|Home | About | Journals | Submit | Contact Us | Français|
Functional magnetic resonance imaging (fMRI) is commonly utilized by neurosurgeons to pre-operatively identify brain regions associated with essential behaviors, such as language and motor abilities. This study investigated the relationship between the distance from tumor border area to functional activations in secondary motor and language cortices to patient morbidity and mortality.
Patients with primary or metastatic brain tumors who underwent pre-operative fMRI motor and language mapping were selected from a large database of tumor patients. The lesion-to-activation distance (LAD) was measured in each subject relative to the supplementary motor area for motor tasks and pre-supplementary motor area for language tasks. The association between LAD and the incidence of deficits was investigated using Fisher’s exact tests of significance. The impact of other variables, including age, handedness, gender, and tumor grade were also investigated. In a subset of subjects, logistic regression was performed to identify the likelihood of deficits based on LAD to primary and secondary regions. Finally, Mantel-Cox log-rank tests were performed to determine whether survival time significantly related to LAD to secondary motor and language areas.
A significant association was observed between LAD to the SMA and the incidence of motor deficits, with the percentage of patients with deficits dropping for those in the LAD > 2 cm group. The relationship between LAD to the pSMA and the incidence of language deficits was not significant. Logistic regression demonstrated that the LAD to primary sensorimotor cortex does affect the incidence of motor deficits, but LAD to SMA does not. Finally, we observed no relationship between LAD to secondary regions and patient mortality.
These results demonstrate that LAD to SMA structures does affect morbidity, although not to the extent of LAD to primary structures. In addition, motor deficits are significantly associated with LAD to secondary structures, but language deficits are not. This should be considered by neurosurgeons for patient consultation and pre-operative planning.
FMRI is emerging as an increasingly reliable technique for the noninvasive preoperative localization of eloquent cortex in relation to pathology. FMRI offers the advantage of spatial localization of language and sensorimotor areas, which allows for non-invasive localization of neural activity as well as the prediction of postoperative patient deficits based on the proximity of fMRI activation to the margin of intracranial tumors. Previous studies have demonstrated consistent agreement between fMRI and the much more invasive WADA testing for language lateralization, as well as between fMRI and electrocortical stimulation, which has led to an increase in the utilization of fMRI for routine preoperative assessment of brain lesion patients2,17. In addition to providing information that can guide tumor resection, the use of preoperative fMRI allows for the investigation of the relationship between the mapping of functional areas and tumor location. Several previous studies, including work in our lab, have demonstrated that incidence of language and motor deficits increases as the distance between tumor border and functional activation in primary language and motor areas, respectively, decreases4,8,16–17.
FMRI has also proven useful in mapping secondary motor (supplementary motor area; SMA) and secondary language (pre-supplementary motor area; pSMA) areas. Generally speaking, surgical resection of the SMA is thought to result in transient motor and language deficits3,11,18. Previous studies have demonstrated the relationship between the overlap of tumor resection and SMA fMRI activity and transient motor and language deficits5–6. However, to our knowledge, only one study has investigated the relationship between SMA (or pSMA) fMRI activity to tumor distance and motor and language deficits. In a relatively small sample of patients (n = 12), Nelson et al demonstrated that the risk of postoperative motor and language deficits patients was 100% when the distance between the SMA and the tumor was less than 5mm, but 0% when this distance was greater than 5 mm9. In this study, we explore the role of fMRI in predicting morbidity as well as mortality related to tumor lesions encroaching upon secondary motor (SMA) and secondary language areas (pSMA) in a large sample of tumor patients.
Subjects were selected from a database of 423 patients who received fMRI as part of pre-surgical planning at the University of Wisconsin Hospital and Clinics between June 1999 and July 2011. Inclusion criteria for this study selected all patients with diagnosis of primary or metastatic tumors in any lobe of the brain who underwent motor and/or language mapping using fMRI. Demographic information can be found in Tables 1 and and2.2. Patients gave informed consent according to study protocol approved by the institutional review board. Patient information was collected from medical records. Any record of preoperative or postoperative weakness (lower extremity, upper extremity, and/or facial) and/or aphasia (Broca’s, Wernicke’s, conduction, global, etc.) was included in the analysis as deficits. Mortality information was collected for all patients using medical records that were cross-referenced with the Social Security Death Index (SSDI). Original data from Social Security Administration, accessed via the http://stevemorse.org/ssdi/ssdi.html, which aggregates information from the following websites: ancestry.com, familysearch.com, familytreelegends.com, geneaology.com, genealogybank.com, newenglandancestors.org, rootsweb.com, and worldvitalrecords.com was utilized for query10. Patients with no listing in SSDI were considered still alive.
The language and motor paradigms used to assess patients are described in more detail elsewhere7,16. In brief, brain activation associated with expressive language was measured with two types of word generation tasks: 1) alternating 20-second blocks of antonym word generation and rest and 2) alternating 20-second blocks of letter word generation task and rest. Similarly, brain activation associated with receptive language was identified with alternating 20-second blocks of text reading and symbol reading. In this task, the patient silently read a short paragraph in the text reading block. During the control or symbols block, the patient was shown a paragraph of symbols and asked to scan and find specific symbols. The symbols block controlled for eye movements during reading, which presumably helped discriminate visual and eye movement-related activity from the true language areas.
Motor activation was determined using a variety of motor tasks, including blocks of unilateral finger tapping, alternative hand finger tapping, unilateral foot or ankle movement, tongue movement, and lip pursing, all contrasted with blocks of rest. All tasks were not performed by every subject. The task that resulted in the most robust language or motor related activation for each subject was selected for analysis.
Imaging data was collected using either a 1.5T or 3T commercial MR scanner (GE Medical Systems, Milwaukee WI) equipped with high-speed gradients. BOLD-weighted single-shot echo-planar images (EPIs) were obtained repeatedly at intervals defined by the repetition time (TR) for each patient during task performance. Technical parameters were: field of view (FOV): 24cm; matrix: 64×64; TR=2000ms; echo time (TE): 40ms (for 1.5T) or 27ms (for 3T); FA: 85 deg (for 1.5T) or 75 deg (for 3T); 6mm coronal plane sections (for 1.5T) or 5mm axial plane sections (for 3T); spatial coverage was sufficient to provide mapping of the entire cortex. The number of images and length of imaging varied with paradigm. Imaging duration ranged from 3 to 5 minutes. Additional high-resolution anatomic scans, including 3D volumetric T1- and T2-weighted sequences were acquired as part of the preoperative assessment. Spatial smoothing was applied to the EPI datasets using a FWHM Gaussian kernel of 8 mm for 1.5T datasets and 6 mm for 3T datasets. Slice time correction, volume registration, and co-registration with subject-specific anatomical volumes were performed in AFNI1. Activation was determined by cross-correlation of the time-course of the EPI signal at each voxel with a generalized least squares fitting algorithm to a smoothed and temporally delayed boxcar reference function modeling the presumed hemodynamic response. This comparison provided a voxel-wise t statistic with which images were thresholded individually to optimize visualization of language or motor areas and overlaid on the co-registered anatomic brain volume maps for each individual subject.
Thresholding was subjectively applied by a trained MR imaging technologist for each patient, with the intent of optimizing specificity and sensitivity to expected regions of task response. "Specificity" means adjusting threshold to minimize spurious voxels that were considered artifact, for example due to head motion. "Sensitivity" means adjusting threshold to highlight the expected responses at a level that displayed a typical supra-threshold extent. This was subjectively adjusted to localize a particular gyrus or region, which represented a statistical probability of greatest confidence as indicated by the t-statistical overlay. Often, a compromise threshold was applied to balance the need to highlight an expected response with the concern toward minimizing artifact. For example, in a dataset that exhibited significant task-correlated head motion, it may not have been possible to minimize the artifacts while still retaining a sufficient sensitivity to the presumed task-related response. The task-related response magnitudes were also dependent on factors such as the patient's ability to perform a particular fMRI task, or whether the BOLD response was compromised by the presence of the tumor. Thus, the threshold was subjectively varied for each individual fMRI scan, based on the quality of the data and medical mapping concerns.
Images used in the analysis were compiled at the time of surgery by a trained MRI technologist and used by the neurosurgical team for pre-surgical planning. T1-weighted, T2-weighted, and contrast-enhanced T1-weighted structural images were analyzed in Picture Archiving and Communication System (PACS). The shortest distance in any plane (coronal, sagittal, or transverse) from the periphery of the tumor to the border of the area of functional activation corresponding to the supplementary motor area (SMA) and pre-supplementary motor area (pSMA), as well as to primary sensorimotor cortex and primary language cortices (Wernicke’s and Broca’s areas) was measured using PACS. Tumor edge was defined as the enhancing margin for tumors that enhance with contrast or the peripheral margin of the solid portion of the tumor as noted in T2-weighted images. Figure 1 depicts an example measurement. When tumor margin was heterogeneous, such for low grade tumors, T1-weighted images were used. Lesion-to-activation distances (LAD) were categorized based on classifications from previous studies12–13,16. These included 1) less than 1 cm, 2) between 1 and 2 cm, and 3) greater than 2 cm.
Fisher’s exact tests of independence were used to compare categorical variables of interest. The primary question of interest was whether there was an association between LAD to SMA or preSMA, and the incidence of motor or language deficits, respectively. As a secondary analysis, the effect of LAD on the existence of postoperative deficits was investigated for patients that had no preoperative deficits. For patients with the relevant information, post-operative deficits were classified as being transient or persistent in nature and the association between the persistence of post-operative deficits and LAD to secondary motor and language areas was assessed using Fisher’s exact tests. Similar Fisher’s exact tests were performed to compare categorical variables that may confound the relationship between LAD and deficits, such as the association between LAD and gender, handedness, or tumor grade, as well as the total number of patients in each LAD group. The association between LAD to primary and secondary motor and language regions and the extent of resection, based on post-surgical radiological notes as being total resection, subtotal resection (<10% residual tumor), partial resection (>10% residual tumor), or no surgery/biopsy only, was also investigated using Fisher’s exact tests. The association between the presence of high grade tumors versus low grade tumors and the extent of resection were also investigated with Fisher’s exact tests. Additionally, Fisher’s exact tests were performed to determine whether the incidence of deficits was associated with gender, handedness, and tumor grade. Chi-square tests were performed to determine whether the number of patients in LAD group for the motor and language subsets were significantly different. Fisher’s exact tests were also performed to compare the distribution of patients with preoperative deficits, postoperative deficits, or both preoperative and postoperative deficits in each LAD group. Finally, independent samples Kruskal-Wallis tests were performed to determine if age significantly differed between LAD groups for the language and motor subsets of subjects.
Binary logistic regression analyses were performed to determine if LAD to primary and secondary motor and language areas affected the likelihood of motor and language deficits, respectively. Two models were run. For the subset of patients in the language-deficit analysis, LAD to primary language regions (Wernicke’s and Broca’s) and LAD to pSMA were included as predictors. For the subset of patients in the motor-deficit analysis, LAD to sensorimotor cortex and LAD to SMA were included as predictors. Only subjects with LAD measures for both primary and secondary regions were included in these analyses.
Mantel-Cox log-rank tests were used to explore relationships between LAD and survival for both the motor (SMA) and language (pSMA) groups. Survival statistics were calculated based on number of months survived from date of diagnosis to date of mortality. All analyses were done using R statistical software15 or Statistical Package for the Social Sciences14. All tests were considered significant at an alpha of 0.05.
Of the 423 patients who underwent fMRI for pre-surgical planning, 52 patients had tumors encroaching on the secondary motor area and had fMRI for motor tasks, and 72 patients had tumors encroaching on the secondary language area and had fMRI for language tasks. Age, handedness, gender, and tumor grade information for each subset of patients are listed in Table 1.
For the subset of subjects included in the motor analyses, a one-way independent samples Kruskal-Wallis test found a significant difference of age between the LAD to SMA groups χ2 (2, N = 52) = 8.2, p = 0.017. There were no significant differences in handedness, gender, or the percentage grade III or IV tumors in LAD to SMA categories (p > 0.1). The extent of tumor resection did not vary due to LAD to secondary or primary motor cortices (p = 0.61 and p = 0.67; Table 3). Similarly, there was no relationship between the frequency of high grade tumors (grade three or four) and the extent of tumor resection for subjects in the language subset (p = 0.33). The number of patients in each LAD to SMA group was significantly different, χ2 (2, N = 52) = 73.0, p < 0.001.
For the subset of subjects included in the language analyses, there was no difference in age between LAD to pSMA groups, χ2 (2, N = 72) = 4.1, p = 0.13. Similarly, there no significant differences in handedness, gender, or the percentage of grade III or IV tumors in LAD to pSMA categories (p > 0.1). The extent of tumor resection did not vary due to LAD to pSMA (p = 0.82), however, the extent of tumor resection did vary due to LAD to primary language cortices (either Broca’s area or Wernicke’s area; p = 0.037; Table 3). This difference was driven by the relationship between LAD to Broca’s area and the extent of resection (p = 0.005), while the relationship between LAD to Wernicke’s area and the extent of resection was not significant (p = 0.77). The frequency of total resection was much higher for patients with LAD to Broca’s area > 2 cm compared to the other LAD groups, while the groups < 2 cm included mostly subtotal or partial resections (Table 3). In addition, there was no relationship between tumor resection and the frequency of high grade tumors (grade three or four) for subjects in the language subset (p = 0.5). The number of patients in each LAD to pSMA group was significantly different, χ2 (2, N = 72) = 21.5, p < 0.001.
Additional Fisher exact tests indicated that subjects with and without deficits in both the motor and language analyses did not differ in gender, handedness, or the percentage of grade III or IV tumors (p < 0.1; see Table 2). Likewise, Mann-Whitney tests demonstrated that subjects with and without deficits did not significantly differ by age for the motor (Z = 0.39, p = 0.69) and language subsets (Z = −0.34, p = 0.73).
Table 4 shows the percentage of patients with preoperative, postoperative, or preoperative and postoperative deficits for the language and motor subsets. Fisher exact tests found no significant effect of LAD group on type of deficit for the motor (p = 0.56) and language subsets (p = 0.89).
For patients with information regarding the persistence of deficits, patients with postoperative deficits were characterized as being either transient or persistent in nature. There was no significant association between LAD and the existence of transient versus persistent postoperative deficits for either the motor (p = 1.0) or the language subsets (p = 1.0). As seen in table 4, the majority of postoperative deficits were transient in nature.
A significant association was found between SMA LAD and the existence of motor deficits (p = 0.04). 62.5% of the subjects in the LAD to SMA < 1 cm group had motor deficits. The percentage of patients with motor deficits was 81.8% in the LAD 1–2 cm group and dropped to 43.4% in the LAD > 2 cm group (Figure 2). However, there was no significant association between SMA LAD and the existence of postoperative motor deficits in patients that were deficit free prior to surgery (p = 0.19). For these patients, 40% in the LAD to SMA < 1 cm group, 33.3% in the LAD 1–2 cm group, and 11.1% of the LAD > 2 cm group had postoperative deficits.
There was no significant association between pSMA LAD and the existence of language deficits (p = 0.81). The percentage of patients with language deficits was 50% in the LAD to pSMA < 1 cm group, 40% in the LAD 1–2 cm group, and 36% in the LAD > 2 cm group (Figure 2). Likewise, there was no significant association between pSMA LAD and the existence of postoperative language deficits in patients that were deficit free prior to surgery (p = 0.22). For these patients, 50% in the LAD to pSMA < 1 cm group, 25% in the LAD 1–2 cm group, and 17.8% of the LAD > 2 cm group had postoperative deficits.
Binary logistic regression was performed in a subset of 56 subjects in the language analyses that had LAD measurements to both pSMA and primary language cortices (Wernicke’s and Broca’s). Neither the LAD to primary language areas (β = −0.469, SE = 0.395, OR = 0.626, p = 0.235) nor the LAD to secondary language areas (β = −0.435, SE = 0.535, OR = 0.647, p = 0.415) significantly affected the likelihood of language deficits.
Binary logistic regression was also performed in a subset of 51 subjects in the motor analyses that had LAD measurements to both SMA and primary sensorimotor cortex. This analysis demonstrated that an increase LAD to primary sensorimotor cortex significantly decreased the likelihood of the existence of motor deficits (β = −1.885, SE = 0.609, OR = 0.152, p = 0.002). In contrast, LAD to SMA did not significantly affect the likelihood of motor deficits (β = −0.317, SE = 0.486, OR = 0.728, p = 0.514).
Mantel-Cox log-rank tests were performed to determine whether survival time significantly related to LAD to secondary motor and language areas. There were no differences in survival rate due to LAD to SMA (χ2 (2, N = 52) = 3.823, p = 0.148), nor were there differences in survival rate due to LAD to pSMA (χ2 (2, N = 72) = 1.281, p = 0.527).
Preoperative functional imaging has become a valuable tool for neurosurgeons, guiding patient consultation and surgical resection of intracranial tumors. Although the use of preoperative fMRI has become more prevalent, the relationship between the distance of tumor from fMRI response and the extent of tumor-induced neurological deficits needs to be resolved. Several studies have demonstrated that the proximity of tumors to functional cortical areas as delineated by fMRI is related to postoperative outcome, with the majority of studies focused on the primary motor and language areas4,8,16–17. However, this relationship is likely dependent on the functional brain region being investigated. This is the first study, to our knowledge, that investigates the relationship between patient morbidity and mortality to tumor distance to secondary motor (SMA) and language (pSMA) areas using a large database of subjects9.
As seen in figure 2, we found a significant relationship between lesion activation distance (LAD) in the SMA and the incidence of motor deficits. The percentage of patients with motor deficits dropped considerably in patients with LAD greater than 2 cm, while the percentage of patients with deficits in the 1–2 cm and < 1 cm LAD groups was relatively high. Furthermore, this observed difference is unlikely to be a result of demographic difference between groups, as the age, handedness, tumor grade, and gender did not differ between the LAD 1–2 cm and LAD > 2 cm groups. In contrast, there was no relationship between LAD and language deficits for the secondary language area (pSMA), although the percentage of patients with deficits did decrease with increasing LAD.
Several previous studies have demonstrated that lesions to secondary motor and language areas are less severe than lesions to primary regions3,5–6,9,11,18. Our results expand on these studies, and demonstrate that LAD to secondary motor areas (SMA) is associated with motor deficits, with relatively high incidence of morbidity in patients with LAD < 2 cm. This information should be utilized by surgeon to inform patient consultation and resection of tumors near the SMA.
We also investigate the association between LAD and postoperative deficits for patients with no preoperative deficits. Although we found no significant association between these factors, the percentage of these patients with postoperative deficits did decrease as the LAD to SMA or pSMA increases. Unfortunately, the small number of patients that did not have preoperative deficits limits the power of this particular analysis.
Information regarding persistent or transient deficits was limited in this dataset. However, for those patients in whom this information was available, we find that the majority of deficits for every LAD group were transient in nature. This is consistent with previous studies which suggest that deficits resulting from tumors or the resection of tumors in secondary motor and language areas are primarily transient in nature, in contrast to those in primary regions3,5–6,11,18.
In this study, we also observed a distinction between the effects of primary and secondary regions. We found that decreasing distance between tumors and fMRI activation in primary sensorimotor cortex increased the likelihood of motor deficits, more so than the effect of distance between tumors and fMRI activation in the SMA. In our previous study drawing from the same database of patients, we found that there was a linear decrease in motor deficits with increase in LAD to primary motor cortex 16. Similarly, for language deficits, the highest incidence of deficits was observed in the LAD (to primary language areas) < 1 cm group, with lower incidence of deficits observed in the 1–2cm and > 2 cm LAD groups. The relationship between LAD to the pSMA and the incidence of language deficits was not significant. We also observed no association between the extent of tumor resection and LAD to secondary motor and language areas. However, we did observe an association between LAD to primary language cortices, driven by LAD to Broca’s area, and the extent of tumor resection for patients used in this study. Incidentally we found no significant association between the extent of tumor resection and tumor grade, although it is likely that tumor grade, the use of intra-operative cortical mapping, and other variables are used by neurosurgeons in combination with LAD measurement. The existence of a relationship between the extent of tumor resection and LAD to primary (Broca’s area) but not secondary language areas could reflect more aggressive surgical approaches for tumors near secondary motor and language areas due to the transient nature of post-operative deficits associated with these areas. The data presented here, in conjunction with our previous study, illustrates the relatively greater plasticity of the brain in response to perturbations to secondary motor and language areas in comparison to primary motor and language areas.
Finally, we investigated the relationship between LAD to SMA and preSMA and the survival rate of patients. No significant relationship existed between survival rate and LAD to secondary motor and language areas. It is possible that survival rate is primarily driven by other factors, such as tumor grade and age.
One methodological issue that should be considered in the current study is the use of subjective thresholding for patient fMRI data. Varying threshold levels does affect the size of the fMRI activations and thus LAD measures. From a basic science standpoint, it might seem desirable to use a set threshold for every patient’s fMRI mapping. However, information regarding the mapping of SMA activity to a hand movement task that does not reach significance at a strict cut-off of 0.05 may still be useful for preoperative planning. The fMRI data used in this study was individually tailored to patients in order to maximize the information afforded by the fMRI for patient care purposes. In this study we calculated LAD from the same subjectively thresholded fMRI activations that the surgeons used for preoperative planning.
This was the first study, to our knowledge, to investigate the relationship between tumor distance to fMRI responses in secondary language and motor areas and the incidence of language and motor deficits in a large sample. In summary, we found a significant association between the distances of lesions to the fMRI responses in the SMA and the incidence of motor deficits. While the incidence of motor deficits was high in patients with LAD less than 2 cm, the incidence dropped considerable for patients with LAD > 2 cm. The relationship between LAD to the pSMA and the incidence of language deficits was not significant. Logistic regression demonstrated that the LAD to primary sensorimotor cortex does affect the incidence of motor deficits, but LAD to SMA does not. Finally, we observed no relationship between LAD to secondary regions and patient mortality. These results demonstrate that LAD to secondary motor structures does affect morbidity, although not to the extent of LAD to primary structures. This should be considered by neurosurgeons for patient consultation and pre-operative planning.
The authors thank M.E. Meyerand for helping with database maintenance and the IRB protocol, and C.H. Moritz for helping with data acquisition and creating clinical fMRI reports.
Research support, including provision of equipment and materials, was provided UW ICTR NIH/UL1RR025011 Pilot Grant from the Clinical and Translational Science Award (CTSA) program of the National Center for Research Resources (NCRR), RSNA seed grant, and KL2 Scholar Award, and R41 NS081926-01 NIH to VP.
This work has not been presented elsewhere.