Quantifying accuracy and precision of diffusion MR tractography of the corticospinal tract in brain tumors

Clinical article

Full access

Object

The aim of this paper was to validate the diffusion tensor imaging (DTI) model for delineation of the corticospinal tract using cortical and subcortical white matter electrical stimulation for the location of functional motor pathways.

Methods

The authors compare probabilistic versus deterministic DTI fiber tracking by reconstructing the pyramidal fiber tracts on preoperatively acquired DTI in patients with brain tumors. They determined the accuracy and precision of these 2 methods using subcortical stimulation points and the sensitivity using cortical stimulation points. The authors further explored the reliability of these methods by estimation of the potential that the found connections were due to a random chance using a novel neighborhood permutation method.

Results

The probabilistic tracking method delineated tracts that were significantly closer to the stimulation points and was more sensitive than deterministic DTI fiber tracking to define the tracts directed to the motor sites. However, both techniques demonstrated poor sensitivity to finding lateral motor regions.

Conclusions

This study highlights the importance of the validation and quantification of preoperative fiber tracking with the aid of electrophysiological data during the surgery. The poor sensitivity of DTI to delineate lateral motor pathways reported herein suggests that DTI fiber tracking must be used with caution and only as adjunctive data to established methods for motor mapping.

Abbreviations used in this paper:ANNull = architectural neighborhood null; dMRI = diffusion MRI; DTI = diffusion tensor imaging; FA = fractional anisotropy; FSE = fast spin echo; IES = intraoperative electrical stimulation; PDF = probability density function; ROI = region of interest; SNR = signal-to-noise ratio; SPGR = spoiled gradient recalled.

Object

The aim of this paper was to validate the diffusion tensor imaging (DTI) model for delineation of the corticospinal tract using cortical and subcortical white matter electrical stimulation for the location of functional motor pathways.

Methods

The authors compare probabilistic versus deterministic DTI fiber tracking by reconstructing the pyramidal fiber tracts on preoperatively acquired DTI in patients with brain tumors. They determined the accuracy and precision of these 2 methods using subcortical stimulation points and the sensitivity using cortical stimulation points. The authors further explored the reliability of these methods by estimation of the potential that the found connections were due to a random chance using a novel neighborhood permutation method.

Results

The probabilistic tracking method delineated tracts that were significantly closer to the stimulation points and was more sensitive than deterministic DTI fiber tracking to define the tracts directed to the motor sites. However, both techniques demonstrated poor sensitivity to finding lateral motor regions.

Conclusions

This study highlights the importance of the validation and quantification of preoperative fiber tracking with the aid of electrophysiological data during the surgery. The poor sensitivity of DTI to delineate lateral motor pathways reported herein suggests that DTI fiber tracking must be used with caution and only as adjunctive data to established methods for motor mapping.

Diffusion tensor imaging (DTI) is the only MRI technique able to noninvasively localize subcortical white matter pathways in the brain in vivo.1 This technique has recently been included in the presurgical workup to aid in mapping functional fiber pathways and prevent surgical damage.5,6,10,16,22,25,28,29 Brain tumors can dislocate, infiltrate, or destroy the normal course and arrangement of white matter fiber tracts. In all of these conditions, the diffusion signal is altered, and the accuracy and sensitivity of the fiber tracking technique decrease.32 There is no real gold standard with which to establish the true patterns of the fibers' course and, in the literature, there is little direct evidence supporting the validity of the fiber tracking method in the human brain.

Fiber tracking relies on algorithms for diffusion parameter estimation and tract reconstruction. The DTI algorithms are typically performed by integrating along the voxel-wise principal diffusion directions constrained by fractional anisotropy (FA) and angle thresholds.26,27 Standard thresholds may be too high to enable tracking into areas within or around lesions characterized by a low FA.3,23,33 Recently, probabilistic fiber tracking algorithms have been introduced that take into account the uncertainty in the diffusion data and quantify tract probabilities over the distribution of possible orientations.2,31 Probability density functions (PDFs) are constructed on local fiber trajectories by voxel-wise diffusion modeling that accounts for noise and signal ambiguity. In this way, fibers can be traced into areas of ambiguous or even undetermined principal diffusion directions with very low FA values. A recent approach introduced for probabilistic DTI tractography is a nonparametric statistical technique based on data resampling called the bootstrap.12 A model-based resampling approach called residual bootstrap is able to estimate uncertainties with higher accuracy compared with the other bootstrap methods such as the repetition bootstrap.9 Repetition bootstrap has previously been used to provide nonparametric quantification of the uncertainty in the inferred fiber orientation information obtained using the diffusion tensor. This approach requires repeated acquisitions of the diffusion data to ensure that derived distributions of the tensor indices are precise.30 This is impractical for clinical applications because it requires long imaging times. Model-based resampling methods such as the wild and residual bootstrapping approaches require only a single image data set to estimate the uncertainty, based on regression residuals.7–9 Recent studies have shown that the model-based residual bootstrapping methods significantly reduce bias and can provide a more precise estimate of uncertainty compared with the wild bootstrapping.8,9,14,37

Intraoperative electrical stimulation (IES) can be used to determine the accuracy of fiber tracking algorithms. It can also determine the algorithms' sensitivity, but the specificity remains difficult to assess. To provide a quantitative approach to the neurosurgical planning, it is very important to take into account several effects including the fiber tractography technique, the neuronavigation system, the brain shift that causes a spatial discrepancy of the information about the location of the actual fiber bundles, and the spread of the current from intraoperative stimulation that could limit the utility of the DTI during the surgery. Despite these limitations, previous studies have shown that deterministic DTI fiber tracking is capable of predicting the location of positive intraoperative stimulation with high precision and at an offset of about 1 cm.5 Therefore, during surgery IES mapping will tend to find positive stimulations at about 10 ± 3 mm from the preoperatively defined fiber tracts. However, DTI deterministic fiber tracking may also suffer from the failure to predict a portion of the motor pathways. Probabilistic fiber tracking may improve on the performance of deterministic DTI fiber tracking by improving the precision, decreasing the offset to the positive stimulation, and delineating more of the existing motor pathways.

In this work, we provide additional data on the validity of deterministic DTI fiber tracking for utility in brain tumor patients and evaluate the possible improvements provided by probabilistic DTI fiber tracking in the same subjects. Intraoperative white matter and cortical electrical stimulation were used as the gold standard for evaluating the accuracy, predictive power, and sensitivity of deterministic and probabilistic residual bootstrap DTI algorithms with varying seed density and regions of interest (ROIs) for preoperative tractography of the pyramidal fiber tract. Although we could not provide exact measures of specificity (absence of a test for false positives), we explore this issue via a novel null tracking algorithm to evaluate the potential that the apparent connection was due to random chance.

Methods

Magnetic Resonance Imaging

Magnetic resonance images were acquired using a 1.5-T Signa scanner (General Electric Medical Systems) in 21 patients (12 males and 9 females, mean age 42 years) who underwent craniotomies for resection of cerebral gliomas. Diffusion tensor imaging was performed using a single-shot spin-echo echo-planar imaging pulse sequence with a diffusion sensitization (b = 1000 seconds/mm2, TR 10,000 msec, TE 100 msec, slice thickness between 2 and 2.3 mm, no gap between slices, voxel volume between 4.5 and 9 mm3, and an average acquisition of 6 ± 1.5 to improve the signal-to-noise ratio [SNR]). Six diffusion sensitizing gradient directions and 1 image set without diffusion weighting (b = 0 second/mm2) were obtained. Acquisition coverage extended from the cerebral peduncle to the brain vertex. High-resolution T2-weighted and postcontrast T1-weighted anatomical MR images were also acquired for use with the stereotactic surgical navigation system. The T2-weighted images were acquired using an axial fast spin echo (FSE) pulse sequence as follows: TR 3 seconds, TE 105 msec, field of view 260 × 195 mm, matrix 256 × 192, and voxel size 1.02 × 1.02 × 1.5 mm with no gap between slices. The T1-weighted images were acquired with a spoiled gradient recalled (SPGR) sequence as follows: TR 34 msec, TE 3 msec, field of view 260 × 195 mm, matrix 256 × 192, and voxel size 1.02 × 1.02 × 1.5 mm with no gap between slices. The FSE and SPGR acquisition coverage included the entire head and the fiducial markers attached to the head.

The DTI b = 0 echo planar volumes were registered to the FSE and SPGR image volume using a 3D affine 12-parameter model fit. The registration between the echo planar volume and high-resolution anatomical volumes was visually checked with reference to the sulci, ventricles, and brain borders in the cerebrum.

As a general assessment we evaluated the SNR in white matter derived from the b0 volume for each subject, and we calculated its means across the group of subjects. The SNR for each subject was calculated by dividing the mean signal from a 100-pixel ROI drawn on an axial slice in the centrum semiovale by the standard deviation of the noise in a 100-pixel ROI in the background, in a region far from artifacts.19

Intraoperative Electrical Stimulations

The surgeon performed intraoperative electrical subcortical white matter and cortical stimulation of motor neurons during the surgical procedure with a 5-mm-wide bipolar electrode producing a 60-Hz square wave with amplitude between 8 and 12 mA.4 White matter and cortical points that elicited a motor response were stereotactically identified on the anatomical images. The motor responses in muscle groups in the extremities were monitored through electromyographic recordings.36 We defined the stimulation points from the screen shots of the anatomical images in the StealthStation stereotactic surgical navigation system (Medtronic Inc.).

Fiber Tracking

Diffusion MRI fiber tracts of the pyramidal tract were generated postoperatively in these patients using 2 different algorithms: the deterministic based on the Fiber Assignment by Continuous Tracking (FACT) method and the probabilistic residual bootstrap DTI technique.9,26

Briefly, for the deterministic fiber tracking method, from the starting point the algorithm calculates a 3D trajectory in continuous space running parallel to the principal eigenvector. At each voxel, the fiber tract's direction is modified to follow the new voxel's primary eigenvector. The algorithm continues until it reaches a voxel with a value of FA less than 0.15, or turns an angle greater than 60°. This process was repeated with varying seeding density from 3 to 13 (odd numbers); the seed density number n refers to an isotropic grid of n × n × n seed points or n3 points within each voxel of the starting region.

The residual bootstrap DTI method was applied to a single diffusion-weighted data set to estimate the uncertainty of the principal eigenvectors in terms of a probability density function (PDF) for each voxel. Starting regions were densely seeded with 33 to 133 (odd densities only) starting points equally spaced throughout each voxel. Fiber trajectories passing through a voxel were propagated by random sampling from the PDF of the principal eigenvector. The same process was repeated for the next voxel until it reached a voxel with a value of FA less than 0.15 (on a scale of 0–1) or turns an angle greater than 60°.

Pairs of regions were used to delineate the motor pathways by requiring the found streamlines to pass through both regions. For comparison with the subcortical white matter stimulation points, fiber tracts were defined by regions of interest in the cerebral peduncle and primary motor area based on anatomical boundaries. This is the process normally followed for preoperative definition of motor pathways. For evaluation of the sensitivity of the fiber tracking methods, the anatomically defined ROI in the cerebral peduncle was combined with ROIs in the motor cortex defined from the intraoperative cortical motor stimulations. For fiber tracking we created a cube of 1 cm3 centered in the coordinates of the stimulation points to take into account the brain shift that can occur during resection. We used alternatively both the cerebral peduncle and precentral gyrus (or the IES site in the precentral gyrus) as seed regions.

We used the white matter and cortical IES sites to quantitatively assess the performance of the deterministic and probabilistic DTI algorithms. The accuracy and precision of the fiber tract courses within the white matter were assessed by measuring the distances between the stimulation sites and the closest border of the DTI fiber tracts. The ability of the algorithms to predict known functional pathways (sensitivity or true positive rate) was assessed by delineating pathways between the cerebral peduncle and the cortical IES sites.

The tracking programs were written in Interactive Data Language (Exelis Visual Information Solutions). The algorithm was run on a Sun Blade 2500 with dual 1.3-GHz processors (Sun Microsystems).

ANNull Tracking

Randomly connected regions may contribute toward greater sensitivity for probabilistic compared with deterministic algorithms and hence provide a false sense of improved performance. To evaluate the potential that tracts were due to purely random chance, we performed a novel null tracking method for all the tracts connected to the cortical stimulation points. Compared with other null tracking methods, our method is based on the architectural neighborhood null (ANNull) and is motivated by the idea that the rate of generating randomly connected regions will be determined by the fiber architectural neighborhood of the tract of interest. This method does not take the noise magnitude into account and therefore only provides the potential for randomly generating connections.

The rate at which random connections occur in a given tracking experiment is described by a simple model. After tractography from a seed region, we have a number (N) of streamlines for each seed point of the region. Let Nc be the number of connected streamlines obtained from the seed point; the hypothetically true connected streamlines is given by Nt = aNc and is assumed to be proportional to the number of connected seed points; Nr is the number of randomly connected streamlines to the target and is also assumed proportional to the number of connected seeded trajectories so that Nr = RNc; R is the random rate; and the rate of true connections is a = 1 – R.

In a second experiment the random rate R can be determined if all connected streamlines are known to be random so that for seeding with the previously known Nc seeds yields Nc2 connected streamlines, then R = Nc2/Nc. Hence the rate of true connections can be determined by a = 1 – R = (Nc – Nc2)/Nc.

The ANNull algorithm is used to estimate the potential value of R as follows. After performing the first tracking, we permuted the tensor in each voxel in plane along the tract obtained and we performed another tracking. We repeated the same process 60 times for each tract and for all the different density seeding reporting the number of streamlines. This algorithm simulates the condition that the noise is sufficient to propagate to the wrong nearest neighbor. Nc2 was approximated by the average over the 60 permuted trials. We considered robustly connected tracts those for which the number of potentially random fibers connected (Nc2) was smaller than the number obtained without doing the permutation (Nc) so that a > 0.

Statistical Analysis

A statistical analysis was performed to evaluate the effects of the algorithm (probabilistic vs deterministic); the seed region (cerebral peduncles vs precentral gyrus); and the seed density on the sensitivity, accuracy, and precision of the DTI tracts.

For the white matter stimulation points we measured the mean and the standard deviation of the distances from all IES points as a function of seed density, seed ROI, and location from the white matter stimulation point (to the edge and to the center of mass of the tract that connects the cerebral peduncle and the motor area relative to the area in consideration). We performed the statistical analysis both for the mean and the standard deviation by location (edge and center of mass) and by seed ROI (cerebral peduncle and motor area) with method as a factor.

For the cortical stimulation points we defined a connected tract (connectivity = 1) if the number of streamlines was more than 0; otherwise, connectivity = 0. We performed a chi-square test of the connectivity with algorithm (deterministic, probabilistic), ROI (cerebral peduncle and cortical IES), dense seeding (3–13), and region (hand, upper extremity, face/mouth) as factors.

Results

A total of 51 motor stimulation points were identified. These included 16 subcortical motor simulation points (4 face/mouth, 5 upper extremity, 3 hand and fingers, and 4 lower extremity) and 35 cortical motor stimulations points (10 hand and fingers, 6 upper extremity, 19 face/mouth). The characteristics of the brain tumors and their relationship with the motor region is reported in Table 1.

TABLE 1:

Description of the brain tumors in each subject*

Case No.Age (yrs), SexTumorBrain LocationMotor Area InvolvedDimensionEdema
140, Mastrocytoma/IIlt frontalinvolvement/compression of lt CST premotor & motor cortex3.5 × 7.7 × 6.0yes
233, Foligodendroglioma/IIlt frontoparietalinvolvement/compression of lt CST precentral gyrus/MC4.5 × 4.6 × 4.3yes
333, Foligodendroglioma/IIlt temporaldistortion/compression of lt CST at the IC/lower MC (face) level4.5 × 7.9 × 4.0yes
444, Foligoastrocytoma/IIlt frontalinvolvement of SMA & upper-extremity motor area1.7 × 3.7 × 3.1no
540, Fglioblastoma/IVrt frontoparietalinvolvement/compression of rt MC & CST precentral gyrus/motor cortex & compression IC by edema5.0 × 4.8 × 3.2yes
649, Mglioblastoma/IVrt frontoparietalinvolvement/compression of rt MC & CST precentral gyrus/motor cortex & compression IC by edema4.3 × 4.2 × 4.5yes
736, Fastrocytoma/IIIlt frontaldistortion/compression of lt CST at the IC/lower MC (face) level6.7 × 2.8 × 5.3yes
822, Mastrocytoma/IIlt frontalinvolvement/distortion/compression of lt MC/infiltrative edema4.0 × 3.2 × 4.4no
965, Mastrocytoma/IIIrt precentral gyrus, frontal lobeinvolvement/compression of lt MC & CST precentral gyrus/MC1.5 × 1.6 × 2.6yes
1038, Mastrocytoma/IIIlt frontotemporaldistortion/compression of lt CST at the IC level4.9 × 3.4 × 3.9yes
1150, Mglioblastoma/IVrt parietal extending to the temporal lobeinvolvement/distortion/compression of rt CST precentral gyrus/MC6.2 × 5.1 × 4.5yes
1231, Moligoastrocytoma/IIrt precentral gyrus & prefrontal cortexinvolvement of rt upper extremities & face MC4.0 × 4.0 × 3.2yes
1349, Moligoastrocytoma/IIIrt temporal & frontal lobesinvolvement/compression of rt CST in its lower aspects (IC)2.7 × 2.3 × 2.3yes
1428, Mastrocytoma/IIlt frontal lobeno motor/CST involvement4.9 × 3.8 × 3.7no
1549, Mglioblastoma/IVrt frontotemporal lobeinvolvement/compression of rt CST in its lower aspects (IC)5.1 × 4.0 × 3.1yes
1635, Foligodendroglioma/IIlt temporalno motor/CST involvement1.8 × 1.7 × 1.1no
1733, Foligodendroglioma/IIlt parietofrontaldistortion/compression of lt MC3.6 × 4.0 × 3.4yes
1858, Mglioblastoma/IVlt frontoparietalinvolvement/compression of lt CST precentral gyrus/MC4.5 × 3.5 × 4.3yes
1948, Foligodendroglioma/IIrt precentral gyrus & prefrontal cortexinvolvement/distortion/compression of rt CST precentral gyrus/MC5.0 × 4.4 × 4.0yes
2045, Mastrocytoma/IIlt precentral gyrus frontal lobeinvolvement/compression of lt CST precentral gyrus/MC2.3 × 2.3 × 2.3no
2150, Fastrocytoma/IIIlt prefrontalinvolvement/distortion/compression of lt MC/infiltrative edema6.5 × 5.0 × 4.5yes

CST = corticospinal tract; IC = internal capsule; MC = motor cortex; SMA = supplementary motor area.

Given as width × length × height.

White Matter Stimulation

The 16 white matter stimulation points were associated with the nearest DTI fiber tract. In all cases, the nearest DTI fiber tract reflected the functional area stimulated. For example, hand motor white matter stimulation was nearest to a pathway delineated from the anatomically defined hand motor cortex to the cerebral peduncle. A statistically significant difference was found between the deterministic and probabilistic DTI fiber tracking methods in assessing the distance between the edge of DTI tracts and the white matter stimulation sites with the seed ROI as the cerebral peduncle. The probabilistic method delineated tract significantly closer to the stimulation points (mean distance 6.5 ± 2.9 mm) compared with the deterministic method (mean distance 7.2 ± 3.7 mm) (p = 0.02) (Fig. 1). No significant differences were found for seeding from the motor cortex using tract reference location as edge or center of mass.

Fig. 1.
Fig. 1.

Representation of the distance from the edge of the reconstructed corticospinal tract and the white matter subcortical stimulation for the deterministic (left) and probabilistic (right) DTI in 1 subject. As shown in the figure, the reconstructed tract with the probabilistic method is larger than the deterministic tract, so it is closer to the subcortical stimulation point.

Cortical Stimulation

We found a significant difference in sensitivity for the 2 algorithms (deterministic vs probabilistic) (p < 0.0001) with 29% of tracts detected with the deterministic method and 52% of the tracts detected with the probabilistic method. An example of successful probabilistic DTI tracking in a patient is shown in Fig. 2. After the null tracking correction, that is, a lower limit estimate correcting for potential random connections, we still had a significant association between the sensitivity and the algorithm (p < 0.000) with 20% of the tracts detected for the deterministic method and 36% for the probabilistic method. In Table 2 we report the percentage of connections found for each motor region over the total of the connections per region before and after the ANNulled estimates, for deterministic and probabilistic methods. ANNulled estimates represent a lower limit because of potential random connections. There was a trend for increasing sensitivity with increasing seeding density up to a seeding density of 7. However, for seeding in the cerebral peduncle, the ANNulled estimates were largely independent of seed density and consistently matched the sensitivity values found with the lowest seed density without correction (Fig. 3). When seeding from the IES, however, there was less impact of the ANNulled procedure on sensitivity, as is evident in Fig. 4.

Fig. 2.
Fig. 2.

Representation of the delineation of the pathway of the corticospinal tract assessed from the cortical stimulation point of the hand motor area in 1 subject. As shown in the figure, it was not possible to reconstruct the tract with the deterministic method (left) but it was possible with the probabilistic method (right).

TABLE 2:

Percentage of IES detected tracks found by DTI tractography (sensitivity) for each motor region*

RegionDeterministicProbabilistic
face27% (25%)43% (37%)
hand18% (9%)55% (38%)
upper extremity53% (21%)79% (32%)

ANNulled estimates appear in parentheses.

Overall connections for each region.

Fig. 3.
Fig. 3.

The percentage of connections of the tracking from the cerebral peduncle varying the dense seeding for both methods. The ANNulled limit estimates are shown in the lower panel. Color gradients represent different levels of seeding density increasing from 3 to 13 by odd numbers. Blue represents the results obtained with the DTI deterministic method, and orange represents the results obtained with the DTI probabilistic method. UE = upper extremity.

Fig. 4.
Fig. 4.

The percentage of connections of the tracking from the IES varying the dense seeding for both methods. The ANNulled limit estimates are shown in the lower panel. Color gradients represent different levels of seeding density increasing from 3 to 13 by odd numbers. Blue represents the results obtained with the DTI deterministic method and orange with the DTI probabilistic method.

We show the percentage of known connections found for tracking from the cerebral peduncle (Fig. 3) and the IES (Fig. 4) with varying seeding density for both fiber tracking methods. The upper panels show the results and at the lower panels show the ANNulled estimates. No significant dependences of sensitivity on seed ROI or seeding density were found.

Discussion

The present study provides a rare assessment of the performance of diffusion MRI (dMRI) fiber tracking for the corticospinal tract. Furthermore, these results directly measure the impact of the choice of algorithm and diffusion modeling on the application of this method for fiber tracking in patients with brain tumors. In short, we demonstrate that the DTI model has poor sensitivity to estimate the full extent of the corticospinal tract, but the use of probabilistic algorithms does improve the sensitivity (Fig. 2). We also show that at a coarse level, the DTI fiber tracking does provide accurate estimates of the white matter course of the corticospinal tract even in the presence of brain tumors.

Accuracy and Reliability of DTI Fiber Tracking

Without true gold standards it has been difficult to obtain realistic measures of the performance of dMRI fiber tracking for in vivo imaging in normal and pathological conditions. There are almost no studies to date that estimate the accuracy of dMRI fiber tracking and rare attempts at determination of reliability in vivo, especially in the presence of pathology. Simulation studies provide some level of understanding of the dependencies on acquisition schemes, diffusion modeling, and algorithms but cannot provide estimates in realistic in vivo cases. Furthermore, the performance of dMRI fiber tracking in terms of accuracy and reliability depends strongly on the specific pathways under consideration and cannot be decided from general arguments. Factors that influence performance include the acquisition scheme (SNR, resolution, number of unique diffusion sensitizing directions, diffusion weighting, and other factors that affect the image distortion and motion artifacts), diffusion model (such as DTI, QBall, diffusion spectrum imaging, and spherical deconvolution), and algorithm/approach (deterministic and stochastic streamlines, nonstreamline approaches, and global fiber tracking). Here we have used intraoperative stimulation data to provide estimates of the accuracy and sensitivity of DTI modeling with deterministic and probabilistic algorithms for the delineation of the corticospinal tract in patients with brain tumors.

Although intraoperative stimulation may not accurately represent the preoperative conditions, it remains the clinical gold standard method. The distance measurements performed in this study include all inherent errors between the stimulation point observed in the neuronavigation system and the real point stimulated. These errors are due to different phenomena that occur during the resection such as brain shift, electrical current, and stereotactic localization errors.

The acquisition scheme used included 6 ± 1.5 averages of 6 unique diffusion-weighted directions at b = 1000 seconds/mm2 with an SNR of 31 ± 9 on the minimally diffusion-weighted image (± SD). Simulation studies suggest that the SNR is the strongest determinant of reliability, with the number of unique diffusion-weighted directions also playing an important role.18 For this combination of SNR and number of directions, the theoretical reliability is very good on the estimates of the diffusion principal directions given the FA of our voxels between 0.4 and 0.6 in the corticospinal tract. In practice, the lower-limb portion of the corticospinal tract can be estimated with 100% reliability using such data in vivo since this pathway has limited detrimental influence from crossing fiber populations.

Our results using cortical stimulation points demonstrate the poor sensitivity of the diffusion model to delineate the lateral pathways of the corticospinal tract in these brain tumor patients. While this is improved with the probabilistic streamline algorithm, the sensitivity remains low. Due to tissue shift and electrical current penetration effects, the current study is not a perfect test of the white matter course of the DTI streamlines to estimate the corticospinal tract. Nonetheless, we find that the nearest streamlines (offset typically less than 1 cm away) do correspond to functional intraoperative cortical homunculi. This offset is expected but does not eliminate the possibility that nearer axonal bundles that follow the same general course were not delineated by the DTI fiber tracking.

Algorithmic factors that affect the DTI fiber tracking results in addition to probabilistic versus deterministic algorithms include seed density and location, choice of anatomical constraints, and stopping constraints. In this study we used anatomical constraints in the cerebral peduncle and motor cortices since these are the well-known limits of the corticospinal tract in the brain. Using these anatomical constraints seems to provide specific delineation of the corticospinal tract as indicated by this study. Most brain tumors occur in the subcortical white matter and therefore it is typically easy to define an ROI in the cerebral peduncle, while sometimes the motor cortex may be distorted from its normal appearance due to the lesion mass effect. No influence of seeding strategy (cerebral peduncle vs motor cortex) was found in this study. Multiple seeds were placed for streamlines, and the number of seeds was evaluated for its impact on reliability/sensitivity. The reliability was seen to improve up to a seed density of 73.

The FA threshold represents an essential parameter for termination of the tracking procedure; another approach is to use a mask of the white matter or brain tissue. In a recent work, it was shown that the distance of the reconstructed fiber bundles to the pathological signal change increases with higher FA threshold.33 We chose a low FA threshold to be sure to reconstruct the fibers also in cases of tumor infiltration and edema. However, in these regions the propagation directions are relatively unreliable with a large variance. Probabilistic approaches are better able to handle voxels with unreliable estimates of the true propagation direction since the deterministic approach will only use a single estimate that is likely to be unduly affected by noise and biased compared with the true direction and will therefore have diminished capability to delineate these pathways. However, there could also be increased apparent ability to delineate fiber pathways with the probabilistic method due to random connections between voxels. The ANNull tracking method provides an estimate of the degree to which random connections may influence the results of fiber tracking, and even for this estimate, the probabilistic approach still yielded improved sensitivity over the deterministic approach. The ANNull approach differs from previous null tracking methods in that it explicitly investigates the degree to which the neighboring architecture can influence random connections. This is an important element since a fiber pathway that is bounded by orthogonal pathways is unlikely to produce a randomly generated pathway that jumps from the true to orthogonal to true pathway. However, a pathway that has similarly oriented pathways at its boundary is very likely to produce erroneous pathways that jump between bundles and lead to false connectivity.

The choice of the threshold could also influence the representation of the size bundle. Electrical stimulation sometimes evokes responses more than 0.5–1 cm away from the estimated tract.5,25 A recent study reduced such mismatches to less than 5% of the cases by lowering the FA threshold for streamline,3 but that simply expands the tracts within the same deterministic estimation. It is reasonable to assume that the probabilistic DTI reconstruction could have an effect to expand the size of the tract, but in the subcortical analysis we showed that the mean distance from the edge was significantly closer only from the cerebral peduncle, and the standard deviation was smaller than the deterministic method, and we did not find any difference in any other case. Moreover, it is well known that tractography underestimates fiber bundle size.21 In that work the authors attempted a preliminary validation of the deterministic tractography and concluded that while tractography can visualize the direction of the fibers, it is less informative with respect to the actual size of the fiber bundle. Therefore, there is an underestimation of the fiber bundle that has to be taken under consideration. The improved matching of probabilistic DTI streamlines to intraoperative mapping may represent better fiber bundle size estimates or an overestimation of the bundle size that effectively cancels some of the expected offset from the stimulation point to the preoperative streamlines.

Implications for Mapping the Corticospinal Tract in Brain Tumor Patients

Brain tumors can involve eloquent areas in both the cortex and subcortical white matter tracts. Knowledge of the location of these eloquent tissues impacts both surgical morbidity and efficacy; precise knowledge of the locations of these eloquent tissues allows for aggressive resection of tumor that has been shown to be the most effective treatment. The clinical gold standard for localization of eloquent tissue is IES. However, this mapping process is laborious and can be improved with in vivo noninvasive preoperative mapping. Several technical advances have been proposed to preoperatively map functional eloquent cortex (for example, functional MRI and magnetoencephalography), but only dMRI currently provides a means to visualize the course of white matter fiber tracts. Recently, DTI-based fiber tractography has been shown to be very useful in visualizing in vivo white matter bundles during neurosurgical planning.5,6,11,16,17,25,34,35 Although powerful due to its unique capability for delineation of white matter bundles, DTI has several known limitations in terms of accuracy and precision, and its use, particularly in brain tumor patients in whom the normal architecture is distorted, must be approached with caution. Although several studies have emphasized its clinical usefulness, few studies have attempted validation of tractography in the context of predicting eloquent pathways for neurosurgery, and optimization of dMRI algorithms for preoperative fiber tracking remains unexplored.10,24 The results reported here represent one of the few attempts at such validation and optimization of preoperative diffusion fiber tracking for intraoperative mapping.

Although the intraoperative coordinates during resection are affected by several effects, such as brain shift, this procedure is the only one available in the present conditions. We took into account all the possible limitations, and we considered an area of at least 1 cm3 for the cortical stimulation point.

In this study, the use of IES was able to give an important contribution to quantify the accuracy and precision of preoperative dMRI fiber tracking methods. In particular, using the cortical motor stimulation sites, we were able to evaluate the number of false negatives for diffusion fiber tracking, and using white matter stimulation points, we were able to assess the accuracy and precision of preoperative dMRI fiber tracking to predict the course of motor fiber tracts. The ANNull tracking was also introduced to probe the potential influence of random connections on the probabilistic and deterministic algorithms. We have used these metrics to compare deterministic and probabilistic DTI methods and show that probabilistic DTI has fewer false negatives (better sensitivity) than deterministic DTI fiber tracking in defining the corticospinal tracts (Fig. 2) and more accurate and precise predictions of the white matter course of motor pathways (Fig. 1). These data also confirm an earlier study that suggested strong predictive power for the white matter course of the pathways that are found.6 However, herein we demonstrate that both DTI probabilistic and deterministic algorithms have poor overall sensitivity to delineate motor pathways.

Impact on Intraoperative Guidance

To discuss the relevance of our findings, it is necessary to first describe the use of preoperative dMRI fiber tracking. This modality is currently used for preoperative planning of resections and for guiding intraoperative mapping with electrical stimulation. Mapping with IES is laborious and is usually performed by stimulation mapping across a virtual grid of points spaced by 1 cm. The standard of care dictates that when an eloquent structure is stimulated intraoperatively, the resection is halted. With dMRI fiber tracking maps, the neurosurgeon is able to focus on the stimulation mapping, thereby reducing the mapping time and potentially also improving detection of eloquent structures. For this latter purpose, the ability of preoperative dMRI fiber tracking to predict the location of the eloquent tissue is of paramount importance. Various factors potentially contribute to the accuracy and precision of preoperative diffusion fiber tracking to predict the white matter course of at-risk eloquent pathways. These factors are tissue shifts during resection, spread of the electrical stimulation current, and preoperative mapping and registration errors. In particular, tissue shifts and current spread lead to inaccuracies in the intraoperative correspondence between preoperative diffusion fiber tracts and intraoperative stimulation.5 For deep white matter this is a positive offset of about 9 mm due to current spread, the movement of tissue toward the resection cavity, and the inability of diffusion fiber tracking to estimate the full extent of the fiber tract. However, predicting the locations of positive intraoperative stimulations depends more on the variance or precision of the predicted locations than the absolute accuracy. For example, with zero variance, the surgeon would know that stimulating 9 mm before reaching the predicted fiber tract locations will result in a positive stimulation. Therefore, the variance in the predicted locations will reflect the predictive power of a dMRI method. In the present work we found a mean distance (± SD) of 7.2 ± 3.7 mm with the deterministic DTI and of 6.5 ± 2.9 mm with the probabilistic DTI, suggesting less bias and more precision with the probabilistic method.

There are other important aspects that have to be taken into consideration. First, it has been shown that the localization error of a frameless surgical navigation system is typically less than 1.5 mm.13,20 Second, intraoperative electric responses are affected by the current spread, electrical conductivity, and resistance that result in errors in electrical stimulations. The electrical current penetrates tissue around the electrodes, while with the neuronavigation system we are able to visualize only a single point related to the localization of the electrodes. We know from the literature that the range of stimulation in the cortex is approximately 5–10 mm,15 but we do not know anything about the underlying white matter. Third, resection can often be accompanied by substantial brain shifts. Sometimes, shifting is determined by relief from the space occupancy of the underlying mass, but its consistency, intraoperative swelling, and other factors turn it into an entity whose size, and direction, cannot easily be predicted. In a study using intraoperative MRI and DTI fiber tractography, Nimsky et al. showed a mean white matter tract shifting of 2.7 mm.28 Intraoperatively acquired images are required for referencing brain shift to any edge or intraaxial point of reference and are possibly entirely sufficient to integrate preoperative fiber tracking analysis into the neuronavigation system. All of these aspects could make the distance between white matter stimulation and the fiber tract difficult to perceive and in this way preserve the white matter bundles from damage.

Fiber tracking, and in particular fiber tracking with DTI, presents several limitations. Errors in the estimation of the fiber tracts can be caused by low SNRs, the selection of the seed ROIs, the choice of the threshold for the anisotropy, the diffusion model (DTI in this case), the fiber tracking algorithm, or the complexity and nature of the fiber architecture, especially the presence of crossing fibers.6,10,25 Moreover, all these aspects became a challenge in the presence of edema and mass effect that affect the diffusion signal.

For neurosurgical planning, the most widely used for the reconstruction of the fibers is deterministic DTI fiber tracking along the principal direction of the tensor. The major limitation of this method is to reconstruct the tract through regions of crossing fibers, resulting in an inaccurate depiction of the motor tract.5,21,25 Motor tracts present a fan-shape configuration at the level of the centrum semiovale. Deterministic DTI is able to distinguish only the principal direction and to depict the fibers traveling to the vertex of the brain lacking of a measure describing confidence or uncertainty of the reconstructed trajectories. More accurate models for estimation of the fibers like QBall and spherical deconvolution could provide an improved solution to this problem. The use of the probabilistic DTI algorithm has been shown here to improved sensitivity and predictive power to determine white matter course of fiber. The results shown here suggest excellent capability for preoperative fiber tracking to predict the white matter course of fiber pathways for the detected all the motor pathways with the DTI model.

Conclusions

Diffusion tensor imaging fiber tracking has been shown to have poor sensitivity to delineate the lateral aspects of the corticospinal tract in patients with brain tumors. Using a probabilistic approach provides some limited improvement in sensitivity. There was very good accuracy of DTI streamlines to predict the white matter course of corticospinal tract bundles. For the validation and improvement of fiber tracking algorithms for neurosurgical planning, the use of electrophysiological data and functional image guidance can be very useful, especially under difficult pathological conditions. Understanding the limitations of diffusion fiber tractography methods for preoperative mapping is vital for safe, reliable, and efficient use of this technique and for optimization of the algorithm for surgical applications.

Disclosure

Dr. Henry is a consultant for Stem Cells Inc., and he received grants from the NIH (Grant Nos. R01 NS066654 and R21 CA121561).

Author contributions to the study and manuscript preparation include the following. Conception and design: all authors. Acquisition of data: Mandelli, Berger, Berman, Henry. Analysis and interpretation of data: all authors. Drafting the article: Mandelli, Bucci, Henry. Critically revising the article: all authors. Reviewed submitted version of manuscript: all authors. Approved the final version of the manuscript on behalf of all authors: Mandelli. Statistical analysis: Mandelli, Henry. Study supervision: Berger, Henry.

This article contains some figures that are displayed in color online but in black-and-white in the print edition.

References

  • 1

    Basser PJPajevic SPierpaoli CDuda JAldroubi A: In vivo fiber tractography using DT-MRI data. Magn Reson Med 44:6256322000

  • 2

    Behrens TEWoolrich MWJenkinson MJohansen-Berg HNunes RGClare S: Characterization and propagation of uncertainty in diffusion-weighted MR imaging. Magn Reson Med 50:107710882003

  • 3

    Bello LGambini ACastellano ACarrabba GAcerbi FFava E: Motor and language DTI fiber tracking combined with intraoperative subcortical mapping for surgical removal of gliomas. Neuroimage 39:3693822008

  • 4

    Berger MSOjemann GATechniques of functional localization during removal of tumors involving the cerebral hemisphere. Loftus CMTraynelis VC: Intraoperative Monitoring Techniques in Neurosurgery New YorkMcGraw-Hill1994. 113127

  • 5

    Berman JIBerger MSChung SWNagarajan SSHenry RG: Accuracy of diffusion tensor magnetic resonance imaging tractography assessed using intraoperative subcortical stimulation mapping and magnetic source imaging. J Neurosurg 107:4884942007

  • 6

    Berman JIBerger MSMukherjee PHenry RG: Diffusion-tensor imaging-guided tracking of fibers of the pyramidal tract combined with intraoperative cortical stimulation mapping in patients with gliomas. J Neurosurg 101:66722004

  • 7

    Berman JIChung SMukherjee PHess CPHan ETHenry RG: Probabilistic streamline q-ball tractography using the residual bootstrap. Neuroimage 39:2152222008

  • 8

    Chung SBerman JIRae CHenry RG: Effect of DTI bootstrap bias on the DTI uncertainty measurements and probabilistic tractography. Proc Intl Soc Mag Reson Med 15:15962007

  • 9

    Chung SLu YHenry RG: Comparison of bootstrap approaches for estimation of uncertainties of DTI parameters. Neuroimage 33:5315412006

  • 10

    Clark CABarrick TRMurphy MMBell BA: White matter fiber tracking in patients with space-occupying lesions of the brain: a new technique for neurosurgical planning?. Neuroimage 20:160116082003

  • 11

    Coenen VAKrings TMayfrank LPolin RSReinges MHThron A: Three-dimensional visualization of the pyramidal tract in a neuronavigation system during brain tumor surgery: first experiences and technical note. Neurosurgery 49:86932001

  • 12

    Efron BTibshirani RJ: An Introduction to the Bootstrap LondonChapman and Hall1993

  • 13

    Grunert PDarabi KEspinosa JFilippi R: Computer-aided navigation in neurosurgery. Neurosurg Rev 26:731012003

  • 14

    Haroon HAMorris DMEmbleton KVAlexander DCParker GJ: Using the model-based residual bootstrap to quantify uncertainty in fiber orientations from Q-ball analysis. IEEE Trans Med Imaging 28:5355502009

  • 15

    Heiervang EBehrens TEMackay CERobson MDJohansen-Berg H: Between session reproducibility and between subject variability of diffusion MR and tractography measures. Neuroimage 33:8678772006

  • 16

    Henry RGBerman JINagarajan SSMukherjee PBerger MS: Subcortical pathways serving cortical language sites: initial experience with diffusion tensor imaging fiber tracking combined with intraoperative language mapping. Neuroimage 21:6166222004

  • 17

    Holodny AISchwartz THOllenschleger MLiu WCSchulder M: Tumor involvement of the corticospinal tract: diffusion magnetic resonance tractography with intraoperative correlation. Case illustration. J Neurosurg 95:10822001

  • 18

    Jones DK: The effect of gradient sampling schemes on measures derived from diffusion tensor MRI: a Monte Carlo study. Magn Reson Med 51:8078152004

  • 19

    Kaufman LKramer DMCrooks LEOrtendahl DA: Measuring signal-to-noise ratios in MR imaging. Radiology 173:2652671989

  • 20

    Keles GELundin DALamborn KRChang EFOjemann GBerger MS: Intraoperative subcortical stimulation mapping for hemispherical perirolandic gliomas located within or adjacent to the descending motor pathways: evaluation of morbidity and assessment of functional outcome in 294 patients. J Neurosurg 100:3693752004

  • 21

    Kinoshita MYamada KHashimoto NKato AIzumoto SBaba T: Fiber-tracking does not accurately estimate size of fiber bundle in pathological condition: initial neurosurgical experience using neuronavigation and subcortical white matter stimulation. Neuroimage 25:4244292005

  • 22

    Krings TCoenen VAAxer HReinges MHHöller Mvon Keyserlingk DG: In vivo 3D visualization of normal pyramidal tracts in human subjects using diffusion weighted magnetic resonance imaging and a neuronavigation system. Neurosci Lett 307:1921962001

  • 23

    Kunimatsu AAoki SMasutani YAbe OHayashi NMori H: The optimal trackability threshold of fractional anisotropy for diffusion tensor tractography of the corticospinal tract. Magn Reson Med Sci 3:11172004

  • 24

    Leclercq DDelmaire Cde Champfleur NMChiras JLehéricy S: Diffusion tractography: methods, validation and applications in patients with neurosurgical lesions. Neurosurg Clin N Am 22:253268ix2011

  • 25

    Mikuni NOkada TNishida NTaki JEnatsu RIkeda A: Comparison between motor evoked potential recording and fiber tracking for estimating pyramidal tracts near brain tumors. J Neurosurg 106:1281332007

  • 26

    Mori SCrain BJChacko VPvan Zijl PC: Three-dimensional tracking of axonal projections in the brain by magnetic resonance imaging. Ann Neurol 45:2652691999

  • 27

    Mori Svan Zijl PC: Fiber tracking: principles and strategies – a technical review. NMR Biomed 15:4684802002

  • 28

    Nimsky CGanslandt OHastreiter PWang RBenner TSorensen AG: Preoperative and intraoperative diffusion tensor imaging-based fiber tracking in glioma surgery. Neurosurgery 56:1301382005

  • 29

    Nimsky CGanslandt OMerhof DSorensen AGFahlbusch R: Intraoperative visualization of the pyramidal tract by diffusion-tensor-imaging-based fiber tracking. Neuroimage 30:121912292006

  • 30

    O'Gorman RLJones DK: Just how much data need to be collected for reliable bootstrap DT-MRI?. Magn Reson Med 56:8848902006

  • 31

    Parker GJHaroon HAWheeler-Kingshott CA: A framework for a streamline-based probabilistic index of connectivity (PICo) using a structural interpretation of MRI diffusion measurements. J Magn Reson Imaging 18:2422542003

  • 32

    Schonberg TPianka PHendler TPasternak OAssaf Y: Characterization of displaced white matter by brain tumors using combined DTI and fMRI. Neuroimage 30:110011112006

  • 33

    Stadlbauer ANimsky CBuslei RSalomonowitz EHammen TBuchfelder M: Diffusion tensor imaging and optimized fiber tracking in glioma patients: histopathologic evaluation of tumor-invaded white matter structures. Neuroimage 34:9499562007

  • 34

    Wieshmann UCSymms MRParker GJClark CALemieux LBarker GJ: Diffusion tensor imaging demonstrates deviation of fibres in normal appearing white matter adjacent to a brain tumour. J Neurol Neurosurg Psychiatry 68:5015032000

  • 35

    Witwer BPMoftakhar RHasan KMDeshmukh PHaughton VField A: Diffusion-tensor imaging of white matter tracts in patients with cerebral neoplasm. J Neurosurg 97:5685752002

  • 36

    Yingling CDOjemann SDodson BHarrington MJBerger MS: Identification of motor pathways during tumor surgery facilitated by multichannel electromyographic recording. J Neurosurg 91:9229271999

  • 37

    Zhu TLiu XConnelly PRZhong J: An optimized wild bootstrap method for evaluation of measurement uncertainties of DTI-derived parameters in human brain. Neuroimage 40:114411562008

If the inline PDF is not rendering correctly, you can download the PDF file here.

Article Information

Address correspondence to: Maria Luisa Mandelli, Ph.D., Department of Neurology, UCSF, 675 Nelson Rising Ln., San Francisco, CA 94158. email: marialuisa.mandelli@ucsf.edu.

Please include this information when citing this paper: published online June 6, 2014; DOI: 10.3171/2014.4.JNS131160.

© AANS, except where prohibited by US copyright law.

Headings

Figures

  • View in gallery

    Representation of the distance from the edge of the reconstructed corticospinal tract and the white matter subcortical stimulation for the deterministic (left) and probabilistic (right) DTI in 1 subject. As shown in the figure, the reconstructed tract with the probabilistic method is larger than the deterministic tract, so it is closer to the subcortical stimulation point.

  • View in gallery

    Representation of the delineation of the pathway of the corticospinal tract assessed from the cortical stimulation point of the hand motor area in 1 subject. As shown in the figure, it was not possible to reconstruct the tract with the deterministic method (left) but it was possible with the probabilistic method (right).

  • View in gallery

    The percentage of connections of the tracking from the cerebral peduncle varying the dense seeding for both methods. The ANNulled limit estimates are shown in the lower panel. Color gradients represent different levels of seeding density increasing from 3 to 13 by odd numbers. Blue represents the results obtained with the DTI deterministic method, and orange represents the results obtained with the DTI probabilistic method. UE = upper extremity.

  • View in gallery

    The percentage of connections of the tracking from the IES varying the dense seeding for both methods. The ANNulled limit estimates are shown in the lower panel. Color gradients represent different levels of seeding density increasing from 3 to 13 by odd numbers. Blue represents the results obtained with the DTI deterministic method and orange with the DTI probabilistic method.

References

  • 1

    Basser PJPajevic SPierpaoli CDuda JAldroubi A: In vivo fiber tractography using DT-MRI data. Magn Reson Med 44:6256322000

  • 2

    Behrens TEWoolrich MWJenkinson MJohansen-Berg HNunes RGClare S: Characterization and propagation of uncertainty in diffusion-weighted MR imaging. Magn Reson Med 50:107710882003

  • 3

    Bello LGambini ACastellano ACarrabba GAcerbi FFava E: Motor and language DTI fiber tracking combined with intraoperative subcortical mapping for surgical removal of gliomas. Neuroimage 39:3693822008

  • 4

    Berger MSOjemann GATechniques of functional localization during removal of tumors involving the cerebral hemisphere. Loftus CMTraynelis VC: Intraoperative Monitoring Techniques in Neurosurgery New YorkMcGraw-Hill1994. 113127

  • 5

    Berman JIBerger MSChung SWNagarajan SSHenry RG: Accuracy of diffusion tensor magnetic resonance imaging tractography assessed using intraoperative subcortical stimulation mapping and magnetic source imaging. J Neurosurg 107:4884942007

  • 6

    Berman JIBerger MSMukherjee PHenry RG: Diffusion-tensor imaging-guided tracking of fibers of the pyramidal tract combined with intraoperative cortical stimulation mapping in patients with gliomas. J Neurosurg 101:66722004

  • 7

    Berman JIChung SMukherjee PHess CPHan ETHenry RG: Probabilistic streamline q-ball tractography using the residual bootstrap. Neuroimage 39:2152222008

  • 8

    Chung SBerman JIRae CHenry RG: Effect of DTI bootstrap bias on the DTI uncertainty measurements and probabilistic tractography. Proc Intl Soc Mag Reson Med 15:15962007

  • 9

    Chung SLu YHenry RG: Comparison of bootstrap approaches for estimation of uncertainties of DTI parameters. Neuroimage 33:5315412006

  • 10

    Clark CABarrick TRMurphy MMBell BA: White matter fiber tracking in patients with space-occupying lesions of the brain: a new technique for neurosurgical planning?. Neuroimage 20:160116082003

  • 11

    Coenen VAKrings TMayfrank LPolin RSReinges MHThron A: Three-dimensional visualization of the pyramidal tract in a neuronavigation system during brain tumor surgery: first experiences and technical note. Neurosurgery 49:86932001

  • 12

    Efron BTibshirani RJ: An Introduction to the Bootstrap LondonChapman and Hall1993

  • 13

    Grunert PDarabi KEspinosa JFilippi R: Computer-aided navigation in neurosurgery. Neurosurg Rev 26:731012003

  • 14

    Haroon HAMorris DMEmbleton KVAlexander DCParker GJ: Using the model-based residual bootstrap to quantify uncertainty in fiber orientations from Q-ball analysis. IEEE Trans Med Imaging 28:5355502009

  • 15

    Heiervang EBehrens TEMackay CERobson MDJohansen-Berg H: Between session reproducibility and between subject variability of diffusion MR and tractography measures. Neuroimage 33:8678772006

  • 16

    Henry RGBerman JINagarajan SSMukherjee PBerger MS: Subcortical pathways serving cortical language sites: initial experience with diffusion tensor imaging fiber tracking combined with intraoperative language mapping. Neuroimage 21:6166222004

  • 17

    Holodny AISchwartz THOllenschleger MLiu WCSchulder M: Tumor involvement of the corticospinal tract: diffusion magnetic resonance tractography with intraoperative correlation. Case illustration. J Neurosurg 95:10822001

  • 18

    Jones DK: The effect of gradient sampling schemes on measures derived from diffusion tensor MRI: a Monte Carlo study. Magn Reson Med 51:8078152004

  • 19

    Kaufman LKramer DMCrooks LEOrtendahl DA: Measuring signal-to-noise ratios in MR imaging. Radiology 173:2652671989

  • 20

    Keles GELundin DALamborn KRChang EFOjemann GBerger MS: Intraoperative subcortical stimulation mapping for hemispherical perirolandic gliomas located within or adjacent to the descending motor pathways: evaluation of morbidity and assessment of functional outcome in 294 patients. J Neurosurg 100:3693752004

  • 21

    Kinoshita MYamada KHashimoto NKato AIzumoto SBaba T: Fiber-tracking does not accurately estimate size of fiber bundle in pathological condition: initial neurosurgical experience using neuronavigation and subcortical white matter stimulation. Neuroimage 25:4244292005

  • 22

    Krings TCoenen VAAxer HReinges MHHöller Mvon Keyserlingk DG: In vivo 3D visualization of normal pyramidal tracts in human subjects using diffusion weighted magnetic resonance imaging and a neuronavigation system. Neurosci Lett 307:1921962001

  • 23

    Kunimatsu AAoki SMasutani YAbe OHayashi NMori H: The optimal trackability threshold of fractional anisotropy for diffusion tensor tractography of the corticospinal tract. Magn Reson Med Sci 3:11172004

  • 24

    Leclercq DDelmaire Cde Champfleur NMChiras JLehéricy S: Diffusion tractography: methods, validation and applications in patients with neurosurgical lesions. Neurosurg Clin N Am 22:253268ix2011

  • 25

    Mikuni NOkada TNishida NTaki JEnatsu RIkeda A: Comparison between motor evoked potential recording and fiber tracking for estimating pyramidal tracts near brain tumors. J Neurosurg 106:1281332007

  • 26

    Mori SCrain BJChacko VPvan Zijl PC: Three-dimensional tracking of axonal projections in the brain by magnetic resonance imaging. Ann Neurol 45:2652691999

  • 27

    Mori Svan Zijl PC: Fiber tracking: principles and strategies – a technical review. NMR Biomed 15:4684802002

  • 28

    Nimsky CGanslandt OHastreiter PWang RBenner TSorensen AG: Preoperative and intraoperative diffusion tensor imaging-based fiber tracking in glioma surgery. Neurosurgery 56:1301382005

  • 29

    Nimsky CGanslandt OMerhof DSorensen AGFahlbusch R: Intraoperative visualization of the pyramidal tract by diffusion-tensor-imaging-based fiber tracking. Neuroimage 30:121912292006

  • 30

    O'Gorman RLJones DK: Just how much data need to be collected for reliable bootstrap DT-MRI?. Magn Reson Med 56:8848902006

  • 31

    Parker GJHaroon HAWheeler-Kingshott CA: A framework for a streamline-based probabilistic index of connectivity (PICo) using a structural interpretation of MRI diffusion measurements. J Magn Reson Imaging 18:2422542003

  • 32

    Schonberg TPianka PHendler TPasternak OAssaf Y: Characterization of displaced white matter by brain tumors using combined DTI and fMRI. Neuroimage 30:110011112006

  • 33

    Stadlbauer ANimsky CBuslei RSalomonowitz EHammen TBuchfelder M: Diffusion tensor imaging and optimized fiber tracking in glioma patients: histopathologic evaluation of tumor-invaded white matter structures. Neuroimage 34:9499562007

  • 34

    Wieshmann UCSymms MRParker GJClark CALemieux LBarker GJ: Diffusion tensor imaging demonstrates deviation of fibres in normal appearing white matter adjacent to a brain tumour. J Neurol Neurosurg Psychiatry 68:5015032000

  • 35

    Witwer BPMoftakhar RHasan KMDeshmukh PHaughton VField A: Diffusion-tensor imaging of white matter tracts in patients with cerebral neoplasm. J Neurosurg 97:5685752002

  • 36

    Yingling CDOjemann SDodson BHarrington MJBerger MS: Identification of motor pathways during tumor surgery facilitated by multichannel electromyographic recording. J Neurosurg 91:9229271999

  • 37

    Zhu TLiu XConnelly PRZhong J: An optimized wild bootstrap method for evaluation of measurement uncertainties of DTI-derived parameters in human brain. Neuroimage 40:114411562008

TrendMD

Cited By

Metrics

Metrics

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 304 304 19
PDF Downloads 210 210 16
EPUB Downloads 0 0 0

PubMed

Google Scholar