Response to letter to the editor

Dear Editor,

We appreciate the compliment of our effort from Backes et al. and their meaningful thoughts regarding the possible explanations of the results. In our study, the clinical constraints forced us to use a dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) protocol with shorter duration, especially the high possibility of severely declined image quality due to the limited tolerance to long scans for elderly participants (average age > 70 years). To fulfill the aim of our study, whole brain coverage and relatively high spatial resolution were chosen, resulting in a temporal resolution of 11.7 s, better than a commonly used protocol¹ (15.4 s) investigating subtle leakage.

We understand the concern of vascular transit effects. However, the sampling rate of our data precludes the possibility of performing the suggested deconvolution, since “a high sample rate is of paramount importance”² for deconvolution and only “time resolution between 1 and 2 s is acceptable.”² The reason is that the first pass (especially the peak) of the time courses, when the vascular transit effects are mostly obvious but still subtle, can hardly be well captured with a 5 times lower sampling rate than needed, given the reported similar arterial transit time (ATT)³ and slightly prolonged mean transit time (MTT) (1–2 s)⁴ compared with aged controls. Due to the same reason, those transit effects, which were buried in the variance introduced by low-rate sampling, should not systematically affect the results. Our study showed increased blood–brain barrier (BBB) permeability similar to that in a previous study,¹ which omitted transit effects with long duration, also supporting the effectiveness of our protocol. Moreover, a previous study⁵ investigating water exchange in multiple sclerosis (MS) patients with subtle BBB leakage scenarios using only 115 s duration also ignored the transit effects. Notably, even if the data have a sufficient sampling rate, differentiating coexisting leakage and transit effects is almost impossible by deconvolution, which assumes only transit effects exists. This is because the inevitably existing leakage can also mimic transit effects, thereby leading to biased estimation.

Furthermore, long duration is not always better for models neglecting contrast reflux to plasma, such as the commonly used Patlak model.^{1, 6} Actually, the physiological feature of BBB breakdown is the passive bidirectional BBB permeability.⁷ The bias (underestimation) introduced by neglecting reflux could be ignored with short durations but will certainly increase as duration becomes longer.⁸ The contrast in tissue may even dissipate rather than “accumulate” when the duration becomes so long that the reflux plays a major role. Cramer and Larsson⁸ found that long duration underestimates the permeability of MS lesions with subtle BBB leakage, whereas “this underestimation is attenuated if the measurement duration is reduced from 15 to 5 min.”⁸ In their simulations,⁸ compared with 15 min duration, the mean permeability values of 5 min duration using the Patlak model were comparable to, sometimes even closer to the ground truth under low leakage scenarios. Thus, an alternative interpretation of the decreasing leakage rate for increasing durations, reported by Wong et al.,⁶ is that the leakage rates were more severely underestimated along with longer durations. Consequently, the “adjusted coefficient of variation (CV)”⁶ considering the leakage rate estimated from 25 min duration data as reference, may actually severely overestimate the variance for short durations. Nevertheless, if the high possibility of a much poorer image in long-duration scans is not considered, we do agree that long duration would have better reproducibility or smaller standard deviation (SD) than short duration, because long duration yields more sampling points, which is always preferable in mathematical fitting. However, minimizing systemic bias is generally prioritized in most studies, and the variance can be mitigated by larger sample size or region of interest (ROI) averaging. ¹We acknowledge that further optimization should be done to find the best protocol with both minimal bias and good reproducibility if subject compliance can be improved. But this optimization will always be a tradeoff among imaging coverage, spatial resolution, temporal resolution, signal-to-noise ratio (SNR), model assumption, patient tolerance, and even ethical concerns. If by any chance a shorter protocol can be used, it would benefit the research and application of sophisticated techniques for BBB leakage quantification.

The “curvilinear structures” in the K^trans map should be large vessels and can also be found in a study using long-duration scan.¹ These structures were excluded in our study and would not influence the results. As for the cross-group analysis, following a precedent,¹ we pooled correlations across groups when the within-group sample sizes were small. We also provided group-stratified scatter plots for readers to interpret. Of note, as shown in those scatter plots, the correlation trends between K^trans and k_bo were not always negative. For example, positive correlation trend can be found for whole brain of aged healthy controls, whose vascular transit function also degraded.⁹ This is consistant with the positive correlation found between K^trans estimated from long-duration DCE-MRI and water exchange rate estimated from arterial spin labeling imaging in a previous study.¹⁰

The authors have nothing to report. Author disclosures are available in the supporting information.