Presurgical functional magnetic resonance imaging in patients with brain tumors

Introduction

Primary benign brain tumors (incidence, 13 in 100,000 per year), including low grade gliomas (34%), meningiomas (21–28%), neurinomas (8–12%), pituitary adenomas (8–12%), and others (14–16%), are associated with a wide range of comorbidities, such as epileptic seizures and neurological deficits (1–3). Each attempt at full or partial tumor removal is associated with a non-negligible risk of additional neurological deficit (4–6). Technical advancements, such as intraoperative neuronavigation systems and functional magnetic resonance imaging (fMRI), have been developed over the past two decades, with the overall objective to remove as much tumor tissue as possible without damaging healthy tissue (4,7). fMRI is a non-invasive MRI technique that, based on the blood oxygenation level-dependent (BOLD) response, is able to connect activated brain regions to various tasks by detecting indirect effects of neuronal activity on local blood volume, flow, and oxygen saturation (8–11). fMRI enables the surgeon to perform a risk assessment based on the relationship between the tumor and eloquent cortex, to make therapeutic decisions, such as selecting patients for invasive intraoperative mapping (IOM), and to carefully advise the patient about the estimated risks and benefits of the procedure (7,12–15). The surgeon is able to develop a resection plan and estimate the position and relationship between the abnormal tissue and the important functional tissue that should be preserved (7,14). Resection borders cannot be reliably determined only based on fMRI data. However, it has been suggested that a tumor can be safely resected at a distance of at least 10 mm from the fMRI activation without causing any decline in neurological function (13,16). The general idea is that the risk assessment of inducing neurological deficits can be achieved by identifying the distance between the tumor margin and the eloquent or essential functional areas. Thus far, different authors have used different methods to determine the threshold of the fMRI activation maps used for the distance measurements (12,13,16). Although the availability of high magnetic field scanners is increasing in the medical environment, clinical fMRI is seldom part of the standard preoperative imaging of brain tumor patients because it is time consuming to perform the post-scan calculations and interpretations. In this respect, clinical fMRI could be categorized as an upcoming diagnostic tool (17). A robust, standardized, and easily used distance measurement method that correlates with the risk of pre- and postoperative neurological deficits could diminish the workload and make fMRI more attractive for clinical use.

The aim of the present study was to evaluate a new approach for distance measurement that was originally introduced in a smaller study of 14 patients (18), to combine both anatomical landmarks and fMRI, and to assess the method’s ability to predict pre- and postoperative neurological deficit outcomes before and after surgery (assisted by intraoperative neuronavigation).

Material and Methods

Postprocessing of the fMRI data

Each fMRI scan was analyzed using dedicated software located on the MRI scanner (BrainwavePA version 1.3.08130, GE Healthcare). The analysis included smoothing with a Gaussian spatial filter at a full width/half maximum (FWHM) of 8 × 8 × 8 mm. The measured BOLD signal was compared to the applied hemodynamic reference function on a voxel-wise basis using multiple regression analysis, which generated a t-test map with a threshold of P = 0.05. The software automatically estimated the effective number of degrees of freedom to account for temporal autocorrelation caused by the smoothness of the hemodynamic response, which was used to convert the t-test map to an “activation” Z map. The activation Z map was then co-registered to the morphological images. For every fMRI task, the neuroradiologist generated functional maps for clinical use by raising the Z score, i.e. increasing the statistical threshold (P < 0.05) to a level that illustrated where each performed task was located in the cerebral cortex (Fig. 1). These maps were made assessable for the neurosurgeon as clinical fMRI maps for presurgical planning and intraoperative neuronavigation purposes. OPEN IN VIEWER

Fig. 1.

Clinical fMRI, with active motor stimulation of the right hand in patient 10. The statistical threshold of the presurgical fMRI was determined by consensus between the neuroradiologist and the neurosurgeon. Activity is observed in the motor cortex, the sensor cortex, and the supplementary motor (SMA) cortex (corresponding to the right hand).

A standardized method for measuring the lesion-to-activation distance (LAD) was retrospectively applied. The method considers the morphology and fMRI activity. Project fMRI maps were made for each task (Fig. 2); the statistical threshold was at a level at which activity was only visible in one axial image plane that corresponded to the relevant morphological area. The relevant morphological area was identified using various anatomical landmarks (12,19). Using a MR workstation (GE-Reformat version 4.2, GE Healthcare), the distance between the high threshold fMRI map and the tumor could then be directly obtained in the oblique three-dimensional (3D) window (Fig. 3). OPEN IN VIEWER

Fig. 2.

Project fMRI, with active motor stimulation of the right hand in patient 10. The statistical threshold was adjusted according to the relevant anatomical landmarks, in this case the “hand knob” (a) in the axial images and “hand hook” (b) in the sagittal images. Therefore, the pixels with activity in the relevant morphological area were only visible in one axial image (a).

OPEN IN VIEWER

Fig. 3.

Project fMRI, with active motor stimulation of the right hand in patient 10. The red dot in the “hand hook” area corresponds to the activity in the motor cortex, according to Fig. 2. The 10.5-mm distance between the tumor and fMRI activity was measured using the oblique 3D window on an MRI workstation.

For each tumor, the volume was roughly estimated from the T1 morphological images using the formula for an ellipsoid, V = 4πabc/6.

After the LAD measurement of each fMRI task, the neurological statuses of the patients were obtained from the hospital records. In all cases, the neurological assessment was performed by a senior neurosurgeon. Based on the findings from the objective neurological examination, each patient was scored pre- and postoperatively for neurological deficits corresponding to the motor and or the language task with shortest LAD. Any relevant motor/language deficits were scored preoperatively as yes/no. Any additional relevant motor/language deficits postoperatively were scored yes/no. No standard protocol was used for neurological testing.

Results

fMRI was successful in all patient sessions; all performed tasks clearly delineated the motor and/or language cortices in the hemisphere with the tumor.

In the preoperative assessment, 27 of the 34 patients met the inclusion criteria. Among the 27 included patients, 26 underwent one or more motor fMRI scans, and one patient underwent only a language fMRI scan (Table 1). There was a significant relationship between the preoperative neurological motor deficits and LAD (Table 2).

OPEN IN VIEWER

Table 1.

Patients included for preoperative assessment.

				LAD (mm)
Patient no.	Age (years)/ sex	Tumor location	Diagnosis	Motor	Language	Language lateralization Dom/Nondom	MDPr	LDPr	Tumor Volume (cm3)
1	37/M	L temporal	Astrocytoma Grade I-II	25.4	0	L	No	Yes	32.3
2	40/M	L frontoparietal	Astrocytoma Grade I-II	15.3	28.3	L	No	No	45.0
3	68/M	L parietal	Metastasis	6.4	*		Yes	No	9.5
4	71/F	R frontal	Metastasis	27.9	*		Yes	Yes	0.6
5	22/F	R frontal	Astrocytoma Grade I-II	18.1	33.7	R/L	No	No	22.2
6	10/F	L frontopariotemporal	Oligodendroglioma	18.1	*		Yes	Yes	*
7	29/F	L frontal	Cavsrnoma	14.3	*		No	No	2.7
8	39/F	L frontal	Oligod endroglioma	12.1	48.4	L/R	Yes	Yes	4.5
9	53/F	R frontal	Astrocytom grade III	12.3	*		Yes	No	2.1
10	39/F	L frontal	Astrocytoma grade I–II	10.5	*		Yes	No	11.2
11	62/F	L frontal	Glioblastoma	27.9	*		No	No	0.9
12	33/M	R temporal	Astrocytoma grade I–II	21.8	0	L/R	No	No	68.1
13	25/M	L temporal	Ganglioglioma	54	21.7	L	No	No	0.8
14	55/M	L frontal	Astrocytoma grade III	0	*		Yes	No	37.6
15	18/F	L frontal	Gliose	29.6	12.3	L/R	No	No	9.8
16	41/M	L frontotemporal	Oligodendroglioma	0	*		Yes	No	102.3
17	35/M	L frontal	Astrocytoma grade I–II	30.4	0	L	No	No	113.7
18	55/F	L temporal	Glioblastoma	24.6	12	L	No	Yes	6.5
19	44/F	R parietal	Meningioma	3.6	*		Yes	No	13.0
20	28/M	R frontal	Astrocytoma grade I–II	6.6	*		No	No	56.7
21	53/F	L frontal	Pleomorft xanthoastrocytoma	25.7	31.2	L	No	No	7.5
22	50/M	R frontal	Oligodendroglioma	35	*		No	No	80.5
23	37/F	R frontal	Oligodendroglioma	21.2	*		Yes	No	25.2
24	34/M	R frontal	Astrocytoma grade I–II	2.6	38.8	R	No	No	18.8
25	44/F	L frontal	Oligodendroglioma	61.4	*		No	No	94.5
26	44/M	L frontal	Astrocytoma grade III	*	9.8	L	No	Yes	198.7
27	39/M	L parietal	Astrocytoma grade I–II	13.5	*		No	No	40.2

OPEN IN VIEWER

Table 2.

Neurological deficits versus fMRI distance and tumor volume.

	Neurological deficits	Neurological deficits/ fMRI distance			Neurological deficits/ Tumor volume
	(n/ntotal)	Odds ratio (1/mm)	CI (1/mm)	P	Odds ratio (1/cm3)	CI (1/cm3)	P
Preoperative assessment
Motor fMRI	10/26	0.89	0.81;0.99	0.03	0.99	0.96;1.01	0.32
Unilateral language fMRI	4/8	0.88	0.75;1.05	0.15	1.01	0.99;1.03	0.40
Postoperative assessment
Motor fMRI	7/16	0.82	0.67;0.99	0.04	0.99	0.96;1.02	0.56
Unilateral language fMRI	3/6	1.06	0.89;1.26	0.50	0.93	0.80;1.09	0.41

CI, 95% confidence interval; n, number of patients with neurological deficits in the group; n_total, total number of patients in the group.

Twelve patients underwent a preoperative language fMRI. Four of these patients had bilateral fMRI activation, and eight had unilateral activation (Table 1). In the unilateral group of language fMRI activation, the tumor was placed in the same hemisphere as the activation in all eight cases. Preoperative neurological language deficits tended to be associated with LAD, P = 0.15 (Table 2).

No significant relationship was found between tumor volume and preoperative neurological motor or language deficits (Table 2).

Seventeen of the 27 patients from the preoperative assessment also met the inclusion criteria for the postoperative assessment. Among the 17 included patients, 16 underwent one or more motor fMRI scans (Table 3). There was a significant relationship between additional postoperative neurological motor deficits and the LAD (Table 2). Among the 16 patients with motor fMRIs, the tumor was radically removed in 10 patients, based on a 48-h postoperative MRI. Logistic regression analysis in this group of 10 patients suggested a perfect fit, with a cutoff LAD of 18 mm. Five patients with a LAD of less than 18 mm experienced additional postoperative neurological motor deficits, whereas five patients with a tumor-fMRI distance greater than 18 mm did not experience additional postoperative neurological motor deficits (P = 0.008, Fisher’s exact test).

OPEN IN VIEWER

Table 3.

Patients included for postoperative motor fMRI assessment.

Patient no.	LAD (mm) motor	MDPr	nMDPo	Tumor volume (cm3)	Resection*
2	15.3	No	Yes	45.0	C
3	6.4	Yes	No	9.5	–
4	27.9	Yes	No	0.6	–
5	18.1	No	Yes	22.2	C
7	14.3	No	Yes	2.7	C
8	12.1	Yes	Yes	4.5	C
11	27.9	No	No	0.9	C
13	54	No	No	0.8	C
17	30.4	No	No	113.7	C
18	24.6	No	No	6.5	P
19	3.6	Yes	Yes	13.0	C
20	6.6	No	Yes	56.7	P
21	25.7	No	No	7.5	C
22	35	No	No	80.5	C
23	21.2	Yes	Yes	25.2	P
25	61.4	No	No	94.5	P

Among the 17 included patients, eight underwent a language fMRI (Table 4). Six of these patients had unilateral fMRI activation in the same hemisphere as the tumor. No significant relationship was found between LAD and postoperative neurological language deficits, P = 0.50 (Table 2).

OPEN IN VIEWER

Table 4.

Patients included for postoperative language fMRI assessment.

Patient no.	LAD (mm) Language	Language Lateralization Dom/ Nondom	LDPr	nLDPo	Tumor volume (cm3)	Resection*
2	28.3	L	No	No	45.0	C
5	33.7	R/L	No	No	22.2	C
8	48.4	L/R	Yes	No	4.5	C
13	21.7	L	No	Yes	0.8	C
17	0	L	No	No	113.7	C
18	12	L	Yes	Yes	6.5	P
21	31.2	L	No	Yes	7.5	C
26	9.8	L	Yes	No	198.7	P

No significant relationship was found between tumor volume and additional postoperative neurological motor or language deficits (Table 2).

The patients (Tables 3 and 4) selected for surgery-assisted (solely by intraoperative neuronavigation) radical tumor removal after the preoperative fMRI had an overall mean LAD of 18.8 mm in contrast to the non-operated patients, who had a mean LAD of 6.9 mm (P = 0.02, non-parametric Wilcoxon-Mann-Whitney test).

Discussion

The present study indicates that the applied standardized method for distance measurement between tumor and preoperative fMRI was able to predict postoperative outcomes in terms of neurological motor deficits following intraoperative neuronavigation surgery. Although the method differs from those of other investigators, the LAD cutoff of 18 mm, beyond which no neurological motor deficits were observed in radically operated patients, aligns with the methods used in similar studies (13,16). Considering the preoperative assessments, a statistically significant relationship (P = 0.03) was found between preoperative neurological motor deficits and LAD, indicating involvement of the motor cortex.

These findings indicate the usefulness of the applied LAD measurement in terms of risk assessment prior to surgery close to the motor cortex.

The applied method was not able to predict postoperative outcomes in terms of neurological language deficits, although preoperative neurological deficits tended to be associated with LAD (P = 0.15). This finding could be attributed to the limited number of patients who underwent preoperative language fMRI and the fact that only six patients had unilateral language activation. Another reason could be the involvement of the white matter tracts between the Broca’s and Wernicke’s areas, which could also explain why both patient 13 (LAD for language =21.7 mm) and patient 21 (LAD for language =31.2 mm) experienced additional postoperative language deficits.

The use of a LAD measurement based only on the fMRI activation has met with some criticism because the LAD is highly determined by the accuracy of fMRI localization and technical factors, such as the statistical threshold (12,20). It is true that the accuracy of fMRI localization is low, but note that the fMRI BOLD signal only serves as a pseudo measure for neural activity. The BOLD contrast originates from hemodynamic changes in the vasculature and is slightly offset toward the venous compartment and, thus, does not precisely localize the electrically active neurons (12). However, it has been shown that there is a high reproducibility of the fMRI localization within the human motor and sensory cortex (21,22) when using methods without threshold adjustment (22) and when different methods of threshold adjustments of the fMRI map were used (21). Therefore, using fMRI for preoperative risk assessment seems feasible, as long as there is a good correlation between the preoperative fMRI LAD measurements and postoperative outcomes.

To address criticism regarding the determination of the statistical threshold, the applied standardized method combines both the data from the fMRI and the morphological images of the T1 3D FSPGR sequence to set the statistical threshold used for the LAD measurement. The inter-observer coefficient of reproducibility was reported in a previous study to be 2.4 mm for the LAD (18). The inter-observer difference included both the differences in outlining of the high threshold fMRI activation and in outlining the tumor.

The presented method for LAD measurement should only be applied in the preoperative risk assessment to determine whether to operate solely supported by a neuronavigation system. In this study, the cutoff value was 18 mm for the distance between the tumor and eloquent motor cortex (i.e. the distance that did not result in any postoperative transient or permanent neurological deficits in patients for whom the tumor was concluded to have been radically excised). Other distance measurement studies have differentiated between transient and permanent neurological deficits (13,16). These studies show that a distance between 0 and 10 mm (between the fMRI-activation and tumor) results in a greater risk of permanent neurological deficits. In the present study, for 10 of the 27 patients, the neurosurgeons choose not to operate solely by neuronavigation. These 10 patients had a mean LAD of 6.9 mm. This group, in which one would expect a high risk of permanent neurological deficits after neuronavigation assisted surgery, was actually recommended for another surgical approach. The assignment was based on the morphological MRIs, the clinical fMRIs (Fig. 1) and the clinical experiences of the neurosurgeons. This result indicates that the presented standardized and easy-to-use method for LAD measurement could serve as a more objective clinical management method. Together with the distance measurement, the clinical fMRI could assist neurosurgeons in assessing the feasibility of radical surgical resection in low-grade tumors. If radical surgical resection is believed to be feasible, three key stages could be proposed for surgical planning: (i) motor LAD ≥ 18 mm, an operation solely assisted by intraoperative neuronavigation, with minimal risk of postoperative neurological deficits; (ii) motor LAD < 18 mm, an fMRI-guided operation assisted by IOM; and (iii) inconclusive motor fMRI, with an operation assisted solely by IOM. It is important to consider how one should present the distance measurement results. If the results are presented as a combination of Figs. 2 and 3, then there is a considerable risk of misinterpretation. As discussed earlier, the fMRI activity is only a pseudo measurement of neuronal activity. A combination of Figs. 2 and 3 could lead the surgeon to the conclusion that the high threshold fMRI activity in motor cortex would correspond to the exact same point if he/she had used IOM. However, the relationship between the presented method for distance measurement and IOM needs further investigation. Meanwhile, the distance measurement could be presented with the clinical fMRI images (Fig. 1).

One limitation of this study is the small sample size, which can limit the generalization of the results. A larger patient cohort is needed in future research. Another limitation is the retrospective study design. Further prospective studies are needed to evaluate the value of this method for motor, language, and sensor areas.

In conclusion, the presented easy-to-use approach for lesion-to-activation distance (LAD) measurement combines both anatomical landmarks and fMRI using clinically approved software, and it was able to predict the presence of preoperative neurological motor deficits, as well as the occurrence of additional postoperative neurological motor deficits, assisted by intraoperative neuronavigation.