Multi-Band Diffusion Tensor Imaging for Presurgical Mapping of Motor and Language Pathways in Patients with Brain Tumors

Abstract

Background and Purpose:

Assessment of the essential white matter fibers of arcuate fasciculus and corticospinal tract (CST) required for preoperative planning in brain tumor patients, relies on the reliability of diffusion tensor imaging (DTI). The recent development of multi-band DTI (mb-DTI) based on simultaneous multi-slice excitation could maintain the overall quality of tractography while not exceeding standard clinical care time. To address this potential, we performed quantitative analyses to evaluate tractography results of arcuate fasciculus and CST acquired by mb-DTI in brain tumor patients.

Methods:

We retrospectively analyzed 44 patients with brain lesions who underwent presurgical single-shot DTI (s-DTI) and mb-DTI. We measured DTI parameters: fractional anisotropy (FA) and mean diffusivity (MD_(mm⁻²_s⁻¹₎) in whole brain and tumor regions; and the tractography parameters fiber FA, MD_(mm⁻²_s⁻¹_), volume _(mm³₎ and length_(mm) in the whole brain, arcuate fasciculus , and CST. Additionally, three neuroradiologists performed a blinded visual assessment comparing s-DTI with mb-DTI.

Results:

The mb-DTI showed higher mean FA and lower MD (r>0.95, p<0.002) in whole brain and tumor ROIs; slightly higher fiber FA, volume, and length; and slightly lower fiber MD in whole brain, arcuate fasciculus, and CST than in s-DTI. These differences were significant for fiber FA in all tracts; length_(mm) in arcuate fasciculus and fiber MD_(mm⁻²_s⁻¹₎ and volume_(mm³₎ in all patients with tumor involved in the arcuate fasciculus, CST, and whole brain tracts (p=0.001). Visual assessment demonstrated that both techniques produced visually similar tracts.

Conclusions:

This study demonstrated the clinical potential and significant advantages of preoperative mb-DTI in brain tumor patients.

INTRODUCTION

Diffusion tensor imaging (DTI) provides information about the architecture and orientation of white matter tracts by using a single covariance matrix to model the diffusivity of water molecules based on the Gaussian assumption.^1,2 Diffusion tensor tractography, which can provide comprehensive non-invasive mapping of eloquent white matter, plays an important role in preoperative brain tumor planning and intraoperative neuronavigation. By displaying the relationship of these tracts to brain tumors, tractography has been shown to improve tumor resection and patient survival while minimizing postoperative deficit.^3,4 DTI may also play a role in distinguishing between brain tumor types based on genetic information to determine the most suitable and effective treatment plan.^3-6 In addition to the anatomical location of the tumor, the localization of brain tissue infiltrated by tumor and overall extension of tumor can be detected and delineated by DTI metrics such as fractional anisotropy (FA) (related to the degree of anisotropy) and mean diffusivity (MD) (related to the total diffusion). This information is often invaluable for selection of the best treatment plan for each patient.^2-9

The current standard single-shot DTI (s-DTI) using single-shot echo-planar imaging has restrictions in terms of scanning time, spatial resolution, and coverage. As part of the standard of care at our institution, the reconstruction of the arcuate fasciculus language pathway and the corticospinal tract (CST) motor pathway are routine for each patient who is referred for presurgical mapping. Other tracts are only performed in a minority of selected patients. We therefore chose to only focus on the language and motor tracts in this paper. Brain tumors and other intracranial mass lesions can displace and distort the CST and arcuate fasciculus. Tumor mass effect can also change the local density of white matter fibers and diffusivity of the perilesional regions due to vasogenic or infiltrative edema, creating additional challenges in the reconstruction of fibers. Improving the quality of DTI and tractography in the clinical setting requires a tradeoff among acquisition parameters, signal-to-noise ratio (SNR), and scan time. Longer scan time, which is difficult for some patients to tolerate, improves the quality of DTI images and SNR, but may increase motion artifacts and reduce the accuracy of overall results. Reducing scan time while improving the quality of imaging is an essential goal in clinical applications.^7-12

Parallel-accelerated simultaneous multi-slice (SMS) imaging facilitates the acquisition of MRI data along the slice dimension to achieve shorter scan time, greater brain coverage, and/or higher spatial and temporal resolution.^12-15 SMS may also reduce misregistration due to subject motion or residual eddy current. Multi-band DTI (mb-DTI) utilizes simultaneous multi-slice excitation. Our mb-DTI acquisition was performed using blipped-controlled aliasing in parallel imaging (blipped-CAIPI EPI) based on SENSE reconstruction with a multiband factor of two. Several slices are simultaneously excited using a single radio-frequency pulse tailored for selective excitation at multiple frequencies. These slices can then be separated based on the coil sensitivity profiles and phase shift method. Thus, a multi-band factor of two acquires two slices simultaneously, halving the repetition time necessary to acquire the same number of slices.^14-17 Nevertheless, while SMS may lead to dramatic reductions in scan time, it may induce new artifacts, including partial slice leakage effects and noise enhancement due to signal loss and lower SNR. The trade-off between shorter acquisition times and increased artifacts is an important consideration when implementing SMS in a clinical setting for presurgical planning.¹⁷ Assessment of the clinical feasibility of mb-DTI remains limited to a few studies involving healthy controls and a small number of clinical cases.^13,14 Further research involving mb-DTI in patient populations remains critical to assess its potential advantages.

To evaluate the clinical advantage of mb-DTI in presurgical planning, we performed a qualitative and quantitative comparison focused on presurgical mapping based on tractography of the arcuate fasciculus and CST. We hypothesized that mb-DTI would provide data equivalent or superior to that obtained with s-DTI for preoperative mapping of motor and language pathways in patients with brain tumors.

METHODS

Patients

This retrospective study was approved by the institutional review board and was conducted in compliance with the Health Insurance Portability and Accountability Act under a waiver of informed consent. The study included 44 patients with brain lesions (mean age=56±14 years, male/female=18/26) who underwent presurgical planning MRI, s-DTI, and mb-DTI (Table 1). Gliomas were classified per the World Health Organization 2016 classification.¹⁸

Table 1:

Patients and tumor types included in the study.

		Distance from
Type		Left AF	CST	Location (lobes)
Gliomas
Anaplastic Oligodendroglioma	2	far	involved	Right temporoparietal
Glioblastoma Multiforme (GBM)	2	involved	far	Left temporal, frontal
GBM	2	involved	involved	Left temporoparietal
GBM	2	far	involved	Right temporoparietal
GBM	2	far	involved	Right frontoparietal
Oligodendroglioma		involved	involved	Left frontoparietal, temporal
Brain Metastases
Brest Carcinoma		involved	far	Left temporal
Colon Adenocarcinoma	2	far	involved	Right parietal
Colorectal and GE junction Adenocarcinoma	2	far	involved	Right frontal
Lung Adenocarcinoma	4	far	involved	Right frontal
Melanoma		far	involved	Right parietal
Nasopharyngeal Carcinoma		involved	involved	Left frontal
Primary CNS lymphoma		far	involved	Right frontal, left parietal
Prostate and Renal Adenocarcinoma	2	involved	involved	Left posterior frontal, right parietal
Salivary gland Adenocarcinoma		involved	involved	Right parietal, left temporal
Uterine Carcinosarcoma		far	involved	Left parietal
Far
Anaplastic Astrocytoma				Right superior frontal
Brest Carcinosarcoma				Right anterior septum pellucidum
CNS lymphoma				Right cerebellum
Endometrial Carcinosarcoma				Right occipital
GE junction Adenocarcinoma				Right sub-occipital
Melanoma				Left occipital
Melanoma				Right superior frontal
Meningioma				Right cavernous sinus
Meningioma				Sellar-suprasellar
Pituitary Adenoma				Right aspect of the sellar
Thyroid Carcinosarcoma				Bilateral occipital
Shunt
Breast cancer; epidermoid cyst		far	far	Right-sided ventriculoperitoneal shunt, posterior fossa
OPGs		far	far	Left -sided ventriculoperitoneal shunt, right cerebellum
Non-brain lesion
Erdheim-chester disease
Lung Adenocarcinoma
Lymphoma

AF: Arcuate fasciculus; CST: corticospinal tract; CNS: Central nerves system; number indicates the quantities.

Image Acquisition

MR images were acquired using a 3T scanner (750W, GE Healthcare, Waukesha, WI) and a 24-channel head/neck coil array. The s-DTI (repetition-time (TR)/echo-time (TE)=11000/74.8_(ms); slice thickness=3_(mm); number of slices=48; field of view (FOV)=240×240_(mmmm²₎; matrix size=128×128; b=1000_(s/mm²; directions=25) and mb-DTI (TR/TE=6500/75.7_(ms); number of multi-bands=2; slice thickness=3_(mm); number of slices=60; FOV=240×240_(mmmm²₎; matrix size = 128×128; b=800_(s/mmmm²₎; directions=25) were obtained in the same session. The acquisition times were 5 min 25 sec for s-DTI and 3 min 2 sec for mb-DTI. The MRI scans also included post-contrast 3D T1-weighted MRI consisting of an inversion recovery spoiled gradient-echo (BRAVO) sequence (TR/TE=53392/2.267_(ms); slice thickness=1_(mm); number of slices=176; matrix size=256×256) and fluid-attenuated inversion recovery (FLAIR) images (TR/TE=10000/106_(ms); inversion time=220_(ms); slice thickness=3_(mm); number of slices=64; matrix size=256×256).

DTI Analysis

DSI Studio was used for DTI image preprocessing (induced by local field inhomogeneity, motion, and eddy current distortion correction), tensor calculation, and deterministic fiber tracking algorithm.¹⁹

DTI Parameters: FA and MD

FSL-BET²⁰ and multi-image analysis GUI²¹ were used to obtain regions of interest (ROI). For each patient, the mean plus standard deviation for FA and MD were measured for each map across the whole brain and tumor ROIs. To determine whole brain measurements, a mask of whole brain was placed on the BRAVO image of the patient and then transferred to each DTI space. To generate the tumor ROI, first an intensity base threshold method was applied, and second further verification was performed to reduce the chance of underestimating and/or overestimating tumor regions. For all glioblastomas and brain metastases ROI was places on the enhancing component. For enhancing lesions, the BRAVO image was overlaid on the anatomical image and a threshold applied to cover the entire contrast region. For non-enhancing lesions, the FLAIR image was overlaid and given a threshold on the BRAVO image to cover the whole FLAIR abnormality. For enhancing metastases, only the BRAVO image was used. In the case of multiple brain metastases, the most clinically meaningful metastasis was evaluated. Finally, tumor ROI as a mask of 6±2 mm² was placed at the center of the selected lesion or tumor region of each patient and was transferred to the DTI space for further measurement (Fig.1).

Figure 1:

Region of interest (ROI) for DTI parameters. (1) T2-weighted and (2) post-contrast T1-weighted images of a patient with right temporal anaplastic astrocytoma tumor. (3–5) Surgical cavity (purple), enhancing (pink), and non-enhancing (yellow) tumor regions and tumor ROI (red circle). Tumor ROI is placed on the enhancing parts of the tumor region. (6) Whole brain ROI (green). Tumor and whole brain ROIs are overlaid on (7, 8) fractional anisotropy and (9, 10) mean diffusivity maps generated by s-DTI and mb-DTI, respectively. All ROIs and maps are superimposed on corresponding post-contrast T1-weighted images.

Diffusion Tensor Tractography

Tractography was performed on each patient for the whole brain, bilateral CST, and left arcuate fasciculus. The termination parameters were: FA threshold <0.2, angular threshold (maximum turning angle) >50°, step size² >0.49_(mm), and min/max tract length <30_(mm) mm or >700_(mm). Each subsequent direction was set to have 0.5 weight contributed from the prior direction. A total of 100000 seed points within the seed region were placed to begin tracking. Tracts began in the seed region and passed through the ROI. ^2,19 These criteria were applied to both s-DTI and mb-DTI. For whole brain tracts, a seed region was placed in the whole brain and passed through the whole brain ROI. For the CST, the seed region was placed at the level of the precentral gyrus in both hemispheres with an average volume size of 4000_(mm³₎. An ROI was placed at the pons with an average volume size of 5000_(mm³₎. For the arcuate fasciculus, the seed region was placed at the superior posterior temporal gyrus (Wernicke’s area) with an average volume size of 1200_(mm³₎ and the ROI at the level of the inferior frontal gyrus between the face area of motor cortex and sylvian fissure (Broca’s area) with an average volume size of 400_(mm³₎. After determining all tracts for each patient, the average values for fiber FA, volume_(mm³₎, length_(mm), and MD_(mm²_s⁻¹₎ in the whole brain tracts, CST, and arcuate fasciculus were measured (Fig.2).

Figure 2:

Region of interest (ROI) for Tractography. ROIs (green): 1) Pons, 2) Broca’s Area, and 3) whole brain. Seed regions (pink): 1) Precentral gyrus, 2) Wernicke's Area. (4–5) Whole brain seeds generated by s-DTI and mb-DTI, respectively. Tracts begin from seeds and pass through ROIs.

Visual Comparison

For qualitative analysis, blinded and randomized visual comparisons of s-DTI and mb-DTI were performed by three board-certified neuroradiologists with 20, 15, and 5 years of experience, respectively. The evaluation was conducted in two categories, anatomically correct (the correct topography of the anatomical location of detected tracts) and fiber extraction (the ability to extract the relevant number of fibers), and was based on three options: mb-DTI equal to s-DTI, mb-DTI better than s-DTI, or mb-DTI worse than s-DTI. In each category, for each selection, the percentage sum of the voting results of each neuroradiologist was calculated, and this number was averaged by the number of neuroradiologists.

Statistical Analysis

Group-level comparisons for DTI scalar parameters were performed by applying two-tailed paired t-tests and by estimating the linear correlation coefficient (r) between s-DTI and mb-DTI for FA and MD maps in the whole brain and tumor ROIs. Patients were divided based on their tumor involvement with the CST and/or arcuate fasciculus pathways. To compare the group-level s-DTI and mb-DTI tractography, average values of fiber FA, MD, volume, and length for the whole brain, CST, and arcuate fasciculus tracts were linearly correlated and compared by applying two-tailed paired t-tests.

To extend the s-DTI and mb-DTI tractography results to other brain pathways, in three patients with tumor involving the left uncinate fasciculus, left optic radiation, and the left portion of the forceps major we reconstructed the left and right uncinate fasciculus, left and right optic radiations, and forceps major and minor. To reconstruct the uncinate fasciculus, seed and ROIs were placed at the anterior portion of the superior temporal gyrus lateral to the amygdala and hippocampus, external capsule, and inferior frontal gyrus at the level of pars orbitalis respectively. Optic radiation was reconstructed from a seed in the lateral geniculate nucleus of the thalamus to an ROI at the level of the calcarine fissure of the primary visual cortex in the occipital lobe. Forceps major fibers were extracted from a seed in the midline of the splenium of the corpus callosum to ROIs at left and right occipital lobes. Lastly, to reconstruct the forceps minor tracts, a seed was placed at the midline of the genu of the corpus callosum and the ROIs at the lateral and medial surfaces of the frontal lobes. The size of the seed and ROI were determined based on the size of the anatomical regions. ROIs were selected to be large enough to cover the whole anatomical area. The average values of tractography parameters measured across patients and all tracts of each patient generated by mb-DTI and s-DTI were compared by applying two-tailed paired t-tests.

Using the Bonferroni correction, significance was set to p<α (α=0.05/(number of tests performed)) when familywise error rate=0.05. For visual comparison, the overall results were reported as mean plus standard deviation, which represents the inter-rater variability for each selection.

RESULTS

Patients

Of the 44 patients, three were diagnosed without brain lesions and were analyzed separately. The locations of lesions were closed/involved with the left arcuate fasciculus in 11 patients, with the left and/or right CST in 25 patients, and with both tracts in eight patients. The location of lesions was far from both tracts in 11 patients.

DTI Acquisition

Both s-DTI and mb-DTI generated adequate results despite small magnetic field inhomogeneities, with a 40% reduction in scan time when using mb-DTI. However, mb-DTI demonstrated decreased sensitivity to disruptions in the magnetic field caused by large susceptibility artifacts in two patients: one who had a posterior fossa epidermal cyst and a right frontal burr hole, and the other who had a cerebellar pilocytic astrocytoma and a left frontal approach ventriculoperitoneal shunt. In these two cases, the shunts in the frontal area, not the tumor, created large artifactual distortions (Fig. 3a). These two patients were analyzed separately, as discussed below.

Figure 3:

Susceptibility artifacts. The presence of a left ventriculoperitoneal shunt (red arrow) rather than the tumor caused different degrees of susceptibility artifacts (yellow arrow) and subsequent complications during the tractography of arcuate fasciculus tracts in s-DTI and mb-DTI. a) Raw DTI (b0 baseline image), b) arcuate fasciculus tracts overlaid on fractional anisotropy (FA) map, and c) color-coded FA map obtained from s-DTI (first row) and mb-DTI (second row). The arcuate fasciculus tracts and FA maps (a, b) are superimposed on the corresponding anatomical image. In this case, mb-DTI demonstrates less artifact and better result. Arcuate fasciculus tract and color-coded FA map are shown in directional colormap (X axis (red): right-left; Y axis (green): anterior-posterior, and Z axis (blue): inferior-superior). L=left, P=posterior, and S=superior.

DTI Parameters

Whole Brain

At the individual level and across the whole brain ROI, the FA was higher in mb-DTI than in s-DTI in 100% of cases, and the MD was lower in 67% of cases. At the group level, in the whole brain ROI, mb-DTI showed a significantly (α=0.01) higher mean FA (s-DTI, 0.39±0.17; mb-DTI, 0.38±0.18; 1.4% increase; p<0.001; r=0.97) and lower mean MD (s-DTI, 1.1±0.77; mb-DTI, 1.09±0.72₍0.001_mm²_s⁻¹₎; 0.6% decrease; p<0.001; r=0.99) (Fig. 4a).

Figure 4:

Diffusion parameter a) in whole brain and b) in tumor ROIs. The boxplot compares average FA and MD of s-DTI and mb-DTI across patients. The midline represents the median. The correlation (r), mean, and p values (α=0.01) were indicated for each compression.

Lesion

At the individual level and across the lesion ROI, the FA was higher in mb-DTI than in s-DTI in 100% of cases, and the MD was lower in 73% of cases. At the group level in the tumor ROI, mb-DTI showed a higher mean FA (s-DTI, 0.21±0.13; mb-DTI, 0.22±0.13; 7% increase; p<0.001; r=0.98) and there was a significant decrease in MD between s-DTI and md-DTI (s-DTI, 1.348±0.38; mb-DTI, 1.26±0.40₍0.001_mm²_s⁻¹₎; p=0.002; r=0.95; α=0.01) (Fig. 4b).

Tractography

The tractography results of the arcuate fasciculus, CST, and whole brain tracts in three patients are presented in Fig. 5. The group-level fiber tractography results of the whole brain, arcuate fasciculus, and CST are described below.

Figure 5:

DTI fiber tractography results for the arcuate fasciculus, corticospinal tract (CST), and whole brain tracts generated by s-DTI (first row) and mb-DTI (second row) methods in three patients. Columns a–c: Arcuate fasciculus for first patient (left temporoparietal glioblastoma) in sagittal view; CST for second patient (right frontal melanoma) in coronal view; and whole brain tract for third patient (left frontal carcinoma in axial view). Red arrows show the location of lesions. All tracts are shown in directional colormap (X axis (red): right-left; Y axis (green): anterior-posterior, and Z axis (blue): inferior-superior). L=left, P=posterior, and S=superior.

Whole Brain Tracts

In all patients with lesion in the brain, while the whole brain tracts generated by mb-DTI were significantly different in fiber FA, MD, and volume, they were well-fitted and highly correlated with the whole brain tracts generated by s-DTI (fiber FA: mb-DTI/s-DTI=0.48±0.02/0.47±0.02, r=0.94, p<0.001; fiber MD₍0.001_mm²_s⁻¹₎: mb-DTI/s-DTI=0.83±0.03/0.83±0.03, r=0.97, p<0.001; fiber volume ₍10000_mm³₎: mb-DTI/s-DTI=48.6±8.11/46.2±8.15, r=0.99, p<0.001). No significant difference was found in fiber length between these two methods (mb-DTI/s-DTI=79±4.6/78±5_(mm), r=0.97, p=0.1; α=0.01). The same results were seen in the three patients without any lesion in the brain (Fig. 6a).

Figure 6a-c:

Quantitative measurements for diffusion tractography. Boxplots show the average fiber length _(mm), volume _(mm³₎, FA, and MD _(mm²_s⁻¹₎ values for each tract across all patients with lesions involved and not involved with the related tract measured from both mb-DTI and s-DTI. The fiber FA obtained from mb-DTI seem significantly (p<0.005) better than those obtained from s-DTI in all tracts. Fiber volume in patients with lesions involving the CST and arcuate fasciculus as well as whole brain tracts of all brain tumor patients are significantly (p<0.001) improved with mb-DTI. The fiber MD in patients with lesions involved with the CST and whole brain tracts (all brain tumor patients) were significantly different between the two methods. The correlation (r), mean, and p values (α=0.01) of tractography outcomes were indicated. The midline represents the median.

Arcuate Fasciculus

In all patients with or without lesion involvement with the arcuate fasciculus, the tracts depicted by mb-DTI showed significant (p<0.001) improvement in fiber FA, volume, and length values. The arcuate fasciculus tracts generated by both mb-DTI and s-DTI were well-fitted and highly correlated: (fiber length_(mm): mb-DTI/s-DTI=70±14/61±13, r=0.89, p<0.001; fiber volume ₍1000_mm³₎: mb-DTI/s-DTI=2.74±1.3/2.06±1, r=0.72, p<0.001; fiber FA: mb-DTI/s-DTI =0.56±0.04/0.53±0.04, r =0.91, p <0.001; α =0.01). No significant difference was found in fiber MD between these two methods in patients with lesion involved or not involved with the arcuate fasciculus (mb-DTI/s-DTI=0.75±0.03/0.76±0.03 ₍0.001_mm²_s⁻¹₎, r=0.85, p=0.03; α=0.025) (Fig. 6b).

CST

Despite the significant increase in fiber FA for the CST tracts estimated by mb-DTI in all brain tumor patients, the CST tracts generated by both mb-DTI and s-DTI were highly correlated (mb-DTI/s-DTI=0.60±0.03/58±0.03, r=0.95, p<0.001). In all patients with lesions involving the CST, the tracts depicted by mb-DTI showed significant (p<0.001) improvement in fiber MD and volume (MD: mb-DTI/s-DTI=0.74±0.03/0.75±0.03₍0.001_mm²_s^−1-₎, r=0.96, p<0.001; volume: mb-DTI/s-DTI =2±0.6/1.6±1₍10000_mm³₎, r=0.9, p<0.001). No significant difference was found between the fiber length of CST in all patients and the fiber MD and volume of CST in patients without lesion involvement with the CST generated by these two methods (Fig. 6c).

Primary and metastatic brain tumors

In whole brain tracts, the tracts generated by mb-DTI showed significant increase only in the fiber MD of all patients with brain metastases (mb-DTI/s-DTI=0.83±0.02/0.83±0.03₍0.001_mm²_s⁻¹₎, r=0.97, p=0.0008) vs. gliomas (mb-DTI/s-DTI=0.84±0.03/0.84±0.04₍0.001_mm²_s⁻¹₎, r=0.96, p=0.2, α=0.025) (Table 1). No differences were found in terms of other fiber parameters. There were no differences between brain metastases and gliomas involving the arcuate fasciculus and/or left/right CST fibers generated by either methods.

Other white matter tracts

In three patients for whom we examined the comparison between s-DTI and mb-DTI in six other important pathways, FA of all tracts extracted by mb-DTI had significantly higher FA value (p<0.01). These tracts depicted by mb-DTI showed improvement in fiber FA (mb-DTI/s-DTI=0.49±0.01/0.47±0.01), fiber MD (mb-DTI/s-DTI=0.86±0.02±/0.86±0.04), fiber length _(mm) (mb-DTI/s-DTI=99±4/95±3), and fiber volume ₍1000_mm³₎ (mb-DTI/s-DTI=54±44/47±35). In Fig. 8 and Table 3, the results of fiber tractography of the left arcuate fasciculus, bilateral corticospinal tracts, whole brain, left and right uncinate fasciculus, left and right optic radiations, and forceps major and minor were compared between two methods in a patient with superior temporal glioblastoma. The fractional anisotropy (FA) was significantly higher in the tracts generated by mb-DTI (p=0.005, α=0.05). In this patient, the left uncinate fasciculus was involved by the tumor, and the tract generated by mb-DTI had slightly higher fiber FA, volume, and length and lower MD values.

Figure 8:

Comparison of nine white matter tracts reconstructed by s-DTI (columns 1–3) and mb-DTI (columns 4–6) in a patient with a left superior temporal glioblastoma multiform tumor. The left uncinate fasciculus was involved by the tumor. The tracts are shown in color-coded direction (columns 1 and 4), fractional anisotropy (FA) (columns 2 and 5), and mean diffusivity (MD) (columns 3 and 6). The following anatomical structures are depicted in rows: 1. arcuate fasciculus, 2. corticospinal tract, 3. whole brain, 4. uncinate fasciculus, 5. optic radiation, 6. forceps minor and major. Arrows show the anatomical terms of location.

Table 3:

Tractography results of nine reconstructed white matter tracts for one subject.

	FA		MD (0.001mm2s−1)		Fiber length (mm)		Fiber volume (mm 3)
	s-DTI	mb-DTI	s-DTI	mb-DTI	s-DTI	mb-DTI	s-DTI	mb-DTI
Arcuate fasciculus	0.49	0.50	0.723	0.718	63	76	8243	15259
Corticospinal tract	0.53	0.55	0.755	0.767	117	133	35331	41640
Whole brain	0.51	0.52	0.817	0.817	101	104	291858	318379
Left uncinate fasciculus	0.40	0.41	0.808	0.804	59	81	2509	4810
Right uncinate fasciculus	0.42	0.43	0.835	0.804	104	74	2939	6670
Left optic radiation	0.48	0.52	0.879	0.866	96	97	8970	13242
Right optic Radiation	0.48	0.50	0.881	0.907	93	93	10372	10000
Forceps minor	0.50	0.54	0.894	0.808	75	89	18552	18675
Forceps major	0.59	0.59	0.808	1.03	130	129	16675	17033
P Value	0.005		0.65		0.43		0.07

Visual Comparison

The results of the blinded visual comparison of s-DTI and mb-DTI are summarized in Table 2. For the arcuate fasciculus, the two methods were determined to be equivalent in 51% of anatomical correctness analyses and 50% of fiber extraction analyses. Mb-DTI was preferred 17% more frequently than s-DTI in arcuate fasciculus anatomical correctness analyses and 32% more frequently in arcuate fasciculus fiber extraction analyses. For the CST, the two methods were determined to be equivalent in 53% of anatomical correctness analyses and 39% of fiber extraction analyses. Mb-DTI was preferred 13% more frequently than s-DTI in CST anatomical correctness analyses and 33% more frequently in CST fiber extraction analyses.

Table 2:

Visual comparison of mb-DTI and s-DTI;

		Equivalent	mb-DTI Preferred	s-DTI Preferred
Arcuate fasciculus	Anatomically Correct	51±11%	33±3%	16±14%
	Fiber Extraction	50±16%	41±10%	9±5%
Corticospinal Tract	Anatomically Correct	53±22%	30±11%	17±12%
	Fiber Extraction	39±18%	47±15%	14±4%

Data represents mean±standard deviation of the percentage sum of the voting results from neuroradiologist responses

Susceptibility-Related Artifact

We further analyzed two patients with prominent signal loss due to susceptibility-related artifact. To measure the regional change in FA value for these patients, the ROI was placed very close to the burr hole instead of the tumor region because artifacts were dominated by the burr hole rather than the lesion. The mean FA values generated by s-DTI and mb-DTI were compared among regions of the brain affected by the tumor and susceptibility artifacts (Figures 3 and 7). The mean percentage FA change between the two methods was +7% across our data but was +30% for the two patients with the greatest signal dropout; the mean percentage FA change dropped to 3% (Fig. 7).

Figure 7:

Susceptibility artifacts based on mb-DTI and s-DTI. FA comparison between regions of the brain affected by lesions and susceptibility artifacts based on mb-DTI and s-DTI. a) ROI in the artifact regions for two patients (1: Epidermoid cyst in the posterior fossa with a right frontal shunt and 2: Pilocytic astrocytoma and left frontal shunt) with prominent signal loss and susceptibility-related artifact due to presence of shunts (red boxes). ROI is placed very close to the shunts rather than the tumor. Tumor ROI for the rest of patients (black boxes). b) The present FA change between the two methods is +49% for both patients with prominent signal loss and susceptibility-related artifact (red boxes) and less than +3% for the rest of the patients (black boxes).

The results of tractography for the two patients are presented in Fig. 3b-c. The mb-DTI tractography demonstrated superior generation of the whole brain, arcuate fasciculus, and CST in both patients.

DISCUSSION

We demonstrated that mb-DTI accelerated by a factor of two with 40% decrease in scan time significantly improved DTI and tractography parameters in terms of FA (p<10⁻³) in the whole brain and tumor regions as well as in the CST, whole brain, and arcuate fasciculus tracts. We showed that susceptibility-related image degeneration artifacts caused by regional abnormality improved up to 30% with mb-DTI. Based on our results, mb-DTI could generate fibers that are well-correlated with standard s-DTI with a shorter repetition time and smaller b-value than s-DTI. Visual comparisons confirmed greater fiber extraction with mb-DTI (although the anatomical course of the fibers was only marginally better for mb-DTI than for s-DTI). Our blinded visual assessment by neuroradiologists further supported our quantitative results, with stronger preference for mb-DTI than for s-DTI, particularly for fiber extraction of language and motor tracts. Equivalent results were found in approximately 50% of patients.

The effect of varying the acquisition parameters of DTI (such as b-value or number of gradient directions on the scalar values of DTI) has been described.^7,8,22-24 While both FA and MD values are sensitive to SNR, they respond differently to changes in voxel resolution, number of gradient directions, and b-value; e.g., the b-value may affect MD while the number of gradient directions may affect FA.^7,8 In fact, increased gradient direction increases the accuracy of the FA value due to improved SNR and decreased partial volume effects while the b-value may affect the MD.^10,24 This may prove important to clinical practice, as FA values have been found to be lower in many pathologies, including brain tumors.^25,26 We demonstrated increased FA in the brain and tumor regions for mb-DTI, which suggests that mb-DTI may be preferable to s-DTI for the quantitative evaluation of patients with brain lesions. Additionally, FA is the major determinant of the success of most tractography algorithms, with many terminating tracking at a preset lower FA threshold. The increased FA in our study probably led to the increased fiber tract volume and length found in the mb-DTI tracts.^1,2 In fact, beyond the arcuate fasciculus and CST, in three patients we also compared s-DTI and mb-DTI in other important fibers in the brain: again, FA of all tracts extracted by mb-DTI had significantly higher FA value (p<0.01).

Like other MRI data, DTI suffers from physiological artifacts including voluntary motion, cardiac pulsation, and respiration- and system-related artifacts. Acceleration and under-sampling in mb-DTI may reduce motion- and system-related artifacts. In patients with brain lesions, the margins of the surgical cavity, craniotomy, tumor in the immediate vicinity of air, non-tissue to tissue interfaces, magnetic field inhomogeneity, and vibration motion may induce additional signal loss due to changes in the susceptibility field.^15,27 We found that mild artifacts resulted in overall small changes, with both methods equally losing signal, while mb-DTI outperformed s-DTI with more prominent artifacts. In two patients with burr holes, the size and shape of the signal loss due to susceptibility and vibration-related artifacts differed between DTI data acquired by mb-DTI and s-DTI, with relatively better recovery of signal loss in mb-DTI. ²⁷ Although the exact measurement of artifact-related signal loss is beyond the scope of our study, the differences between these two methods are illustrated in Fig. 7. The mean percentage FA change between the two methods was +7% across all patients and decreased to 3% when excluding the two patients noted above.

Potential limitations of this study include its retrospective design and relatively small cohort size. We did not specifically examine the data acquired using the same b-value or at higher multi-band factors. Instead, we analyzed two fixed acquisition parameters currently used in clinical practice with mb-DTI set to use the shorter b-value (800 vs. 1000 s/mm²) and a multi-band factor of two. The shorter repetition time may also be leveraged to increase the number of gradient directions to increase the SNR without increasing scan time as compared to s-DTI. These benefits of mb-DTI are offset by slightly decreased diffusion contrast, which is usually tolerable for clinical imaging.

We found that implementing mb-DTI in patients with brain lesions could lead to a shorter scan time and qualitatively and quantitatively improved results. Mb-DTI and s-DTI produced congruent results for fiber tractography of arcuate fasciculus and CST, two white matter tracts of critical importance during presurgical planning.