVet system analyzed the ECG and respiratory signals in on-line mode and produced trigger events as input to the PET scanner, which encoded them into the PET list-mode data stream. In addition the BioVet system allowed storage in a CSV file of the entire ECG and respiratory signals with a sampling rate of 1 ms together with the trigger events.
Data preparation
Visualization and analysis of the trigger quality and processing of the ECG- and respiratory-signal was provided thorough MATLAB (The MathWorks, Natick, USA) software developed in-house. The cardiac trigger events stored in the BioVet’s CSV file and the CSV file extracted from the PET list-mode data were synchronized for further processing as described previously(1). The following sections describe the essential processing steps for generating ECG-based respiratory trigger events. The proposed algorithm exploits the modulations of the ECG signal which are assessable by finding fluctuations in the ECG-amplitude and the time-of-occurrence of the cardiac beats. The processing parameters reported in these various steps (e.g. filter frequencies, etc.) were chosen based on earlier experience, and set as identical for all animals. Illustrations in figures 1 and 2 show a 20- and a 5-second period of time from the measurement of animal 1 in table A.1. ECG signals were initially corrected for bias by filtering with a high-pass frequency filter with a cutoff frequency of 5 Hz. A global bias correction of the respiratory signals (RRS) was performed by subtracting their global mean, and noise caused by signal spill-over from the ECG was suppressed by applying a moving average filter of 51 ms width. An example of the data resulting after preprocessing is depicted in figure A.1 (top: ECG; bottom: RRS).
Peak detector
Trigger events were generated on each R-peak in the ECG signal and on each maximum in the RRS signal. The peak detector algorithm was designed to be simple and robust. It was based on finding upward peaks in a signal exceeding a certain threshold. Since biases were already removed from both signals during preprocessing, we defined the threshold for the peak search from each signal’s global mean plus 1.5 times its standard deviation. Any maximum in a signal found above the corresponding threshold was considered as a valid peak (i.e. R-peak, or end-inspiration) if two more conditions were fulfilled: the signal following the maximum dropped about 1.5 (ECG), 2.0 (RRS) or 0.5 (ERSt and ERSa) times the standard deviation of the global mean, and the previous peak was found at least 30 ms (ECG) or 100 ms (RRS) prior to that event, as is implemented as the so-called “inhibit” in many on-line peak detectors. Hereby artefacts or artificial TEs (e.g. small maxima in signals between the true respiratory peaks) are excluded by using the inhibit during the processing of the respiratory signal.
ECG-based respiratory curves
Figure 1 illustrates the basic signals which were generated and processed in this work. Clearly noticeable are the respiration-induced changes of R-peak amplitude (fig. 1, top, red line) and the almost simultaneously occurring alternations of the cardiac cycle period (fig. 1, middle), both in approximate synchrony with the measured respiratory signal (fig. 1, bottom). This figure demonstrates that two surrogate signals could be generated from the detected ECG trigger events to estimate ECG-based respiratory signals: the first consisted of the amplitudes of the R-peaks in the ECGs, the second consisted of the duration of each cardiac cycle. The sequences of R-peak amplitudes and cardiac cycle periods were extracted, normalized by division by a moving average filter with a window width of 51 ms so that signals varied around a value of 1, and interpolated using a cubic spline function with 1 ms sampling rate to generate the two ECG-based respiratory signals ERSa (amplitude) and ERSt (time or duration). Both signals were subtracted by their global mean to receive signals varying around a value of 0. If necessary a signal was inverted so that the respiratory peaks associated with the end-inspiration phase pointed upward.
ECG-based respiratory triggers
The above described peak detector was applied to both signals to generate the ECG-based respiratory trigger events. Figure 2 illustrates the final ECG-based respiratory signals ERSa (top), ERSt (middle) and the original respiratory signal RRS (bottom) from the belt measurements. For each signal the peak detector’s threshold and the resulting trigger events are superimposed. As final step the time of each ECG-based respiratory trigger event was stored in a list in CSV file format. The respiratory gating packets which were contained in the original PET list-mode file were removed and the new ECG-based gating packets encoded to allow for reconstruction of ECG-based respiration-gated images.
Representative ECG-based respiratory triggers processing of human data
We acquired representative ECG and respiratory signals from a healthy young human subject using a clinical ECG trigger device (101NR patient monitor, Ivy Biomedical Systems, Branford, CT) and respiratory trigger device (AZ-733V, Anzai Medical Co., Ltd, Tokio, Japan). The signals from both devices were collected by software written in LabView (National Instruments, Austin, TX) and stored in CSV files similar to the methods described above. Processing of these data was performed as above for the generation of ECG-based respiratory trigger events, with slight adaptions of the peak detector criteria as required for this specific setup: in particular, the signal drop following a maximum had to be 2.0 times the standard deviation of the global mean for the RRS, and the inhibit durations were set to 300 ms (ECG) and 500 ms (RRS). Since a paradoxical respiratory sinus arrhythmia had been observed in isoflurane anaesthetized rats, the ERSt signal had to be inverted prior to searching for respiratory peaks in awake human data.
In analogy to figure 1, figure 3 depicts the ECG and respiratory signals of the human subject for a period of 120 seconds. Figure 4 illustrates the final ECG-based respiratory signals ERSa (top), ERSt (middle) and the original respiratory signal RRS (bottom) from the belt measurements. The peak detector’s threshold and the resulting trigger events, which were utilized for retrospective respiratory gating of PET list-mode data, are superimposed.
Figure 1. Illustrative 20 second recordings of the physiological signals of rat 1 as investigated in this work. Top: ECG signal (green), cardiac trigger events (blue dots) and spline interpolation of cardiac trigger event amplitudes (red). Middle: Spline interpolation of cardiac cycle periods. Bottom: Respiration signal from belt measurements.
Figure 2. Final respiratory signals of rat 1 based on ECG-amplitude (ERSa, top) and cardiac cycle period (ERSt, middle) and the original respiratory curve from belt measurements (bottom). Peak detector thresholds are shown as blue lines, the corresponding trigger events as blue dots.
Figure 3. Illustrative 120 second recordings of the physiological signals of the human subject. Top: ECG signal (green), cardiac trigger events (blue dots) and spline interpolation of cardiac trigger event amplitudes (red). Middle: Spline interpolation of cardiac cycle periods. Bottom: Respiration signal from belt measurements.
Figure 4. Final respiratory signals of the human subject based on ECG-amplitude (ERSa, top) and inverted cardiac cycle period (ERSt, middle) and the original respiratory curve from belt measurements (bottom). Peak detector thresholds are shown as blue lines, the corresponding trigger events as blue dots.
Table 1:
Mean cardiac and respiratory cycle period are presented along with the corresponding standard deviation (SD) for all 15 rodents. For respiratory cycle period the relative standard deviation (RSD) is presented as well. Inadequate triggered datasets (marked in red) were not used for calculating the overall means and standard deviations in the bottom row. Blank cells indicate cases for which a reasonable delay could not be calculated.
Table 2:
Mean cardiac and respiratory cycle period are presented along with the corresponding standard deviation (SD) in seven healthy volunteers. For respiratory cycle period the relative standard deviation (RSD) is presented as well.
Table 3:
Estimates for EDV, ESV, SV and EF are presented as mean ± standard deviation for respiratory gate 1, 2 and 3 in a subset of animals (N=8) which demonstrate the effects on the heart during breathing. From gate 1 to gate 2 volumes increase in the expiration phase while decreasing again during inspiration from gate 2 to 3, as discussed in the manuscript. Due to the motion within the gates 1 and 3 (see also figure 1 in the manuscript) data seems to be less reliable then data from gate 2 due to artifacts. Thus we opted for gate 2 for further evaluation of the respiratory trigger signal.
For detailed information regarding the triggering algorithm please contact:
Dr. Guido Boening
Department of Nuclear Medicine, University of Munich
Marchioninistr. 15, 81377 Munich, Germany
+49-(0)89-4400-74657 (phone)
+49-(0)89-4400-74676 (fax)
(1)Boning G, Todica A, Vai A, Lehner S, Xiong G, Mille E et al. Erroneous cardiac ECG-gated PET list-mode trigger events can be retrospectively identified and replaced by an offline reprocessing approach: first results in rodents. Phys Med Biol 2013;58:7937-59.