EF is one of the most important parameters to characterize cardiac function, and its "gold" standard measurement method is invasive ventriculography. In contrast, the commonly accepted "gold" standard of the non-invasive method is cardiac magnetic resonance(CMR). The former is invasive and costly, while the latter is limited in clinical use because of its complicated techniques and many contradictions. As a non-invasive examination, 99mTc-RBC MUGA using radionuclide to label autologous red blood cells is equivalent to simulating the change of blood volume in the ventricle, which is the closest quantitative method to the concept of EF. Tomography can reconstruct left and right ventricles in 3D mode, which is undoubtedly much closer to the complete standard measurement mode.
In this study, the LVEF obtained by the two methods of the two cameras had excellent correlation and no significant statistical difference, which was coincide with the previous studies[15–17]. The correlation of RVEF was good but worse than that of LVEF (r value: 0.619 vs 0.831), and there was a statistically significant difference between them. This result was consistent with previous studies[17], but in which Chen et al. thought that the reason for this situation was that the algorithm of QBS software was inaccurate due to the influence of LV size and shape when identifying RV boundary in CZT-SPECT, which leaded to the reduction of correlation with the planar method. Although It could be part of the reason, the influence of RV in the planar method seemed to be more obvious, such as it was more difficult to accurately identify the position of the RV pulmonary trunk, which tend to cause more errors in the delineation of the RV basal part than that in the CZT-SPECT tomographic method. Therefore, concerning the correlation and difference of RVEF, we thought that the reason based on the C-SPECT planar method may be more significant, which can be indirectly proved by the results of repeatability of their respective methods in the later part. In addition, previous studies comparing CZT-SPECT tomographic imaging with CMR, RVEF, and right ventricular volume had a reasonable correlation[16, 18–19].
There are few studies on PER and PFR in tomographic MUGA with CZT-SPECT, which may be related with that there is no public-recognized "gold" standard for these two parameters and CZT-SPECT myocardial perfusion imaging (MPI) can also provide many functional parameters of left ventricles, such as PER and PFR, and MPI is more convenient and popular than MUGA, so there are many studies on this aspect in recent years[20–21], especially the study of left ventricular systolic and diastolic function under stress[22], but there are few studies on right ventricular PER and PFR in various examination techniques until now. In our study, the LVPER and LVPFR obtained by the two cameras and methods had a reasonable correlation (r = 0.672 and 0.700), but the correlation was worse than LVEF, which was still within our expectations. Compared with LVEF obtained from 2 frames of EDV and ESV targeted images, the cumulative deviation of LVPER and LVPFR obtained from 16 frames of reconstructed count (volume)-time curve will be more significant. However, the correlation between RVPER and RVPFR was only 0.463 and 0.253 between the two methods, and the RVPFR between the two methods had a noticeable statistical difference, which was lower than our expectation. Therefore, we compared RVPER(P), RVPFR(P), RVPER(T) and RVPFR(T) with RVEF, RVEDV, and RVESV, respectively, and got the trend that these four parameters were positively correlated with RVEF and negatively correlated with RVEDV and RVESV, which was in line with the essential physiological function of the heart and excluded the possibility of human error in data processing by the observers. Even so, we could see that the correlation of RVPER(P) and RVPER(T) in each group was good or general, and the correlation of RVPFR(P) and RVPFR(T) in each group was only general or poor, even RVPFR(P) had no apparent correlation with RVEDV and RVESV. To sum up, the deviation of measuring RVPER and RVPFR both in C-SPECT and CZT-SPECT showed evident uncertainty and no clear rule. Sometimes the former measured too much, and sometimes the latter measured too much, which was particularly obvious in RVPFR, and the deviation value was significant. The above may indicate that we need a third-party comparison. However, before that, we are unsure whether it was related to our sample selection because of this low correlation, so we thought it may be necessary to have a group of samples with completely normal right heart or none-cardiovascular diseases to be the normal group for control.
In terms of repeatability, the intraclass correlation coefficient of RVPFR(P) was good (ICC = 0.698), while the intraclass correlation coefficient of other groups of data was excellent (0.823 ~ 0.989) which was similar to previous studies[13, 15, 17, 23]. It could be drawn from the Bland-Altman diagram that the repeatability confidence intervals of LVEF and RVEF of CZT-SPECT were narrower than those of C-SPECT, indicating that CZT-SPECT had better repeatability. The repeatability confidence interval of LVEF of the two cameras was narrower than the corresponding confidence interval of RVEF, which showed that the repeatability of LVEF was better than RVEF. The repeatability confidence intervals of LVPER and LVPFR of the two cameras and methods were not much different in width, the LVPER group was slightly better than the corresponding LVPFR group, and the RVPER and RVPFR were similar. However, the repeatability confidence intervals of the four parameters in the LV group were all narrower than those in the RV group, which showed that the repeatability of LVPER and LVPFR in both cameras was better than that of RVPER and RVPFR. As the repeatability of PER and PFR in left and right ventricles in their respective cameras and methods was good, combined with the problem of imperfect correlation between the above two cameras and methods, it further showed that we need to compare third-party data or establish CZT-SPECT reference standards for PER and PFR. The left and right ventricular volume parameters of CZT-SPECT were not compared with the reference group but evaluated the number of cases outside their confidence intervals, which was considered to be still within the clinical acceptance range (5.2% ~ 7.8%).
In this study, coefficient of variation (CV), an index used in previous studies by Jensen et al. [15, 23], was introduced, but the definition of CV in their studies seemed to be different from the conventional definition, and CV was regarded as an index reflecting the discreteness of a group of data. When the standard deviation (SD) is not suitable for direct comparison between two groups of data, CV can be used for comparison instead. It is defined as the ratio of standard deviation to the average value (SD/AV) of this group of data. However, in Jensen et al.' s research, CV value was obtained by dividing the standard deviation of the difference between paired data groups by the average value of the paired data groups, that was, standard deviation (A2-A1)/ average value ((A2 + A1)/2). They used the CV value obtained to evaluate the repeatability of each paired data set. For example, they obtained LVEF, RVEF, left and right ventricular volumes measured by observers 1 and 2, to calculate the CV values by the above method, then they used the CV to evaluate the repeatability of the single parameter and also compared with other parameters.
In our study, the corresponding CV values of each group were obtained by this method, and the trends of EF and volume parameters of left and right ventricles were roughly the same as those in Jensen et al.' s study, but the values were higher, which was caused by the difference in case sample selection. They chose tumor patients without cardiovascular diseases, and their heart function was normal. All of our samples were patients with cardiovascular diseases, and there were many patients with a severe decrease in heart function caused by ischemic cardiomyopathy, dilated cardiomyopathy, or heart failure due to other reasons. In our results, the CV values of LVEF(P), LVEF(T), RVEF(P), and RVEF(T) were 11.1%, 6.7%, 14.9%, and 9.5%, respectively, which were significantly lower in CZT-SPECT group than in C-SPECT group, while LVEF group was lower than corresponding RVEF group respectively. The above was also indicated in the Bland-Altman diagram, and here we could see that the CV value of RVEF in the CZT-SPECT group was even lower than that of LVEF in the C-SPECT group, indicating that CZT-SPECT was better than C-SPECT in terms of repeatability among observers. The CV values of LVEDV(T), LVESV(T), LVSV(T), RVEDV(T), RVESV(T) and RVSV(T) were 6.1%, 10.2%, 14.1%, 16.0%, 23.4% and 14.7%, respectively. The results showed that the repeatability of left ventricular measurement was better than that of the right ventricle. The first reason is that the right ventricular boundary recognition, especially the right atrial ventricular boundary and pulmonary trunk location, mentioned above, and it has always been a problem in cardiac blood pool imaging, from planar imaging to tomographic imaging. Another reason is that the repeatability for small chambers is worse for larger chambers, which was also mentioned in Jensen et al.' s research. Most of the samples in this study had normal right heart function, and some right heart chambers were relative smaller, which also had a certain influence on the measurement repeatability. In addition, most of these samples had serious left ventricular involvement and enlarged heart chambers; the ratio of left and right cardiac chambers was increased. The right ventricular recognition was more obviously affected by the enlarged left ventricle, which further affected the repeatability of right ventricular measurement. However, the effects for RVEDV and RVESV were closely similar, so the CV values of RVSV and LVSV were very close, which explained why the repeatability of RVEF(T) was not significant compared with LVEF(T)(CV value: 9.5% vs 6.7%).
For PER and PFR, the CV values of LVPER(P), LVPFR(P), LVPER(T), and LVPFR(T) were 12.6%, 17.2%, 11.6%, and 15.7%, respectively, and the values of RVPER (P), RVPFR(P), RVPER(T) and RVPFR(T) were 18.6%, 25.3%, 18.6% and 21.7%. PER group was better than PFR group, and LV group was better than RV group. There are 8 cases for observer 1 and 7 cases for observer 2 of LVPFR(T) in the processed data, respectively, which could not be obtained(4 same cases could not be obtained both). RVPFR(T) could not be obtained in 5 cases for observer 1 and 6 cases for observer 2(3 same cases in both). The software identifies it as NA, which meant unrecognized or invalid data. The reasons for this situation were not mentioned in previous studies and software instructions, and we cannot exactly explain these reasons either. By observing and analysing the volume(and filling) -time curve(Fig. 1) of each sample, we tried to find the potential trend to explain the NA data emerged. We preferred one possibility based on distinguishing the early peak filling rate(PFR) and late peak filling rate(PFR1) by BPGs software. In the volume(and filling) -time curve, BPGs recognized the maximum negative value of the filling curve, which obtained from the volume curve, as PER(the deepest trough in the filling curve), then the following two positive peaks (peak 1 and peak 2) would be further identified as the value points of PFR and PFR1 (or only identified PFR if only single peak). We found that most of these NA data have the following characteristic: there was a trough with negative value between peak 1 and peak 2, and the absolute value of this trough was not much different from peak 1 or significantly larger than peak 1. In the few of these NA data, there might be another characteristic: peak 1 and peak 2 were not significantly different in absolute value, and there was a less obvious trough with positive value between them. The cause of this waveform of the filling curve should be related to the sample we selected, based on the waveform of the volume curve disorder caused by significant impairment of ventricular diastolic function in patients with severe cardiovascular disease. Compare with CZT-SPECT, there was no NA data of PFR in C-SPECT, which reason might be, the processing of planar PFR was relatively simple, and only the highest peak in the filling curve in the whole diastolic process would be selected. According to the different calculation methods and principles, we chose the larger one between PFR and PFR1 of CZT-SPECT mentioned above as the PFR data for the comparative analysis, to match the calculation mode of C-SPECT planar method. In any case, this finding in our study suggested that PFR parameters obtained by CZT-SPECT might be not a stable parameter for monitoring ventricular diastolic function, but PER was much more stable, and it was also reported from our analysis that its repeatability was better than PFR. However, previous studies had shown that the diastolic dysfunction of the ventricle was earlier marker than the systolic dysfunction for heart diseases[24]. Therefore, we need further study in processing data with BPGs or other relative software in order to obtain stable PFR parameters.
Study limitations
This study had some limitations. Firstly, due to the limitation of clinical situations, there was no CMR and/or left ventricular angiography results as a "gold" standard for comparison; Secondly, the diastolic function was still lacking "gold" standard reference techniques. In addition, due to ethical restrictions, there was no normal group with non-cardiac diseases. However, as far as we know, in CZT-SPECT tomographic MUGA studies published so far, this study included the most significant number of samples of patients with cardiovascular diseases.