Feasibility of ¹⁵O-water PET studies of auditory system activation during general anesthesia in children

Functional imaging of brain activation due to visual or auditory stimuli using radioactive biomarkers of blood flow or metabolism has been implemented for the first time in the 1980s [1–4]. With the advent of fMRI in the 1990s, most research in this context especially in healthy subjects shifted to this method avoiding radiation exposure [5, 6]. However, in adult patients with cochlear implants, radioactive biomarkers still retained their role in research since fMRI is not possible for safety reasons [7, 8]. In the last 25 years, the field of brain stimulation in general has witnessed an exponential growth in clinical applications—including diverse neurological, psychiatric, behavioral, and cognitive conditions—and research investigations [9]. Functional imaging using PET provided substantial evidence with respect to the mechanisms of action in these therapies [10]. Regarding auditory implants especially the two developments have occurred in parallel: (i) it has been recognized that implantation in children at an earlier age provides the best outcomes, since it takes advantage of sensitive periods of auditory development [11] and (ii) new approaches with respect to targets more central in the auditory pathway (brainstem, midbrain) have been developed [12, 13]. In consequence to the second development, functional imaging using PET and SPECT has been used as a monitoring tool helping to understand functional changes during auditory rehabilitation in adult users of novel types of implants [14–17]. However, small children with auditory implants have not yet been included in such studies due to their inability to cooperate in functional imaging while awake. This raises the question if those functional imaging studies are feasible during general anesthesia to circumvent this and how reliable such investigations would be.

Functional imaging studies during general anesthesia have been done in adult healthy subjects during auditory stimulation with words using fMRI [18–20]. Specifically, these studies revealed a reduced but maintained activation in auditory regions of the superior temporal cortex especially during light anesthesia while results were heterogeneous during deep anesthesia. Moreover, activations related to auditory stimuli in the auditory networks beyond the temporal cortex remained present at a light state of anesthesia [19]. Nevertheless, higher levels of auditory processing in the brain such as comprehension and memory were clearly impaired during anesthesia [18–20]. With respect to the used anesthetic, there is evidence from patients undergoing cardiac surgery that in opioid (fentanyl)-based general anesthesia (combined with the benzodiazepine flunitrazepam): (i) auditory evoked potentials are similar to the awake state and (ii) implicit memories of auditory stimuli can be registered in a higher proportion as compared to other combinations of anesthetics [21]. Correspondingly, studies in non-human primate showed with electrophysiological methods using auditory stimuli during opioid-based anesthesia activations of the primary auditory cortex and belt areas [22]. In accordance with these findings, opioid-based anesthesia has been recommended for intraoperative monitoring of cochlear implant function [23].

In parallel to increasing challenges for functional imaging with PET in patients receiving electrical stimulation therapy to their brain, a substantial improvement in imaging technology occurred, with respect to sensitivity, spatial resolution, and signal to noise ratio [24, 25].

Against this background, we systematically reanalyzed PET auditory activation studies of patients obtained in the context of patient care at Hannover Medical School. The studied patient population encompassed (awake) adult patients as well as a group of (anesthetized) children receiving functional ¹⁵O-water PET with auditory stimulation during general anesthesia. Besides generating for the first time data on the feasibility of this approach (PET auditory activation studies during general anesthesia), we intended to optimize the procedure.

There are several issues with respect to PET acquisition and data analysis, in which different parameters have been suggested for activation studies with ¹⁵O-water. These parameters are selected empirically without systematic reasoning and differ from one PET center to the other. One parameter is the acquisition duration. Early studies used relatively short durations of 40 s [2, 4]. In many studies up to now, intermediate durations between 60 and 90 s are employed [14, 17, 26–31]. Working groups from Japan often use 120 s acquisition time [8, 32–34].

A further issue is smoothing of the three-dimensional (3D) data set during preprocessing. Filter kernels between 1 and 2 times of the spatial resolution (FWHM, full width at half maximum) of the used PET scanner (7–12 mm) tended to be more frequently employed [15, 17, 27–31] as compared to kernels between 2- and 3-fold the FWHM (15–20 mm) [14, 35–37].

Finally, the cut-off used for statistical inferences from statistical parametric maps is variable. Although a p value less than 0.001 uncorrected for multiple comparisons is frequently employed [14, 15, 29, 38], less rigid thresholds have been applied as well [28, 30, 31, 37].

Consequently, the aims of the present study are to (i) explore the feasibility of functional imaging studies of the auditory system with PET in children during general anesthesia, (ii) assess the impact of the acquisition duration after ¹⁵O-water application on the statistical outcome measures, and (iii) elaborate adapted parameters for ¹⁵O-water PET studies of auditory cortex activation with respect to pre-filtering and cut-off for statistical inferences — in particular for impaired activations during general anesthesia.

The functional performance after cochlear implantation can vary between simple sound detection and very good speech intelligibility [42]. The underlying mechanisms in the individual case often remain unclear. Furthermore, in the situation of unilateral implantation, the potential benefit of a second implant can be difficult to predict [11]. Therefore, objective functional measures reflecting individually either intactness of the central auditory pathway or correlates of central auditory processing are desirable. ¹⁵O-Water PET has been employed for that purpose using stimuli presented either via auditory implants or promontory needle electrodes [42, 43]. The necessity of an objective measure is given all the more in patients not able to cooperate in testing procedures of auditory function like small children.

Therefore, we investigated the feasibility of ¹⁵O-water PET activation studies of the auditory system during general anesthesia. During a flat opioid-based EEG conducted general anesthesia, we could demonstrate in children activations of the auditory cortex during stimulation via a promontory needle electrode or a cochlear implant. However, when applying the same parameters for PET data analysis as in awake adults, activations during anesthesia were less frequent and significant (with respect to peak height and extent). This observation corresponds to findings in healthy subjects using fMRI [18–20]. Nevertheless, it would be desirable that diagnostic accuracy of auditory activation studies during anesthesia at least partially approaches that of studies in waking state.

The present data indicate that specifically adapted parameters for PET images are helpful in this regard. Particularly, similar sensitivity can only be obtained if the cut-off for statistical inferences is lowered for studies during anesthesia. Typically, for single-subject analyses of PET studies in awake patients, an uncorrected p value less than 0.001 is used for cut-off [14, 44]. Other investigators have suggested less rigid cut-off values of p < 0.05 or p < 0.01 particularly for studies of the auditory system [37, 38]. One argument in favor of this is that the open hypothesis with respect to the location of detected activations typically associated with a level of p < 0.001 is not fitting to the situation of an auditory activation study with a high likelihood of an activation predominantly in the auditory system. We observed a similar sensitivity in the detection of auditory cortex activation during anesthesia compared to waking state using a cut-off level of p < 0.05 or p < 0.01 instead of p < 0.001.

Furthermore, the present study indicates that in activation studies during anesthesia, a relatively high degree of smoothing (e.g., filter kernel equivalent to 2- to 3-fold the FWHM) results into considerably less frequent “activations” in extra-auditory regions (Table 4, Fig. 2). These extra-auditory “activations” during anesthesia and stimulation via promontory needle electrode or CI (music) are unexpected and likely to be primarily artificial [18, 19, 45]. On the other side, adequate filtering to detect the expected signal in the auditory cortex (despite narcosis) might be achieved by setting it according to the matched filter theorem. This theorem suggests that the optimal smoothing filter kernel should match the spatial size of the signal to be detected [46, 47]. In auditory activation studies, the main structure in which a signal should be detected is obviously the primary auditory cortex and it should be possible to separate it from the secondary and association areas. The extent of the primary auditory cortex (Heschl’s gyrus, Brodmann area 41) might be assumed to be about 20 to 30 mm — despite a considerable inter-individual variation [48, 49]. Thus, matching the filter kernel to the extent of the primary auditory cortex according to the filter theorem results again into a relatively high degree of smoothing (equivalent to 2- to 3-fold the FWHM of the scanner used in our study). Employing this in the present study produced similar frequencies of auditory cortex activations with and without anesthesia (with the respective cut-offs). Higher frequencies of auditory cortex activations during aesthesia achieved with less smoothing filter kernels are likely to reflect in part false positive findings. This is further illustrated in Fig. 2 [0 mm]³ displaying activations in auditory and non-auditory regions appearing as comparable “noise” consisting of small accidentally “significant” clusters. Further evidence that the detection of a diagnostically relevant signal (in that case of reduction) is not hampered by stronger filtering comes for studies in neurodegenerative diseases. In these, it has been demonstrated for a large range of filter kernels (between 8 and 18 mm), that a higher degree smoothing had only a very small impact on the detection of the relevant signal, i.e., diagnosis [47]. Nevertheless, it has to be kept in mind that with very large filter kernels (e.g., [30 mm]³, also evaluated in the present study), signals might be compromised. This is illustrated by consistently lower mean t_max values observed with [30 mm]³ compared to [20 mm]³ for both acquisition durations in all subgroups (see Table 2). Moreover, too large filter kernels might be detrimental with modern generations of PET scanners — providing in particular at least twice as much spatial resolution compared to the PET scanner used in the present study. These modern scanners enable sufficient recovery of signal for substructures of the auditory system of half of the size compared to the scanner used in the present study, which could be lost by excessive smoothing.

As mentioned above, another issue potentially influencing the accuracy of ¹⁵O-water PET activation studies of the auditory system is scanning duration. Shorter as well as longer acquisition durations have been proposed. Kanno et al. [50] demonstrated an increase of signal to noise (S/N) ratio with acquisition time. In the mean, they found S/N ratio to be about 1/3 higher at 120 s compared to 40 s scan time. On the other hand, Volkow et al. [51] showed the largest activations in the occipital cortex due to light stimulation based on acquisitions during “uptake phase” (the initial 30–35 s). Moreover, they demonstrated that activations depend on the duration of stimulation. Continued stimulation after the peak of uptake phase for further 40 s during the “washout phase” resulted into lower uptake compared to stimulation restricted to the uptake phase. The authors explained this observation and also lower activation with longer acquisitions by increased clearance of radioactivity from areas of high blood flow. In our study, we did not observe a significant difference in activation height (T_max) dependent on the acquisition duration (60 vs. 90 s). This is particularly in contrast to a study by Silbersweig et al. [52] reporting improved significance (greater signal to noise ratio and higher Z-score) of results obtained with 90 s compared to 60 s. One reason for these conflicting results might be differences in the injection procedure—particularly with respect to the duration of the bolus injection. Silbersweig et al. [52] described a slow bolus injection (over 20 s), which is considerably slower compared to Volkow’s and our study (3 and 7 s, respectively). A slower injection might promote improved results with longer acquisition durations. Thus, a final conclusion with respect to the most favorable scan duration is not yet possible based on the presently available data. Nevertheless, future studies employing modern PET equipment with the capability of list mode acquisition enabling flexible retrospective definition of reconstructed time frames would allow to address this issue more precisely. Furthermore, due to achievements in PET technique like time of flight measurements, an enhanced signal to noise ratio for a given acquisition duration can be expected [24]. Finally, this will help to adapt the acquisition duration rather on biological/physiological grounds than on technical limitations of the scanner for improved results of PET activation studies.

Some limitations of our study have to be considered. First, that it is a retrospective study. A prospective study would have offered the possibility, e.g., to select age-matched patients studied with and without anesthesia. Moreover, auditory stimulations were performed routinely on the left side first (without randomization); therefore, an order effect cannot be excluded. A further limitation is caused by the use of the standard SPM template for adults for spatial normalization in children. Muzik et al. [53] explored this issue. They found acceptable errors in children down to 6 years. In younger children, mean deviations of the brain contour around 2–3 mm have to be expected. However, this might be more relevant in the localization of epileptogenic foci (as in Muzik’s [53] patient population) as compared to more extended auditory cortices. Furthermore, some beneficial effect might be expected due to higher degrees of smoothing as applied in the present study. And, finally, at least visual inspection of the normalization result in all cases of the present study did not reveal any artifacts. Furthermore, with respect to the impact of the acquisition duration, it would have been desirable to compare more than two time periods. However, this was unfortunately not possible retrospectively since our scanner did not allow performing list mode acquisitions, and therefore, we had to stick with the originally used frame sequence. At last, an impact of the use of CI’s from different vendors could not be completely excluded. However, due to the facts that music was used as a robust stimulus not that dependent on the speech processor and most of the implants reflected the technique available during a relatively short period between 2001 and 2007, a major impact is not expected. Despite the abovementioned limitations, we considered it as an added benefit to obtain insights regarding the feasibility of auditory activation studies with ¹⁵O-water PET during anesthesia based on available data obtained on clinical grounds.

The present data indicate that auditory activation studies with ¹⁵O-water PET in children during general anesthesia are feasible. They can approach results of studies obtained during waking state in adults if an adapted less rigid cut-off for statistical inferences (p uncorrected < 0.01 or p < 0.05) is employed and strong smoothing during preprocessing (filter kernel ≥ [20 mm]³) — like in studies without anesthesia. Nevertheless, further research is needed to ascertain whether activation studies with ¹⁵O-water PET in children during general anesthesia are a valuable diagnostic tool before and after auditory implantation in clinical routine situations. The issue of optimal scan duration is still not clarified on the basis of the presently available data. However, modern PET scanners providing fundamentally higher signal to noise ratios — in comparison to the generation of PET scanners employed to elaborated procedures for auditory activation studies with PET decades ago — will open novel possibilities to explore the circuitry of the auditory system in implant users.