Introduction

With an incidence of 2–3/100,000 people per year in Europe and the USA (www.cbtrus.org), glioblastoma is the most frequent and malignant primary brain tumor in adults [17]. Due to the infiltrative nature of this disease, surgery fails to remove the entire tumor cell population, and despite a postoperative combination of radiotherapy and chemotherapy, overall survival (OS) is poor with a median OS of 14.6 months [30, 31].

Regarding surgery for malignant brain tumors, gross total resection (GTR) is associated with longer survival than subtotal resection or biopsy [1, 13, 16, 18, 23, 29, 35, 36]. One of the main challenges in achieving the largest extent of resection (EOR) without causing damage to the surrounding functional brain parenchyma is the distinction between tumor and normal brain. This distinction is especially difficult in primary brain tumors, which often show a diffuse growth pattern resulting in a transitional zone at the tumor margins where malignant and normal cells are intermingled. Much effort has been invested in the development of technologies that might improve EOR and that might help to distinguish tumor from normal brain. In daily practice, neuronavigation, intraoperative magnetic resonance imaging (ioMRI), intraoperative ultrasound (ioUS), intraoperative computed tomography (ioCT), and 5-aminolevulinic acid (5-ALA) are the most frequently used ones.

This study includes only patients who had microsurgical GTR (GTR; here defined by no residual nodular contrast enhancement in the early postoperative MRI within 72 h). This study does not aim to analyze the impact of intraoperative resection control modalities on EOR. We investigate whether OS and progression-free survival (PFS) in these patients might depend on the intraoperative imaging modality used to achieve GTR.

Patients and methods

Medical charts of patients with primary glioblastoma operated at our institution between the years 2009 and 2012 were reviewed retrospectively. According to federal regulations, no written informed consent was required for this study. A waiver was obtained from the local ethics committee (KEK-StV-Nr. 27/14). We included data of patients with GTR of these lesions as evaluated by an independent neuroradiologist in the early postoperative MRI, defined as cerebral MRI with contrast within 72 h after surgery. We excluded patients with residual tumor (defined by residual nodular contrast enhancement) in the early postoperative magnetic resonance imaging (MRI), as well as patients followed for less than 6 months. The primary endpoint of this study was OS depending on which intraoperative resection device was used, and the secondary endpoint was PFS as defined by radiographic recurrence. Intraoperative technical resection control modalities comprised ioUS and ioMRI and a group of patients without intraoperative resection control (only neuronavigation, which was used in all cases). We used a small-range high-frequency (7–15 MHz) ioUS probe (L15-7 probe iU22 Ultrasound Systems, Philips, Bothell, USA) as shown in our previous publications [7, 26] as well as in an illustrative case (Fig. 1) and a low-field ioMRI, PoleStar N-20 0.15 T (until 09/2010) and N-30 0.15 T (after 09/2010; both Medtronic, Louisville, USA). In addition, we assessed whether 5-ALA (Gliolan®, medac Hamburg, Germany) was used. Tumor volume was calculated using measurements of the contrast-enhancing rim on gadolinium-enhanced T1-weighted MR images and applying the formula: tumor volume = 4/3 ∗ p ∗ 1/2x ∗ 1/2y ∗ 1/2z), where x, y, and z are the maximum diameters within the three axes.

Fig. 1
figure 1

Intraoperative ultrasound imaging in an illustrative case. a, b Preoperative T1-weighted gadolinium-enhanced magnetic resonance (MR) images show a ring-enhancing parietal mass. c Transdural intraoperative ultrasound (ioUS) imaging before the resection in a coronary plane shows the lesion to be homogenously hyperechogenic, even in MR-hypointense, non-enhancing areas. d, e Postoperative T1-weighted gadolinium-enhanced MR images and a corresponding postresection ioUS surface scan (f) in a coronary plane. g, e Intracavital ioUS scanning of the resection cavity shows a complete resection

Full size image

Statistical analysis

Descriptive statistics were used to present continuous variables as mean with standard deviation and categorical variables as counts and percentages of total. OS and PFS are presented as median and 95 % confidence interval (CI). Continuous variables were compared between groups using the Mann-Whitney test, and categorical variables were compared using the Fisher’s exact test.

The primary endpoint OS was first addressed graphically using Kaplan-Meier plots. To quantify the effect of the intraoperative imaging device (ultrasound versus no ultrasound) on survival, we fitted a multiple Cox proportional hazards model. The association between the imaging device and the outcome was adjusted for the effect of the potential confounders age, 5-ALA, tumor volume, and preoperative Karnofsky performance status. After fitting this model to the full dataset, we performed additional subgroup analyses within the patients with glioblastoma.

The level of statistical significance was set at 5 %. Statistical analyses were performed using SPSS 20 (IBM, Chicago, IL, USA) and R (R Core Team (2013). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/).

Results

Patients

We included data of 76 glioblastoma patients (30 females [39.5 %], 46 males [60.5 %], mean age 57.4 ± 11.6 years) with GTR and adequate follow-up. The mean follow-up time was 623.8 ± 346 days. 5-ALA was used in 19 (25 %) glioblastoma cases. Intraoperative technical resection control modalities comprised ioUS (n = 48), ioMRI (n = 22), and a control group without either modality (n = 11). In five cases, both ioUS and ioMRI were used. Basic patient characteristics are shown in Table 1. The overlap of different technical resection devices and 5-ALA is shown in a VENN diagram (supplementary Fig. S1).

Table 1 Basic patient characteristics
Full size table

Primary endpoint—OS

OS is illustrated in a graphical approach using Kaplan-Meier plots for all patients (n = 76) as depicted in Fig. 2a. When looking at the use of intraoperative resection control devices (Fig. 2b), OS was longer in patients with ioUS (either ioUS alone or ioUS in combination with ioMRI). Thus, in further graphical and statistical analyses, we focused on the comparison between patients that had GTR with ioUS compared to those without ioUS. Median OS for all patients (n = 76) was 20.4 months (95 % CI 18.5–29.0) as depicted in Fig. 2a. In the group of glioblastoma patients with ioUS (n = 48), median OS was 21.9 months (95 % CI 18.5–33.1)—30 patients died during follow-up (see Fig. 3a). In contrast to that, median OS in those patients without ioUS (n = 28) was 18.8 months (95 % CI 15.8–NA) and 20 patients died (see Table 2). After fitting the Cox proportional hazards model to all patients, adjusting for age, preoperative KPS, tumor volume, and 5-ALA, we found a lower hazard for “death” in the group of glioblastoma patients with ioUS (hazard ratio = 0.62; 95 % CI 0.32–1.24; p = 0.180), however not statistically significant (see Table 3).

Fig. 2
figure 2

Kaplan-Meier plots showing overall survival (OS) (a) and progression-free survival (PFS) (c) in all glioblastoma patients (n = 76) including the 95 % confidence intervals (dashed lines). OS (b) and PFS (d) of all patients depending on the intraoperative resection control device used

Full size image
Fig. 3
figure 3

Kaplan-Meier plots showing overall survival (OS) (a) and progression-free survival (PFS) (b) comparing the intraoperative ultrasound (ioUS) group and the group operated without ioUS

Full size image
Table 2 Primary and secondary end points
Full size table
Table 3 Results of the Cox proportional hazards model for OS and PFS
Full size table

Secondary endpoint—PFS

Prolonged PFS was seen for the ioUS cohort as depicted in a graphical approach (Fig. 3b). Median PFS in the ioUS group was 7.1 months (95 % CI 5.6–11.7 months) compared to patients operated without the use of ioUS (median PFS 3.4 months; 95 % CI 3.4–11.2 months; see Table 2). The estimated hazards ratio for ioUS versus no ioUS was 0.59 (95 % CI 0.33–1.05; p = 0.072) when adjusting for the set of confounders (see Table 3).

Additional analyses

We performed additional analyses in order to assess whether the more favorable results regarding OS and PFS in glioblastoma patients operated with ioUS compared to glioblastoma patients operated without ioUS are explained by different preoperative tumor volumes or due to different postoperative treatment (chemotherapy and/or radiation). Mean preoperative tumor volume within the ioUS group (n = 48) was 50,822 mm3 (interquartile range (IQR), 7789–81,325 mm3), and mean tumor volume in glioblastoma patients operated without ioUS (n = 28) was 42,016 mm3 (IQR, 14,700–56,089 mm3). There was no statistical significant difference between both groups (p = 0.540; Mann-Whitney U test). In addition, we looked at the postoperative treatment modalities following microsurgical complete resection in the glioblastoma group (shown in Table 4). Both, the ioUS group and the group of patients without ioUS had a similar distribution of treatment modalities with a rate of 83 % and 82 % of patients treated with a combined regimen (any chemotherapy at any time point during disease course and radiotherapy). There was no significant statistical difference between the ioUS group and the non-ioUS group regarding this distribution (p = 0.852; two-sided Fisher’s exact test). Also, the groups did not differ regarding the rate of patients that received standard therapy (radiotherapy plus concomitant and adjuvant temozolomide) as first-line treatment (p = 0.306, chi-squared test; see Table 4) nor regarding cumulative radiation doses (p = 0.546, Mann-Whitney U test).

Table 4 Distribution of initial postoperative treatment modalities (chemotherapy, radiotherapy, combined radio-chemotherapy) after microsurgical glioblastoma complete resection
Full size table

We also looked at the distribution of the molecular markers isocitrate dehydrogenase (IDH1) R132H mutation and O(6)-methylguanine-DNA methyltransferase (MGMT) promoter methylation. IDH1 R132H mutation status was assessed in 41 patients, and none of the tested individuals was found to carry the R132H point mutation—as expected in primary glioblastoma. MGMT promotor methylation status was assessed in 52 patients. In 11/52 patients, the MGMT promotor was methylated (nine patients in the ioUS group and two patients in the non-ioUS group). The difference between the groups regarding MGMT methylation status was not statistically significant (p = 0.431; two-sided Fisher’s exact test) (see also Table 4).

Discussion

Primary endpoint

In this study, OS and PFS were longer in glioblastoma patients that had a GTR using ioUS compared to those patients without ioUS. In a Cox model, this difference did not reach statistical significance. However, a trend toward a benefit of ioUS use can be seen, especially concerning PFS. Regarding glioblastoma, it is known that the tumor burden goes well beyond the contrast-enhancing part seen on MRI. Recent positron emission tomography (PET) studies showed that glioblastoma tumor volume is usually greater when assessed with PET compared to gadolinium-enhanced MRI [2, 10, 11]. By convention, the EOR in glioblastoma is calculated by defining all contrast-enhancing tumor volume as 100 %. Of course, even removing all contrast-enhancing tumor parts (100 % by conventional definition) does not mean that all tumor cells have been removed. None-enhancing parts at the border zones frequently remain. These non-enhancing tumor parts are not detected using standard gadolinium-enhanced ioMRI sequences since ioMRI resection control is classically based on contrast enhancement alone. On the contrary, it is well known that low-grade gliomas (usually non-enhancing lesions) are well delineated with ioUS [14]. In fact, we propose that glioblastoma resections with ioUS go beyond the contrast-enhancing parts, enabling a more complete removal of the tumor (supratotal resection), which is one of the possible reasons for the effects of ioUS on OS and PFS that we observed. The same effect, fluorescence beyond MRI contrast enhancement was recently shown for 5-ALA [24].

Intraoperative resection control devices

IoMRI was introduced in the late 1990s starting with a 0.5-T low-field “double-doughnut system” [4]. High-field ioMRI (1.5–3.0 T) was started later—the choice between the two setups is usually a tradeoff between image quality and integration into the clinical workflow as well as costs [12]. In our series, an intraoperative low-field system was used as described in detail above and published previously [3, 5]. In a randomized controlled trial by Senft et al. assessing the role of ioMRI in brain tumor surgery, a total of 58 patients were included and 29 patients (22 glioblastomas) were randomly allocated to the intraoperative ultra-low-field MRI group (same ioMRI system that was used in this study) [25]. Patients operated with ioMRI had longer PFS (226 vs. 154 days), but this difference did not reach statistical significance (p = 0.083). It is difficult to compare studies like this to our own results, since we purposely included only patients with complete resections and since we mainly focused on OS instead of PFS. When looking at EOR, it has to be taken into account that ioMRI has a huge intrinsic advantage over all other guidance modalities: It is the very same modality that is used to evaluate EOR postoperatively (usually high-field early postoperative MRI). However, it is questionable whether this judgment (contrast enhancement on MRI) is a fair estimate of glioblastoma tumor burden, since it is well known that non-enhancing tumor zones are present in addition to the contrast-enhancing core [34].

5-ALA is a precursor in the biosynthesis of heme that accumulates in malignant gliomas. It carries fluorescent properties that make tumor tissue better visible to the surgeon when using a specially modified microscope [27]. In our series, 5-ALA and ioUS had similar effects on OS and PFS. However, our 5-ALA group was very small (n = 19) and we did not look at all glioblastoma cases up-front, but only at those who had GTR. Moreover, the phase III randomized controlled trial by Stummer et al. (2006) found a difference in PFS, but not for OS which is in accordance to our results [28]. Since 5-ALA fluorescence is based on metabolic effects rather than blood-brain barrier breakdown (as in gadolinium-enhanced ioMRI), using 5-ALA certainly adds additional information and, like ioUS, has the advantage of real-time and low-cost imaging. However, ioUS might be especially useful in glioblastoma patients with no active 5-ALA fluorescence.

IoUS is a rather old technology introduced in the 1960s that was not used frequently over a long period of time due to suboptimal image quality. However, in recent years, ioUS had a comeback mainly because of technical improvements (reduced probe size, improved resolution with high-frequency probes, and the option of 3D imaging) [6, 26, 33]. In brain tumor surgery, ioUS has been shown to be especially helpful in surgery for cystic gliomas [9]. In addition, ioUS can be combined with neuronavigation resulting in improved orientation [15, 19] and it can add valuable information on the vasculature during tumor surgery [21, 32]. The data on ioUS in glioblastoma surgery is still very limited. Coburger et al. recently compared navigated high-resolution linear array intraoperative ultrasound (lioUS) to conventional intraoperative ultrasound (cioUS) in a prospective cohort of 15 glioblastoma patients and 44 resection sites. LioUS (L15-7 probe that was also used in this study) detected residual tumor after microsurgical complete resection in all cases and in 33 of 44 sites (75 %). Histopathology workup revealed solid tumor in 66 % and infiltration zone in 34 % of the cases—there were no false-positive or false-negative findings [8]. A group from Trondheim could show that survival after glioblastoma surgery significantly improved within the same period that ultrasound and neuronavigation were introduced in their department [22]. However, this was just a pure observation without any detailed analyses. This is the first analysis that looks at the influence of ioUS on OS and PFS in glioblastoma. Advantages of ioUS over ioMRI are obviously low costs as well as the fact that images can be obtained repeatedly and fast, whereas ioMRI is usually obtained once at the end of the resection because of the time delay.

Neuronavigation was used in all cases of our series. It cannot incorporate intraoperative anatomic changes (e.g., brain shift), for it does not provide real-time information [20]. However, it should be evaluated whether navigated ioUS with preoperative or intraoperative MRI images leads to improvements in both EOR and safety as well as patient outcome, especially when used by surgeons without long-standing experience with ioUS.

Limitations

We are fully aware of some limitations that are associated with the retrospective design of this study and with the limited number of patients. Since in some patients, more than one resection control modality was used and since subgroups are rather small, it is impossible to study the effect of each device. However, the main analysis looking at ioUS usage is based on two clearly separated groups without any overlap. Due to the retrospective design, it is not possible to judge post hoc whether GTR was intended by the surgeon. Therefore, the intraoperative resection control devices cannot be linked to EOR or to survival in general. However, this information is not necessary with regard to our research question (Is the outcome of patients who had GTR dependent on the intraoperative resection device used?). It is also possible that there might be a bias due to the personal preferences of each surgeon. For instance, it might be possible that one surgeon preferred ioMRI, whereas another surgeon chose ioUS more frequently. One could argue that intraoperative high-frequency ultrasound is preferentially used for superficial lesions since higher frequency leads usually to enhanced axial resolution at the expense of tissue penetration. However, the distance between the core of the lesion and the closest cortical surface was comparable between both groups (see Table 1). In addition, in our institution, high-frequency ioUS is routinely used for deep lesions. Due to the reduction of probe sizes in recent years, ioUS probes can be introduced into the resection cavity as described by Serra et al. (2012). Thus, the ioUS cohort is not a selection of superficial supposedly more favorable cases.

Conclusion

Glioblastoma patients who had GTR have prolonged OS and PFS if ioUS was used. Larger studies (ideally multicenter and randomized) should compare ioMRI to ioUS and confirm these findings with greater statistical power. At this point, routine experienced use of ioUS in microsurgical resections of glioblastomas is recommended. Experienced use of modern ioUS remains a competitive intraoperative imaging device in glioblastoma surgery.