Presurgical Localization of the Primary Sensorimotor Cortex in Gliomas

Data from 71 patients (31 women) with an intra-axial primary brain tumor recruited prospectively between January 2014 and June 2018 were retrospectively reviewed. Patients were recruited prior to planned neurosurgery following review by a multidisciplinary neuro-oncology team. Patients came from two sites. Additionally, 14 healthy volunteers (6 women) were recruited at 1 site for normative and test-retest comparisons. Inclusion criteria were: aged 18 years or above, normal or corrected to normal vision. Exclusion criteria were: contraindications to MRI, prior resections and radiotherapy/chemotherapy. All participants gave prospective informed written consent to take part in the study. The study was approved by the South Central-Oxford B Research Ethics Committee in Oxford and the Local Ethics Committee in Heidelberg (based on a previous approval by the Local Ethics Committee of the University of Würzburg).

The surgical strategy was established for all patients according to a structured protocol for clinical decision making, which was tailored on a case by case basis according to tumor location and extent (as well as potential multifocality and the presence of subependymal or leptomeningeal spread, if applicable). In general, patients were primarily evaluated on an outpatient or consultation basis by a consultant neurosurgeon, who reviewed the available clinical as well as imaging information and examined the patient neurologically and with respect to performance/compliance status. At this point, all treatment options including no surgery, biopsy only, debulking/tumor reduction or gross total resection were considered. Predicated on this initial evaluation, each patient was then reviewed in a multidisciplinary neurosurgical/neuro-oncological tumor board. Based on this joint case discussion, informed by any supplementary imaging when meanwhile obtained, the patient’s likely tolerance of intraoperative monitoring (e.g. electrophysiology and/or awake surgery) and if relevant, adjuvant treatment options, the surgical plan was then ratified or amended. The final plan was communicated with and decided on informed consent of the patient. Importantly, fMRI results were used to inform but not to decide the surgical strategy; indications for intraoperative monitoring (motor evoked potentials [MEP], somatosensory evoked potentials [SEP] or spinal somatosensory evoked potentials [SSEP] phase reversal and/or awake surgery with cortical/subcortical stimulation mapping) according to standardized protocols were always based on proximity to cortical/subcortical structures identified anatomically.

Patients were scanned on one of three Siemens 3T MRI (Siemens Healthineers, Erlangen, Germany) systems; a Verio (n = 23), Trio (n = 22) or Prisma (n = 26) scanner. Healthy volunteers (n = 14) were scanned twice on the Prisma, 6 months apart (online resource methods).

Scans consisted of a 1 mm³ T1-weighted MPRAGE (Magnetization Prepared RApid Gradient Echo) scan, followed by blood oxygen-level dependent gradient-echo echo-planar imaging (EPI) rsfMRI covering the entire brain. The rsfMRI data were collected once per patient, either at standard unaccelerated (TR = 3.5 s), intermediate (TR = 1.56 s) or fast (TR = 0.72 s) temporal sampling (Table 1). For the latter data acquired by simultaneous multislice acquisitions (Trio/Prisma scans), a reference volume providing higher tissue contrast was collected to facilitate registration of each individual’s functional to structural images. For rsfMRI, participants were requested to lie still and rest while watching a fixation cross. The tfMRI data were acquired after rsfMRI in 45 patients and all 14 controls. Dominant hand knob activations were evaluated using motor trials implemented in a fMRI adaptation of the Corsi block tapping test (online resource Fig. S1).

The CS was localized in each hemisphere on every individual’s anatomical scan by two experienced (>15 years) board-certified neuroradiologists working independently and blinded to fMRI results. It was identified using four landmarks: (i) inverted omega of the precentral gyrus [13], (ii) the inverted T sign at the termination of the superior frontal sulcus at the precentral gyrus [14], (iii) marginal ramus of the cingular sulcus or „pli de passage fronto-parietal superieur“ [15] and (iv) termination of precentral gyrus behind the pars opercularis ([16]; Fig. 1). The lowest extension of the precentral gyrus posteriorly turns into the subcentral gyrus, or „pli de passage fronto-parietal inferieur“, which can also be helpful for orientation. Rarely, the pre- and postcentral gyrus are connected by a third „pli de passage fronto-parietal moyen“ of Broca at the level of the hand knob.

The fMRI data were analyzed using the FMRIB Software Library (FSL) [17]. Preprocessing consisted of brain extraction, high pass filtering (cut-off at 90 s), spatial smoothing at 5 mm (full width at half maximum), distortion correction using field maps, and affine registration to structurals using boundary-based registration. Spatiotemporal components were automatically estimated from each individual’s data using independent component analysis (ICA) [18].

For tfMRI, post hoc regression was used to identify ICA components where temporal signal fluctuations correlated with task trial timings (modeled using gamma convolved hemodynamic response functions) [18, 19]. Resulting task ICA components were thresholded using default Gaussian gamma mixture modeling (p > 0.5, denoting a higher probability of reflecting true signal than noise).

In rsfMRI, individual (sensori)motor networks were identified in two ways: first, spatial maps from the single subject resting ICA were inspected to identify a characteristic network overlapping the precentral and postcentral gyri. Visual inspection is practical in clinical settings, but subjective. Therefore, a second automated dual regression (DR) approach [20] was also performed, which objectively extracts spatial components in individual datasets that match given template resting state networks (RSNs). Each subject’s rsfMRI data were regressed against 10 extensively validated RSNs [8]. From the resulting spatial maps, the sensorimotor network was selected and Gaussian Gamma mixture model thresholding (p > 0.5) was applied to match the task analysis.

For tfMRI and rsfMRI, the extent of spatial overlap between fMRI maps and the probabilistic location of the primary motor cortex (M1) were determined based on the Jülich histological template atlas [21]. Overlap between fMRI and M1 was measured based on a voxel count and Dice similarity metrics (online resource methods).

The impact of temporal fMRI resolution was evaluated by statistically comparing frequencies of sensorimotor RSN detections according to temporal sampling rate, and the number of functional subdivisions identified from single-subject resting ICAs between scan groups.

Potentially confounding effects of tumor grade and location were evaluated by comparing fMRI network detection rates according to tumor location (distal/affecting the central region) and histopathology (World Health Organization grading).

Muscle strength was evaluated using the Medical Research Council (MRC) scale before surgery, during hospitalization and up to 3 months after surgery (Table 2).

Statistical analyses were performed using SPSS v25 (IBM Statistics SPSS Version 25; IBM Corp., Armonk, NY, USA; [22]). Assumptions around equality of variance were verified using Levene’s test within SPSS. Paired t‑tests were used to assess correspondence between resting and task-derived measures within groups and reproducibility among controls. Independent sample t‑tests were used to compare clinical and imaging metrics between participant subgroups and χ²-tests were used to compare frequencies of successful CS identifications by tfMRI and rsfMRI according to temporal sampling or clinical variables (tumor location, histology). Statistical significance was set at p < 0.05.

Age did not differ between patients (mean 41.9 ± 13.9 years) and controls (37.9 ± 12.7 years, t = −1.05, p = 0.31). Clinical data are reported in Table 3. Glioma diagnosis was histologically confirmed in 70/71 patients (98.6%), 1 patient refused a biopsy. In 34/71 (47.9%) patients, the tumor directly involved the precentral or postcentral gyrus (n = 23) or displaced/distorted the CS (n = 11).

Anatomical CS localization was consistent between raters in 70/71 patients (98.6%). In one patient, severe effacement due to tumor infiltration confounded CS identification (Fig. 1b). In this case, CS was localized by consensus on re-review of landmark criteria.

Task-correlated motor activations were detected in all 14 controls and all 45 (100%) patients undergoing tfMRI. Finger tapping with the dominant hand activated expected regions of the contralateral precentral and postcentral gyri, ipsilateral precentral gyrus and bilateral supplementary motor area (Fig. 2). At the individual level, no evidence for functional reorganization away from the expected sensorimotor cortex was observed (Fig. 3).

For rsfMRI data, DR reconstructed a bilateral network matching the extended sensorimotor network in 14/14 healthy controls (100%) and 64/71 patients (90.1%). In the remaining seven patients, Gaussian gamma mixture modeling and thresholding resulted in few and sparsely distributed voxels (Online Resource Fig. S2). By comparison, visual inspection of individual rsfMRI ICAs identified at least one clear sensorimotor network in all healthy controls and 61/71 patients (85.9%). This difference between DR and visual identification did not reach significance (\(\chi\)(1) = 1.35, p = 0.25). Within healthy controls, both tfMRI and rsfMRI were highly reproducible, but tfMRI was less variable than rsfMRI over time (Online Resource results).

Task-fMRI activated on average 19% of the M1 mask volume (8785 ± 2477 mm³) in controls and 41% (19,108 ± 7390 mm³) in patients (t = −7.9, p < 0.001). Patients with CS involvement/displacement engaged a larger extent of M1 than patients with tumor remote to the CS (t = −2.4, p = 0.020). By comparison, the resting (sensori)motor map included on average 26,932 ± 8003 mm³ of the atlas M1 mask (58%) in controls and 21,428 ± 10,656 mm³ (46%) in patients (t = 1.8, p = 0.07, Online Resource Fig. S3). Patient subgroups (CS affected or not) also did not differ in that respect (t = 1.8, p = 0.077). In patients with both task and resting fMRI available (n = 45), amount of M1 overlap did not differ between tfMRI and rsfMRI (t = −1.33, p = 0.19). Additional Dice similarity metrics are reported in Online Resource results.

Since tfMRI succeeded in all participants, the impact of temporal acceleration was determined for rsfMRI only. Of the seven patients in whom no resting sensorimotor network was found using DR, four (57.1%) were scanned using basic EPI (TR = 3.5 s) and the other three (42.9%) using intermediate acceleration (TR = 1.56 s). The DR succeeded in all patients with fast acquisition rsfMRI (TR = 0.72 s). Due to the high overall success of DR (90.1%), this difference in template-driven resting network detection did not reach significance (\(\chi\)(2) = 4.67, p = 0.09). The slightly fewer successful visual identifications from single-subject resting ICA reflected differences among scan groups (\(\chi\)(2) = 17.84, p < 0.001). Specifically, sensorimotor networks were detected more frequently in accelerated (n = 47/48, 97.9%) than in basic rsfMRI scans (n = 14/23, 60.8%). To further explore this finding, one Trio patient was scanned using both a sequence without and with an intermediate acceleration for an equivalent number of volumes (250). Accelerated rsfMRI offered enhanced sensorimotor network detectability in this individual (Online Resource Fig. S4).

In terms of subdivisions along the CS, a single extended resting (sensori)motor network was found in 15/61 (24.6%) patients with successful rsfMRI. Spatially segregated sensorimotor networks were identified in the remaining patients. Segregated components localized to the expected face (38/61, 62.3%), hand (38/61, 62.3%) or foot (26/61, 42.6%) motor homunculus subregions (Fig. 4). At least 2 sensorimotor networks were seen in 33 patients (all scanned with temporal acceleration, versus 0/23 patients scanned using a standard sampling rate, \(\chi\)(2) = 29.6, p < 0.001). All 3 functional zones were distinguished in 20 patients, all scanned using temporal acceleration (9/22 Trio, 11/26 Prisma, 0/23 Verio, \(\chi\)(2) = 13.4, p = 0.001). Correspondence to tfMRI available in the same patients varied by sequence acquisition. Patients scanned with temporal acceleration at rest (n = 24) showed better correspondence between template-derived sensorimotor and task maps than patients scanned with no acceleration (n = 21; t = −2.73, p = 0.008). Similarly, spatial concordance between tfMRI and rsfMRI single-subject ICA results was better when selecting the resting map localized to the CS omega sign among accelerated rsfMRI scans (n = 17), compared to using the single widespread sensorimotor resting maps identified in basic rsfMRI (n = 11; t = −2.68, p = 0.013).

The tfMRI succeeded in all cases, while visual detections from single-subject resting ICA were as frequent when a tumor involved or displaced the CS (28/71, 39.4%) as when it did not (33/71, 46.5%; \(\chi\)(1) = 0.68, p = 0.41). The rsfMRI network detectability did not differ between WHO grades (\(\chi\)(3) = 2.10, p = 0.55).

Prior to surgery, 5/71 patients (7%) presented with mild or moderate weakness (MRC grading 3–4/5), each with a tumor involving or displacing the peri-Rolandic cortex (5/34, 14.7%; Table 2). Of the patients eight did not undergo resection, four of whom (50%) had a tumor involving the precentral/postcentral region. Reasons for not undergoing resective surgery included refusal of the patient due to associated risks (n = 4), rapid tumor progression into multifocal disease (n = 2), high seizure activity precluding awake intraoperative stimulation (n = 1), and unsuitability for awake surgery (n = 1). Among the 63 patients operated on, motor function declined immediately following surgery in 12 (19.1%), including 4/5 (80%) patients with pretreatment weakness. Therefore, new muscle weakness emerged postsurgery in 8/30 patients (26.7%) with tumors affecting or effacing the CS, compared to 0/33 (0%) patients with tumors not affecting the CS (\(\chi\)(1) = 10.08, p = 0.001): 3 months after surgery, seven patients recovered fully, two partially, one had persisting motor weakness, and one patient declined further due to ischemia affecting the corticospinal tract.

The 12 patients with postoperative motor deterioration (Table 4) each had a tumor anatomically involving (n = 9) or displacing (n = 3) the CS. Among these, rsfMRI failed to detect a sensorimotor network in 4 (33.3%). In 2/4 cases, tfMRI data were available and succeeded when rsfMRI had failed (Online Resource Fig. S5). All 12 patients had undergone surgery with intraoperative mapping/monitoring.