The records of 83 patients who underwent MVD surgery for typical HFS at our hospital between January 2018 and December 2019 were retrospectively reviewed. Brain perfusion was assessed by preoperative DSC MRI and neurovascular conflict around CN VII was identified by axial three-dimensional (3D) proton-density volume isotropic turbo spin-echo acquisition (VISTA) sequence in 60 of these patients. The patients’ demographic data, such as age, sex, and side of spasm, were collected and analyzed. The patients were divided into two groups according to their postoperative course : group A consisted of 32 patients (53.3%) who had early recurrence, which means their spasm disappeared right after surgery but reappeared within four days and lasted for a long time; group B consisted of 28 patients (46.7%) who had no early recurrence, which means their spasm resolved completely after surgery and did not recur during the follow-up period. The demographic characteristics of the participants with early recurrence and no early recurrence in the HFS patients were summarized in Table 1. We also evaluated the occurrence of delayed reliefs.

MR images were obtained using a 3.0 Tesla MRI system (Ingenia, Philips Healthcare, Best, the Netherlands) with the 32 channer sensitivity encoding (SENSE) coil. The MRI protocol for HFS patients in our institute hospital consisted of the following sequences : sagittal 3D T1-weighted (T1W) image, axial 3D time-of-flight, axial 3D proton-density VISTA, axial 2D T2-weighted turbo spin-echo, axial fluid-attenuated inversion recovery (FLAIR), axial DSC MRI, and axial contrast-enhanced 3D T1W VISTA. The sagittal 3D T1W image was acquired using the turbo-field echo sequence with the isotropic voxel size for evaluation of brain lesions, image registration, and brain tissue segmentation with the following imaging parameters : repetition time (TR)/echo time (TE)/inversion time (TI)=8.1/3.7/1013 ms; flip angle=8°; matrix=236×236; and voxel size=1×1×1 mm³.

The DSC MRI image was obtained with a 2D axial single-shot gradient-echo echo-planar imaging (EPI) sequence with following parameters : TR/TE=1800/32 ms; flip angle=35°; field of view (FOV)=210×210; acquisition matrix=132×135; reconstruction matrix=256×256; acquisition voxel size=1.67×1.74×5.00 mm³; reconstruction voxel size=0.92×0.92×5.00 mm³; slice thickness=5 mm without gap; number of slices=24; SENSE factor=3; EPI factor= 45; number of average=1; number of dynamic scans=130; and total scan duration=4 minutes. Gadolinium-based MRI contrast agents with a standard dose of 0.1 mmol/kg were injected intravenously at 4.5 mL/sec.

DSC MRI data were processed using nordicICE software (version 4.1.2; NordicNeuroLab, Bergen, Norway). A motion correction was applied to minimize artifacts from participant’s movement. The temporal smoothing, which was a low-pass filter, was performed using Gaussian kernel of 2×2×2 pixels to reduce effects of noise and spikes in the dynamic signal response. The signal change was converted to relative contrast agent concentration. The resulting concentration time-curve was used or further analysis using an automatic detection of an arterial input function. We used a standard singular value decomposition (SVD) method with iterative thresholding Tikhonov regularization [30,44]. The relative CBV (rCBV) map referenced to white matter tissue was estimated by using residue area-under curve. The leakage (extravasation) correction was performed to correct both positive (T2 effect) and negative (T1 effect) effects with the parametric residue leakage fitting method [2,3,42] and to map leakage or contrast agent permeability (K^tran) map. Output maps were rCBF, rCBV, relative mean transit time (rMTT), leakage or K^tran map (1/min), EF, and mean baseline signal intensity maps.

To analyze the DSC maps on the standard brain template space, the following processing was performed using Statistical Parametric Mapping-version 12 (SPM12) software (Wellcome Department of Imaging Neuroscience, University College, London, UK). First, the 3D T1W image and the mean baseline signal intensity map for each participant were co-registered. As a result, all DSC maps were also co-registered into 3D T1W image. Second, the 3D T1W image was segmented into gray matter and white matter using the CAT12 tool (http://www.neuro.uni-jena.de/cat/) to obtain brain tissue information and was spatially normalized to the standard brain template provided in the CAT12 tool. All DSC maps were also spatially normalized into the standard template using the deformation field information of 3D T1W. Finally, Gaussian smoothing using a full-width at half maximum of 10×10×0 mm³ was performed for the voxel-based statistical analysis of all maps, including the gray matter (GMV) and white matter (WMV) volumes.

No significant difference in age (p=0.602) or sex (chi-square=0.417, p=0.519) was found between the two participant groups. Right side operations (Lt/Rt, 10/22) had a very significant higher occurrence of reappearance compared to no reappearance (16/12). The demographic data of the participants and the results of the statistical analysis were summarized in Table 1. Although no significant difference for either age or sex was found between the two groups, we used both parameters as control factors when we performed data because perfusion parameters are usually depended on age and sex in human.

Of the patients included in our study, 32 patients experienced HFS recurrence more than a day after MVD surgery, despite having immediate symptom relief. The mean time to symptom recurrence was 2.6 days after surgery. Of these 32 patients, 23 patients had their symptom disappearance during the follow-up period, with a mean time of symptom resolution of 2.7 months (1 week to 12 months) postoperatively. The other nine patients did not see any improvement in their recurrent symptoms during the follow-up period, although eight of them had milder symptoms than before surgery. One patient’s symptom recurred on the fourth day after surgery, and remained the same as before surgery despite 33 months of follow-up. In addition, six out of nine patients with persistent symptoms were lost to follow-up before 1 year had passed, so it was not possible to know whether their symptoms had improved later. There were no statistical significances between symptom recurrence and hypertension and diabetes, which are related factors of HFS (p=1.000 and p=0.254, respectively).

Fig. 1 shows representative DSC maps obtained from two male HFS patients : one with reappearing symptom (58-year-old) and one without reappearing symptom (56-year-old) after the MVD surgery. DSC maps provide detailed information of brain structures. The patient without reappearing symptom had higher rCBF than the patient with reappearing symptom. The patient with reappearing symptom had higher EF than the patient without reappearing symptom.

Fig. 2 shows results of the voxel-based independent t-test analysis between HFS patients with and without reappearing symptom groups after the MVD surgery for each DSC map. In the most of brain areas, especially in the cerebellum culmen, cingulate gyrus, and precuneus, rCBF was reduced in the symptom reappearance group. rCBV was also reduced in the symptom reappearance group, especially in the cerebellum declive and brainstem. However, rMTT was increased in the symptom reappearance group in the most of brain areas, especially in the supramarginal gyrus. EF was also increased in the symptom reappearance group, especially in the medial and superior frontal gyrus, and temporal lobe. Leakage, GMV, and WMV maps did not differ significantly between the patient groups. Supplementary Table 1 lists the detailed locations of the significant differences between the two patient groups for rCBF (A), rCBV (B), rMTT (C), and EF (D).

Table 2 lists the result of the independent sample t-test between HFS patients with and without reappearing symptom groups after the MVD surgery for rCBF values in the specific brain areas. The rCBF value was significantly decreased in most of the selected brain areas in the reappearing group compared to the no-reappearing group, except the amygdala and supramarginal gyrus. Results of the independent sample t-test for other DSC indices were summarized in Supplementary Table 2 for both rCBV and rMTT and Supplementary Table 3 for both leakage and EF. In all selected areas, rCBV and rMMT values did not differ significantly between the two groups. Only in the supramarginal gyrus, both Leakage and EF were significantly different between the two groups.

One of the main findings of our study was that early recurrence patients had significantly lower rCBF than no early recurrence patients in most brain regions. This suggests that the neurovascular compression of the facial nerve, which is the main cause of primary HFS, impaired the perfusion of the brain tissue [12]. The offending arteries that compress the facial nerve root entry zone (REZ) may include the anterior inferior cerebellar artery (AICA), posterior inferior cerebellar artery (PICA), vertebral artery (VA), or multiple offending arteries. The recurrence of HFS after MVD surgery may be contributed by several factors, such as : the blood flow in their supply area may be reduced by the cumulative effect of compression by multiple perforator arterioles [38]. Reduced blood flow and recurrence may be led by the inappropriate insertion of Teflon, which may fail to absorb the pulse energy or induce local inflammation due to direct compression of the facial REZ [23]. The contact area and demyelination of the nerve may be increased by the ventrocaudal location of the offending vessel, resulting in a wider range of nerve dysfunction [28]. The degree of interaction between offending vessels and nerve may be affected by these factors, which can be reflected by the perfusion parameters measured by DSC MRI. A measure of the rate of delivery of arterial blood to a capillary bed in the brain tissue is rCBF. The early recurrence group had significantly decreased rCBF in the territories of AICA, PICA, and VA compared to the no early recurrence group. Local hemodynamic changes of some macro- and micro-vessels caused by increasing tortuosity of vessels were indicated by higher mean blood flow velocities in patients with the HFS, as found by a previous study [33]. The CBF value after the MVD surgery might be increased when the CBF value in the HFS patients before the MVD surgery is low, which causes to increase in the deformative force of the offending arteries and increased tortuosity of vessels. HFS symptoms might be caused to reappear by this [5,6].

In summary, we may explain reasons why HFS patients with low CBF have more recurrent than those with high CBF after MVD surgery : first, low CBF may indicate a chronic and severe compression of the facial nerve by the vessel, which may cause more damage and dysfunction to the nerve and its nucleus. Second, low CBF may also reflect a reduced blood supply to the brain regions involved in facial movement, such as the cerebellum, the temporal lobe, and the cingulate gyrus, which may impair the recovery of the nerve function after surgery. Third, low CBF may increase the deformative force and tortuosity of the offending vessel, which may cause it to recontact facial nerve after surgery and trigger the spasm again. However, more studies are needed to confirm the exact pathophysiology and prognosis of early recurrence. Our hypothesis that rCBF can be used as an indicator of neurovascular compression and recurrence risk in HFS patients was confirmed by this. The failure of MVD can be attributed to the recurrence of symptoms within such a short timeframe. This scenario often occurred when the medial exposure of the armpit’s facial nerve root was insufficient, and the conflict site remained inadequately detached from the offending artery. Due to the reliance of HFS episodes on precise compression with specific amplitude and frequency corresponding to blood pressure and heartbeat, a loss of accurate impact on the facial root may result from any slight alteration in the relative positioning of the artery and nerve during cerebellum elevation.

We used machine learning methods to predict early recurrence patients after MVD surgery based on the DSC indices. We found that the EF parameter, which reflects the vascular permeability and the breakdown of BBB, was a good predictor of early recurrence patients using the NB and the 2nd SVM methods. EF describes the fraction of the contrast agent that is extracted during the first passage through the tissue. Our results showed that the EF value was higher in early recurrence patients than in no early recurrence patients (Fig. 2 and Supplementary Table 3), indicating that there was a certain degree of BBB damage in the early recurrence patients. A previous DSC MRI study has shown that EF has the highest diagnostic performance for differentiating different types of tumors [1]. Although we found that rCBF significantly decreased in the early recurrence patients (Fig. 2 and Table 2), the machine learning analysis showed that the AUC value was lower with rCBF than with EF.

We also identified some brain regions that were important for predicting early recurrence patients based on the EF value, such as the brainstem, MTG, amygdala, and PC. These regions are related to the facial nerve function and emotional processing, which may be affected by neurovascular compression and recurrence of HFS. The facial nerve REZ in the brainstem is the most directly involved in the development of HFS, as it leads to hyperexcitability of facial motor nucleus and functional changes of brainstem [20,21]. The MTG is involved in dynamic facial expression and emotional processing, along with other regions such as superior temporal sulcus and amygdala [17,29,34,35]. The PC is involved in executive control of movement initiation and integration of motor and attention processes, which may be impaired by neurovascular compression and recurrence of HFS [15,24,32,41]. A previous study showed that GMV in these regions was significantly abnormal in HFS patients compared with controls [1]. Moreover, emotional upheaval may trigger spasms by increasing vessel pulsation and nerve abrasion [45]. Amygdala and parahippocampal gyrus are implicated in stress responses and emotional stimuli processing [9]. Therefore, we hypothesize that brain damage in these regions may lead to emotional instability, resulting in early recurrence. The emotional upheaval may not be directly related to blood flow or recurrence of HFS after surgery. However, we considered that other brain areas related to facial movement would also have an effect on facial spasms, so we included brain areas such as the amygdala and parahippocampal gyrus in this study. In general, previous studies showed that age and sex did not associate with predicting early recurrence patients after MVD surgery. However, our machine learning models showed better prediction when age and/or sex were combined with DSC indices. This result may suggest that age is associated with CBF changes.

We found that the machine learning methods of NB model (EF; AUC, 0.845), the TB model (MTT; AUC, 0.830), and the 2nd SVM model (leakage; AUC, 0.801) were good tools for predicting early recurrence patients. First, the NB model is trained that the method estimates the parameters of a probability distribution, assuming predictors are conditionally independent given the class. It computes the posterior probability of that sample belonging to each class. It was classified to estimate the parameters required for accurate classification while using less training data than many other classifiers. A major advantage to NB classifiers is that they are not prone to overfitting, thanks to the fact that they “ignore” irrelevant features. Second, the TB model is an ensemble of bagged decision trees for either classification or regression. Individual decision trees tend to overfit. Bagging, which stands for bootstrap aggregation, is an ensemble method that reduces the effects of overfitting and improves generalization. Finally, the KernelSVM model classifies data by finding the best hyperplane that separates all data points of one class from those of the other class. The Kernel-SVM can effectively handle non-linear decision boundaries, control overfitting by adjusting the regularization parameter and Kernel parameters and performed well even with small datasets.

Predicting early recurrence patients after MVD surgery is important for improving surgical outcomes and reducing complications. MVD is the treatment of choice for HFS, due to its high rate of complete resolution. However, some patients experience recurrent or persistent symptoms after surgery, which may require reoperation. The second surgery is reported to be more difficult and riskier than the first surgery, due to adhesion, narrow FOV, and nerve stretching injury [16,38,46]. Therefore, it is very important to predict recurrence of HFS after the first surgery.

The main causes of HFS persistence or recurrence are missed compression zones or improper use of Teflon [23,31]. Most missed compression zones are located in the ventrolateral pons area or bulbopontine sulcus, besides the REZ region. Improper use of Teflon can lead to insufficient decompression or local inflammation due to direct compression of facial REZ. These factors may result in multiple offending vessels or trunk vessel compression rather than single perforator arteriole compression. DSC MRI can detect these situations by measuring the changes in rCBV, rCBF, and MTT, which may suggest multiple offending vessels or main vessel compression. Thus, neurosurgeons should pay attention to these regions and use proper amount of Teflon to displace the responsible vessels during MVD surgery. Using DSC MRI to predict whether patients have potential multi-vessel or main vessel compression is very potential to predict the possible situation of surgery and prevent recurrence of HFS.

We found that 23 patients of the 32 patients who had symptom recurrence after MVD surgery had their symptoms disappear during the follow-up period, with a mean time of symptom resolution of 2.7 months postoperatively. Various studies have reported delayed relief of HFS after MVD surgery. According to the study of Li [22], 41 (7.4%) out of 545 patients had persistent spasms after surgery, of whom 33 had resolution within 1 year. On the other hand, Ishikawa et al. [10] observed a delayed relief in half of their 175 patients; the delayed relief was within 1 week in 25%, 1 month in 50%, and 8 months in 90%. In the study of Sindou [40] of 147 patients, complete relief of the spasm was delayed in one-third of the patients. The delay was less than 6 months in 50%, between 6 months and 1 year in 38%, and more than 1 year in 12%, of whom two patients took up to 3.5 years. In another study about the failure of MVD, the authors recommended waiting at least 3 months before deciding on reoperation even if the outcomes were unsatisfactory after MVD [45]. Some authors suggested that reoperation should be considered after at least 1 year of follow-up [4,39]. Most of the patients in our study were cured of recurrent symptoms after 2.7 months, but it was difficult to determine MVD failure or delayed relief in patients whose symptoms did not improve within 1 year of follow-up. One patient’s symptoms remained the same as before surgery even after 33 months of follow-up, which could indicate MVD failure.

Although we provided evidence for DSC parameters alteration in early recurrence patients, our study has some limitations that may affect the generalization of the results. First, the recurrence rate after our intraoperation is much higher than expected. Second, we did not perform MVD again on the early recurrence patients, so it is difficult to determine the specific reason for the recurrence. Teflon adhesions and secondary granulomas may distort nerves and cause compression again [38]. This situation is not suitable for predicting the probability of recurrence.