Research article Received: 18 September 2014,
Revised: 19 November 2014,
Accepted: 4 January 2015,
Published online in Wiley Online Library: 23 February 2015
(wileyonlinelibrary.com) DOI: 10.1002/nbm.3266
Tractography of the optic radiation: a repeatability and reproducibility study Michael Dayana*, Sylvia Kreutzera,b and Chris A. Clarka Our main objective was to evaluate the repeatability and reproducibility of optic radiation (OR) reconstruction from diffusion MRI (dMRI) data. 14 adults were scanned twice with the same 60-direction dMRI sequence. Peaks in the diffusion profile were estimated with the single tensor (ST), Q-ball (QSH) and persistent angular structure (PAS) methods. Segmentation of the OR was performed by two experimenters with probabilistic tractography based on a manually drawn region-of-interest (ROI) protocol typically employed for OR segmentation, with both standard and extended sets of ROIs. The repeatability and reproducibility were assessed by calculating the intra-class correlation coefficient (ICC) of intra- and inter-rater experiments, respectively. ICCs were calculated for commonly used dMRI metrics (FA, MD, AD, RD) and anatomical dimensions of the optic radiation (distance from Meyer’s loop to the temporal pole, ML-TP), as well as the Dice similarity coefficient (DSC) between the raters’ OR segmentation. Bland–Altman plots were also calculated to investigate bias and variability in the reproducibility measurements. The OR was successfully reconstructed in all subjects by both raters. The ICC was found to be in the good to excellent range for both repeatability and reproducibility of the dMRI metrics, DSC and ML-TP distance. The Bland– Altman plots did not show any apparent systematic bias for any quantities. Overall, higher ICC values were found for the multi-fiber methods, QSH and PAS, and for the standard set of ROIs. Considering the good to excellent repeatability and reproducibility of all the quantities investigated, these findings support the use of multi-fiber OR reconstruction with a limited number of manually drawn ROIs in clinical applications utilizing either OR microstructure characterization or OR dimensions, as is the case in neurosurgical planning for temporal lobectomy. Copyright © 2015 John Wiley & Sons, Ltd. Additional supporting information may be found in the online version of this article at the publisher’s web site. Keywords: optic radiation; Meyer’s loop; diffusion imaging; tractography; reproducibility; reliability; repeatability
INTRODUCTION
NMR Biomed. 2015; 28: 423–431
* Correspondence to: M. Dayan, Developmental Imaging and Biophysics Section (DIBS), Institute of Child Health, University College London, 30 Guilford Street, London, WC1N 1EH, UK. E-mail: [email protected] a M. Dayan, S. Kreutzer, C. A. Clark UCL Institute of Child Health, London, UK b S. Kreutzer Institute of Neuroscience and Medicine, INM-3, Research Center Jülich, Jülich, Germany Abbreviations used: AD, axial diffusivity; dMRI, diffusion MRI; DSC, Dice similarity coefficient; DT, diffusion tensor; FA, fractional anisotropy; ICC, intra-class correlation coefficient; LGN, lateral geniculate nucleus; MD, mean diffusivity; ML, Meyer’s loop; NEX, bumber of excitations; OR, optic radiation; PAS, persistent angular structure; PICo, probability index of connectivity; QSH, Q-ball spherical harmonics; RD, radial diffusivity; ROI, region-of-interest; SNR, signal-to-noise ratio; ST, single tensor; TP, temporal pole; WM, white matter; RATI _ TPJ _ EXPK, rater I, timepoint J, experiment K.
Copyright © 2015 John Wiley & Sons, Ltd.
423
One set of white matter (WM) tracts of special interest in neuroimaging is the optic radiation (OR), a bundle of fibers connecting the lateral geniculate nucleus (LGN, part of the thalamus) to the primary visual cortex. This WM bundle has been the subject of much focus using neuroimaging, as it is a main component of the visual system. Of crucial importance is the ability to characterize its topology for neurosurgery planning, e.g. for temporal lobectomy (1) to avoid visual field defects in patients with intractable epilepsy, as well as its state when affected by diseases, e.g. in multiple sclerosis (2). WM tracts can be reconstructed in vivo with diffusion magnetic resonance imaging (dMRI) combined with tractography. The motion of water molecules in the brain causes an attenuation of the MRI signal in each voxel of the imaged volume and is constrained by WM fiber microstructure and arrangement (3). From dMRI measurements in several directions, one can estimate the water molecular diffusion profile (4), the peaks of which are interpreted as corresponding to the WM fiber directions. The simplest diffusion model is the diffusion tensor (DT) (5), the diffusion profile of which can be represented by an ellipsoid. Tensor elements are used to provide rotationally invariant metrics characterizing the shape of this ellipsoid and a number of important diffusion metrics can be derived from it. They include fractional anisotropy (FA) and mean diffusivity (MD), as well as axial and radial diffusivities (AD and RD, respectively), the values of which can be interpreted as characterizing the underlying microstructure. In the case of the DT, the ellipsoid can provide only a single
peak and therefore a single fiber-bundle orientation. Model-free methods such as Q-ball spherical harmonics (QSH), first introduced by (6), and persistent angular structure (PAS), described by (7), calculate the diffusion profile directly from the measurements without assumptions on its shape and allow the estimation of multiple bundle directions per voxel. Connecting fiber directions from voxel to voxel to reconstruct entire white matter tracts is the basis of tractography (8,9), which initiates tracts from a ‘seed’ region of interest (ROI) and often relies on other selection and exclusion ROIs to reconstruct specific WM tracts.
M. DAYAN ET AL. Since the introduction of tractography, in vivo anatomical depiction of the OR has been possible. Most studies dedicated to OR tractography have relied on careful ROI positioning by experimenters (10–14,2,15–20). The main drawback of tractography is the difficulty of validating in vivo reconstructions of a particular subject. The gold standard for this purpose is comparison of the subject’s in vivo segmentation with histology, and thus is only possible in animals. As a result, human studies investigating tractography validation rely on known anatomy in the population obtained from dissection and histology studies (21,22). In the case of the OR, typical comparisons involve the distances separating its anterior tip, Meyer’s loop (ML), from well-defined landmarks (10,11,13,15,1,19). However, not only is the accuracy of tractography reconstructions of importance, but also its reliability (how much variation characterizes repeated measurements), which includes reproducibility (how much variation characterizes results from different experimenters) and repeatability (how much variation characterizes results from the same experimenter under the same conditions). These issues are particularly important for translation of OR tractography into the clinical setting. The aim of the present work was to assess the repeatability and reproducibility of an OR tractography method based on manually drawn ROIs by comparing results obtained from the same experimenter and from two different experimenters, respectively. The outcome of this study informs us regarding the usability of OR reconstruction methods similar to the one described, for both research and practical clinical purposes.
Data processing Raw DICOM slices and volumes from the scanner acquisition were sorted and converted into Analyze volumes with a custom script based on the Tractor software (23). dMRI volumes were corrected for eddy currents and small head movements by registration with the first non-diffusion-weighted volume with a 12-parameter affine transformation using FSL (24). The diffusion gradient directions were updated accordingly by applying to them the rigid-body transformation associated with the previous transformation. The brain was then segmented with the BET utility (25), in which parameters were adjusted after iterative visual inspection. The diffusion tensor (DT) coefficients were computed with weighted linear least-squares regression with the Camino software (26) and then used in the calculation of the dMRI metrics, namely FA, MD, AD and RD. The main diffusion directions were estimated with the single tensor (ST) model as well as with QSH and PAS, in order to not only compare single-fiber and multi-fiber methods but also evaluate the importance of the method choice for estimating fiber directions. The maximum order of the spherical harmonic series for QSH was set at 6, while the number of basis functions for the PAS representation was chosen to be 16 (27). Tractography
METHODS Subjects The study took place at Great Ormond Street Hospital, London, UK. This work was granted ethical approval by the local ethics committee. 14 healthy adults without any known medical condition took part in the study. Informed consent was obtained in all subjects before their participation. The cohort included 10 males and 4 females, with an age range from 21–65 years. Imaging
424
Each participant underwent a dMRI protocol on a Siemens Avanto 1.5 T clinical system (Siemens Healthcare, Erlangen, Germany), using a self-shielding gradient set with maximum gradient strength of 40 mT m 1 and standard ‘birdcage’ quadrature headcoil. Echoplanar diffusion-weighted images were acquired for an isotropic set of 60 non-collinear directions, using a weighting factor of 1000 s mm 2, along with three T2 weighted (b = 0) volumes and with a repetition factor (or number of excitations (NEX)) of 2 to increase the signal-to-noise ratio (SNR). 45 contiguous axial slices of thickness 2.5 mm were imaged, using a field of view of 240 × 240 mm 2 and 96 × 96 voxel acquisition matrix, for a final image resolution of 2.5 × 2.5 × 2.5 mm 3. The echo time was 89 ms and the repetition time was 6300 ms. In addition, a T1 -weighted 3D FLASH structural image was acquired using 176 contiguous sagittal slices, a 256 × 224 mm 2 field of view and 1 × 1 × 1 mm 3 image resolution. The echo time in this case was 4.9 ms and the repetition time was 11 ms. Once the data were acquired (timepoint 1), the same protocol was repeated during a second scan session undertaken approximately one week later
wileyonlinelibrary.com/journal/nbm
(timepoint 2). The dMRI sequence was also repeated within the first session and, as a result, the overall scan time for the acquisition was approximately 50 min for the first session and 30 min for the second session.
The tractography algorithm run was the PICo algorithm, as implemented in Camino. This algorithm relies on repeating the tractography process, and at each iteration changing the fiber direction(s) according to the associated uncertainty. The typical output of PICo is a probability map, in which each voxel contains the number of times a tract went through that voxel, divided by the total number of tracking iterations (28). In order to perform tractography for the OR, ROIs were defined on the color-coded FA maps, on which the main diffusion directions were overlaid. As described in a previous work (19), a 16-voxel seed region was placed on one coronal slice, slightly anterolaterally to the LGN, and a waypoint ROI including the striatum sagittale (of which the OR is part) was drawn on another coronal slice. Both ROIs were drawn in regions of high FA, where the main diffusion directions were aligned along the anteroposterior directions (i.e. with high green intensity on the FA color-coded map). Two sets of exclusion ROIs were defined solely according to the main diffusion directions, in order to exclude neighbouring tracts to ML, and both included whole plane exclusion masks in inferior, superior and mid-sagittal sections. A ‘standard’ set of exclusion ROIs (Fig. 1, rows 1 and 2), abbreviated thereafter as ‘std set’, featured ROIs placed laterally to prevent tracking of the acoustic radiation, medially to avoid the generated tracts following the anterior commissure and forceps major, and anteriorly to prevent tracking of bundles continuing along the inferior occipitofrontal fasciculus and uncinate fasciculus. An ‘extended’ set, abbreviated as ‘ext set’, was also defined, as some extra artifactual tracts could remain after the implementation of the ‘std set’ in the tractography process. The ‘ext set’ included an additional medial ROI to remove fibers connecting to the cingulum (Fig. 1, row 3) and lateral ROI to exclude fibers coursing to V5 (Fig. 1, row 4).
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY
Figure 1. Sagittal, coronal and axial slices of a single subject FA map colored according to the principal diffusion direction (green: anterior–posterior, blue: superior–inferior, red: left–right), illustrating the placement of the ROIs used in the tractography analysis. ROIs belonging to the ‘std’ set are shown in orange in rows 1 and 2 and those of the ‘ext’ set are shown in blue in rows 3 and 4. Overview of the ‘std’ set ROIs, which are all (1) whole-plane exclusion ROIs, except for (2) the medial ROI, which does not include the occipital lobe, and overview of the ‘ext set’, which includes an additional (3) medial and (4) lateral ROI. Please note that, in all panels, the position of the sagittal slice is indicated by the thin green cursor lines in the associated coronal and axial slices.
NMR Biomed. 2015; 28: 423–431
gray matter regions, whilst avoiding too high a number of spurious tracts associated with voxels exhibiting high uncertainty in the diffusion directions. The maximum intensity projection of the probabilistic maps generated for each seed voxel was displayed with a linear scale, from red to yellow, and with a
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
425
Once the ROIs were created, the PICo algorithm was run for each fiber reconstruction method (ST, QSH, PAS) without any angular threshold (similarly to (15) and (1)) to account for the high curvature of ML. A relatively low FA threshold, 0.1, was used to allow the tracts to reach the extremities of the OR ending in
M. DAYAN ET AL. threshold of 1 % to eliminate artifactual tracts visually without removing plausible OR bundle reconstructions. Calculation of dMRI metrics The parameters selected to quantify the microstructural characteristics of the segmented OR were the average FA, MD, AD and RD. The computation of their values was achieved using a custom MATLAB® script. It consisted of selecting only voxels with probabilities greater than the chosen PICo threshold of 1 %, which resulted in the creation of a binary mask. This threshold was chosen, as it was considered to correspond to the most visually convincing reconstruction of the OR. The next step was to select the area corresponding to the OR in each of the dMRI metrics volumes (e.g. the FA volume) and keep only voxels with PICo probability greater than 1 %. This step was carried out by a voxel-by-voxel multiplication with the binary mask previously computed. An average over the resulting thresholded area was then calculated and provided the dMRI metric mean corresponding to a 1 % PICo threshold. Neuroanatomical distances of the OR To assess the spatial location of ML and compare the results with reference dissection studies, the distance from its tip to the temporal pole (ML-TP) was measured. The tip of ML was identified by visual inspection of the PICo probability maps displayed with a threshold of 1 %. The main diffusion directions were superimposed on to these maps to confirm that the most anterior voxels forming ML, highlighted on the probability maps, were indeed part of the reconstructed OR. The position of the most anterior voxel belonging to the OR was chosen as the location of the tip of ML. The location of the temporal pole (TP) was evaluated on the color-coded FA maps and chosen as the coronal slice coinciding with the most anterior extent of the temporal lobe – including the TP – and the beginning of the frontal lobe (19). In cases where several coronal plane positions along the anterior–posterior axis fit this criterion, the most anterior position was systematically selected. The ML-TP distance was computed as the difference between the tip of ML and the TP.
at timepoint 1: INTER = ICC(RAT1 _ TP1 _ EXP1, RAT2 _ TP1 _ EXP1). The metrics calculated for reproducibility were as follows: • the Dice similarity coefficient (DSC), which informs us regarding tract shape similarity and is only calculated for reproducibility measures involving two different raters (INTRA_SAMESCAN and INTER), DSC ðEXP1 ; EXP2 Þ ¼
2ðEXP1 voxels∩ EXP2 voxelsÞ EXP1 voxels þ EXP2 voxels
• dMRI metrics (FA, MD, AD, RD), which are often relied upon in clinical research using tractography, and • the ML-TP distance, which plays an important role in neurosurgery planning. It is important to underscore that rater 1 had no prior experience of MRI imaging and was trained during a single day by rater 2, who had substantial experience in dMRI and OR anatomy. Tractography reproducibility was further assessed graphically by generating Bland–Altman plots. These plots represent the difference M1(Si) M2(Si) between two repeated measures M1 and M2 on a subject Si as a function of this measure mean, [M1(Si) + M2(Si)]/2, and feature three constants shown as horizontal lines: a solid line for the mean difference between the repeated measures on all N subjects, 1 N M1 ðSi Þ þ M2 ðSi Þ ∑ N i¼1 2 and dashed lines defined such as to include 95 % of the points, called the 95 % lines of agreement. Such plots can reveal any systematic bias (if the mean difference represented by the solid line is different from 0) and allow us to estimate whether the amount of variability is acceptable, i.e. if the interval of variation within the 95 % lines of agreement is not much larger than the effect to be measured.
RESULTS Tractography reconstruction
Reproducibility analysis
426
To assess the reproducibility of the tractography method, the data obtained at timepoint 1 and timepoint 2 (cf. Imaging subsection) were analyzed by two raters to evaluate three reproducibility measures estimated with the intra-class correlation coefficient (ICC). The ranges ICC < 0.40, 0.40 ≤ ICC < 0.60, 0.60 ≤ ICC < 0.75 and 0.75 ≤ ICC indicate poor, fair, good and excellent reliability, respectively (29). The intra-rater reproducibility (INTRA_SAMESCAN) for data acquired from the same scanning session at timepoint 1 (TP1) was evaluated by comparing two tractography experiments (EXP1 and EXP2) undertaken by rater 1 (RAT1) and separated by a time interval of at least one week: INTRA _ SAMESCAN = ICC(RAT1 _ TP1 _ EXP1, RAT1 _ TP1 _ EXP2). The intra-rater variability including scanning variability (INTRA_RESCAN) was estimated by comparing results obtained by the same rater (RAT1) from data acquired at timepoint 1 and timepoint 2: INTRA _ RESCAN = ICC(RAT1 _ TP1 _ EXP1, RAT1 _ TP2 _ EXP1). Finally, the inter-rater reproducibility was evaluated from the comparison between results obtained by two different raters (RAT1 and RAT2) on the same data acquired
wileyonlinelibrary.com/journal/nbm
An example of probability maps obtained for the ‘std’ and ‘ext’ sets from one subject (with fiber directions estimated with the PAS model and PICo threshold 1 %) is shown in Figure 2. The value of the tractography probability in each voxel is displayed on a linear scale from just above 0 (one track went through that voxel) to 1 (all the tracts generated passed through that voxel), while a grayscale FA map is shown in the background. For the ‘std set’, the probability map color scale was chosen from red to yellow and for the ‘ext set’ from dark blue to cyan. As the ‘std set’ excludes fewer fibers, the associated probability map features more voxels than the probability map of the ‘ext set’. The entire OR could be reconstructed with ML clearly visible in all subjects, irrespective of the method chosen for estimating the fiber directions. Tract shape similarity The DSC values of most repeatability and reproducibility measures (INTRA_SAMESCAN, INTRA_RESCAN, INTER) were in the [0.6, 1] range, corresponding to values ranging from good ([0.6, 0.75]) to excellent ([0.75, 1]) agreement (Fig. 3).
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY
Figure 2. Example of a probabilistic map of the OR with PICo threshold of 1 % obtained for the ‘std’ and ‘ext’ sets, represented in sagittal (left), coronal (middle) and axial (right) planes. The OR reconstruction is displayed using a linear color scale ranging from red to yellow for the ‘std’ set and from dark blue to cyan for the ‘ext’ set. The ‘std’ set resulted in more voxels included in the OR, due to the less stringent exclusion ROIs. The extra voxels tended to be part of tracts coursing towards the cingulum (circled in green) and V5 (circled in orange).
Figure 3. Dice similarity coefficient (DSC) for the INTRA_SAMESCAN (top), INTRA_RESCAN (middle) and INTER (bottom) reproducibility measures and the ST, QSH and PAS reconstruction methods. Results of the left and right hemispheres are shown in the left and right columns, respectively. Values of the DSC associated with the ‘std’ and ‘ext’ mask sets are shown in blue and pink. Brown, orange and green dashed lines indicate threshold values separating fair, good and excellent DSC: 0.4, 0.6 and 0.75, respectively.
NMR Biomed. 2015; 28: 423–431
categorized as ‘good’. For the INTRA_RESCAN measure, PAS and QSH methods provided ‘good’ DSC, while the ST model values were exactly at or just below the threshold for the ‘good’ category. The general trend for DSC values was to be larger for
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
427
For the INTRA_SAMESCAN and INTER measures, excellent DSC were obtained for the PAS and QSH methods, although for the latter the extended set of masks resulted more often in ‘good’ DSC, and DSC for the ST model corresponded to values
M. DAYAN ET AL. multifiber reconstructions (QSH and PAS) – with PAS resulting in the largest DSC for most cases – and larger for the standard set of masks compared with the extended one. Anatomical measurements For a given tractography experiment, the ML-TP distance was found to be the same for each fitting method (ST, QSH and PAS). The ICC for the ML-TP distance was found to be good (0.60 ≤ ICC < 0.75) for all repeatability and reproducibility measures (INTRA_SAMESCAN, INTRA_RESCAN, INTER) in both hemispheres, except in the left hemisphere of the INTRA_SAMESCAN, where it was excellent (0.75 ≤ ICC, Fig. 4).
Figure 4. ML-TP distance ICC for the INTRA_SAMESCAN, INTRA_RESCAN and INTER reproducibility measures. The ICC for left and right hemispheres are indicated in blue and green, respectively.
The ICC was shown to be systematically smaller in the right hemisphere. The Bland–Altman plots did not demonstrate any bias, while the 95 % lines of agreement were around ± 1 voxel (2.5 mm) in the left hemisphere and around ± 2 voxels (5 mm) in the right hemisphere. dMRI metrics For all repeatability and reproducibility measures, all reconstruction methods and both hemispheres, the ICC of all dMRI metrics were in the good ([0.6, 0.75]) to excellent ([0.75, 1]) range (Fig. 5), with the only exception being for the INTRA_SAMESCAN repeatability of FA within the right OR reconstructed with fiber directions estimated with the PAS method. For the INTRA_RESCAN reproducibility, dMRI metrics were good to excellent for ST, mostly excellent for QSH and all excellent for PAS. For QSH and PAS, the standard set of masks tends to give slightly more reproducible results, however the opposite was true for the ST model. For the INTER reproducibility, dMRI metrics were good to excellent for both ST and QSH and mostly excellent for PAS, with the exception already mentioned of FA in the right OR. For the INTER reproducibility, the dMRI metrics ICC were all excellent. A trend between standard and extended mask sets similar to that of the INTRA_SAMESCAN repeatability was observed in the INTRA_RESCAN and INTER measures. The Bland–Altman plot of dMRI metrics from the INTRA_SAMESCAN measure had no apparent systematic bias, all mean differences between raters being close to zero and the 95 % lines of agreement globally similar in absolute value (see Fig. S1 in the Supporting Information). A general trend was for the ST model to have larger variation between raters compared with the QSH and PAS methods: FA differences had 95 % lines of agreement larger than ±0.05 and axial, radial and mean
428
Figure 5. dMRI metrics ICC for the INTRA_SAMESCAN (top), INTRA_RESCAN (middle) and INTER (bottom) reproducibility measures and the PAS (left), QSH (middle) and ST (right) reconstruction methods. For each given pair of reproducibility measure and reconstruction method, the ICC for the left and right hemispheres are shown on the left and right, respectively. Values of the ICC associated with the ‘std’ and ‘ext’ mask sets are shown in blue and pink. Brown, orange and green dashed lines indicate threshold values separating fair, good and excellent ICC: 0.4, 0.6 and 0.75, respectively.
wileyonlinelibrary.com/journal/nbm
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY diffusivities larger than ± 0.5 × 10 3 mm 2 s 1 for the ST model, while these values were smaller for QSH and PAS, more markedly for the latter. The left and right hemispheres presented globally the same pattern. The Bland–Altman plot related to the INTRA_RESCAN and INTER_RESCAN reproducibility measures exhibited the same behaviour, with slightly more variation between raters for the INTRA_RESCAN reproducibility (see Figs. S2 and S3 in the Supporting Information).
DISCUSSION Tractography reconstruction shape and dimensions
NMR Biomed. 2015; 28: 423–431
dMRI metrics extracted from the OR The fact that dMRI metrics demonstrated excellent ICC (0.75 ≤ ICC) for QSH and PAS for the INTRA_SAMESCAN and INTER reliability measures, compared with good ICC (0.60 ≤ ICC < 0.75) for ST, concurred with the DSC findings. Another common characteristic with the shape reliability is that the extended set of masks provided more variability than the standard set, again probably due to the added complexity of positioning the extra exclusion masks. It can also be seen in Figure 5 that FA in general had lower ICC compared with other dMRI metrics. This could be due to the higher sensitivity of FA to change in microstructure, which for this very reason is also the most often used metric in clinical studies based on diffusion imaging. As with the DSC results, the INTRA_RESCAN reliability measure is generally lower than both the INTRA_SAMESCAN and INTER measures. This was counterintuitive, as more variation was expected when comparing different raters. Since the influence of subject re-scanning is assumed to be minimal, this difference was thought to arise from the multitude of landmarks to consider when positioning ROIs and the inexperience of rater 1, who was given a single day of training. Consequently, results were more similar between the trainer and trainee than within the own trainee experiments. It can be inferred that, although a single training day was enough to obtain reproducible results, it is highly likely that higher reproducibility could be obtained with a longer training period. More than a day of training therefore seems recommended. The Bland–Altman plots illustrated the highest reliability gained from using multi-fiber reconstruction methods. The ST model had the most spread 95 % lines of confidence for all reliability measures, although in absolute value these were around 0.03 for FA, ± 0.5 × 10 3 mm 2 s 1 for MD and RD, and ± 0.1 × 10 3 mm 2 s 1 for AD. These values are small, but still larger than the effects that can be found as significant in clinical studies. This is, however, not an issue, since the Bland–Altman plots demonstrated that no systematic bias existed in either reproducibility or repeatability and, as a result, the importance of these variations would decrease with the size of the sample studied, such that it should be possible to detect effects even an order of magnitude smaller than these values with a large cohort.
CONCLUSION We demonstrated in this work that, despite being challenging to segment, the OR could be reconstructed from ROI-based tractography methods, similar to the most common methods
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
429
Tractography reconstruction of the optic radiation is notoriously difficult, due to the high curvature of ML and the high variability of individual trajectories. Tractography methods of the OR based on ROI locations can differ subsequently, notably due to the choice of seed ROI (30). In this work, positioning the seed ROI not exactly in the LGN but lateral to it avoided GM voxels where FA is low and thus where diffusion directions are associated with high uncertainty. Importantly, this was of particular interest for estimating the subsequent segmentation repeatability and reproducibility, as many studies have employed a similar seed placement (11,31,15,32–35,17). Selecting a single coronal slice and defining an ROI spanning the entire cross-section of the OR enabled us to recover both its medial and lateral aspects. The visual inspection of the tracts and topological measurements from surrounding anatomical landmarks demonstrated that such a seed ROI was effective for reconstructing the general shape of the OR – including ML – when combined with a waypoint ROI in the sagittal stratum and specific exclusion masks, whatever the method used to estimate the fiber directions, be it ST, QSH or PAS. The standard set of exclusion masks was relatively simple to define, while the extended set required more attention given to the surrounding structures. This was reflected in the results, as discussed subsequently, for which the extended set of masks consistently resulted in higher variation in repeatability and reproducibility. The DSC results demonstrated that there was generally good repeatability and reproducibility, although the ST model was at the boundary between moderate and good agreement for the INTRA_RESCAN reliability measure. The ST only allows for one fiber direction reconstructed per voxel and therefore imposes more restriction during the tracking process. A change in the ROIs would thus have more effect with the ST model compared with multi-fiber methods and so provides a possible explanation for these findings. The good to excellent reliability of the OR DSC concurred with visual inspection, according to which all tracts appeared correctly segmented, with the same anterior tip position found for all fiber direction methods used (ML-TP was the same for ST, QSH and PAS). The ML-TP distance was categorized as excellent or close to excellent (0.75 ≤ ICC) for all reliability measures in the left hemisphere, while it was categorized as good (0.60 ≤ ICC < 0.75) in the right hemisphere. This result was probably caused by an asymmetry in the OR itself (19), likely to have a slightly different fiber configuration – perhaps more complex with a higher level of voxels containing multiple fibers in the right OR. A similar trend was demonstrated in the Bland– Altman plots, where higher variability was found between raters in the right OR (± 2 voxels, i.e. 5 mm) compared with the left (± 1 voxel, i.e. 2.5 mm), although no bias was apparent in either case. The amount of variability in extent is consistent with a slice
position uncertainty of ± 1 voxel for defining both the TP position and ML position. As mentioned above, visual inspection ascertained that all different methods provided a satisfactory reconstruction of the OR and that the most anterior tip of ML was at the same location for each of them. The outputs of all these methods were therefore globally comparable and thus all deemed acceptable for conventional OR tractography. We did not investigate multiple-tensor fitting, though it is available in Camino, as this method requires us to detect multi-fiber population voxels, is unstable when fitting more than two fiber directions (36) and is therefore outperformed by most existing multi-fiber algorithms.
M. DAYAN ET AL. reported in the OR literature, with good to excellent repeatability and reproducibility as assessed from its shape, dimensions and averaged dMRI metrics. Reliability was shown to be enhanced when the complexity of the set of ROIs drawn was reduced and when tractography was performed on data obtained with multi-fiber reconstruction algorithms rather than the ST model. These results are encouraging and further support the use of OR tractography methods based on manually drawn ROIs, for studies investigating microstructural changes with dMRI metrics as well as for neurosurgery planning, in which knowledge of the ML-TP distance is important. The fact that the ML-TP distance was not found to differ between the ST model and the multi-fiber methods suggests that OR tractography with the simpler ST model is adequate when only a limited number of diffusion directions is available for dMRI acquisition.
16. 17. 18.
19.
20.
Acknowledgements This work was supported by the UK Medical Research Council (G0300117) and Engineering and Physical Sciences Research Council (EP/C536851/1). CAC acknowledges European Union Grant FP7-ICT-2009-C 238292.
REFERENCES
22.
23.
430
1. Winston G, Mancini L, Stretton J, Ashmore J, Symms M, Duncan J, Yousry T. Diffusion tensor imaging tractography of the optic radiation for epilepsy surgical planning: A comparison of two methods. Epilepsy Res. 2011; 97: 124–132. 2. Reich D, Smith S, Gordon-Lipkin E, Ozturk A, Caffo B, Balcer L, Calabresi P. Damage to the optic radiation in multiple sclerosis is associated with retinal injury and visual disability. Arch. Neurol. 2009; 66: 998–1006. 3. Beaulieu C. The basis of anisotropic water diffusion in the nervous system – a technical review. NMR Biomed. 2002; 15: 435–455. 4. Yablonskiy D, Sukstanskii A. Theoretical models of the diffusion weighted MR signal. NMR Biomed. 2010; 23: 661–681. 5. Basser PJ, Mattiello J, Le Bihan D. Estimation of the effective selfdiffusion tensor from the NMR spin echo. J. Magn. Reson., Ser. B 1994; 103: 247–254. 6. Tuch D. Q-ball imaging. Magn. Reson. Med. 2004; 52: 1358–1372. 7. Jansons K, Alexander D. Persistent angular structure: New insights from diffusion magnetic resonance imaging data. Inverse Probl. 2003; 19: 1031–1046. 8. Mori S, Crain BJ, Chacko VP, VanZijl PCM. Three-dimensional tracking of axonal projections in the brain by magnetic resonance imaging. Ann. Neurol. 1999; 45: 265–269. 9. Lazar M. Mapping brain anatomical connectivity using white matter tractography. NMR Biomed. 2010; 23: 821–835. 10. Yamamoto T, Yamada K, Nishimura T, Kinoshita S. Tractography to depict three layers of visual field trajectories to the calcarine gyri. Am. J. Ophthalmol. 2005; 140: 781–785. 11. Nilsson D, Starck G, Ljungberg M, Ribbelin S, Jönsson L, Malmgren K, Rydenhag B. Intersubject variability in the anterior extent of the optic radiation assessed by tractography. Epilepsy Res. 2007; 77: 11–16. 12. Xie S, Gong GL, Xiao JX, Ye JT, Liu HH, Gan XL, Jiang ZT, Jiang XX. Underdevelopment of optic radiation in children with amblyopia: A tractography study. Am. J. Ophthalmol. 2007; 143: 642–646. 13. Sherbondy AJ, Dougherty RF, Napel S, Wandell BA. Identifying the human optic radiation using diffusion imaging and fiber tractography. J. Vis. 2008; 8: 1–11. 14. Chen X, Weigel D, Ganslandt O, Buchfelder M, Nimsky C. Prediction of visual field deficits by diffusion tensor imaging in temporal lobe epilepsy surgery. Neuroimage 2009; 45: 286–297. 15. Yogarajah M, Focke NK, Bonelli S, Cercignani M, Acheson J, Parker GJM, Alexander DC, McEvoy AW, Symms MR, Koepp MJ, Duncan JS. Defining Meyer’s loop-temporal lobe resections, visual field
wileyonlinelibrary.com/journal/nbm
21.
24.
25. 26.
27.
28.
29.
30. 31. 32.
33.
deficits and diffusion tensor tractography. Brain 2009; 132: 1656–1668. Hofer S, Karaus A, Frahm J. Reconstruction and dissection of the entire human visual pathway using diffusion tensor MRI. Front. Neuroanat. 2010; 4: 1–7. White M, Zhang Y. Three-tesla diffusion tensor imaging of Meyer’s loop by tractography, color-coded fractional anisotropy maps, and eigenvectors. Clin. Imaging 2010; 34: 413–417. Wu W, Rigolo L, O’Donnell L, Norton I, Shriver S, Golby A. Visual pathway study using in vivo diffusion tensor imaging tractography to complement classic anatomy. Neurosurgery 2012; 70: 145–156 (Discussion 156). Dayan M, Muñoz M, Jentschke S, Chadwick M, Cooper J, Riney K, Vargha-Khadem F, Clark C. Optic radiation structure and anatomy in the normally developing brain determined using diffusion MRI and tractography. Brain Struct. Funct. 2015; 220: 291–306. Kuhnt D, Bauer M, Sommer J, Merhof D, Nimsky C. Optic radiation fiber tractography in glioma patients based on high angular resolution diffusion imaging with compressed sensing compared with diffusion tensor imaging – initial experience. PLoS One 2013; 8(7). Seehaus A, Roebroeck A, Chiry O, Kim DS, Ronen I, Bratzke H, Goebel R, Galuske R. Histological validation of DW–MRI tractography in human postmortem tissue. Cereb. Cortex 2013; 23: 442–450. Assaf Y, Alexander DC, Jones DK, Bizzi A, Behrens TEJ, Clark CA, Cohen Y, Dyrby TB, Huppi PS, Knoesche TR, LeBihan D, Parker GJM, Poupon C. The CONNECT project: Combining macro- and microstructure. Neuroimage 2013; 80: 273–282. ISSN 1053-8119; http:// dx.doi.org/10.1016/j.neuroimage.2013.05.055 Clayden JD, Jentschke S, Muñoz M, Cooper JM, Chadwick MJ, Banks T, Clark CA, Vargha-Khadem F. Normative development of white matter tracts: similarities and differences in relation to age, gender, and intelligence. Cereb. Cortex 2012; 22: 1738–1747. Smith S, Jenkinson M, Woolrich M, Beckmann C, Behrens T, Johansen-Berg H, Bannister P, DeLuca M, Drobnjak I, Flitney D, Niazy R, Saunders J, Vickers J, Zhang Y, DeStefano N, Brady J, Matthews P. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 2004; 23: S208–S219. Smith SM. Fast robust automated brain extraction. Hum. Brain Mapp. 2002; 17: 143–155. Cook PA, Bai Y, Gilani NS, Seunarine KK, Hall MG, Parker GJ, Alexander DC. Camino: Open-source diffusion–MRI reconstruction and processing. In 14th Scientific Meeting of the International Society for Magnetic Resonance in Medicine. Seattle, USA; 2006, p. 2759. Sweet A, Alexander D. Reduced Encoding Persistent Angular Structure. In 14th Scientific Meeting of the International Society for Magnetic Resonance in Medicine. Stockholm, Sweden; 2010. Parker GJ, Haroon HA, Wheeler-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 2003; 18: 242–254. Cicchetti D. The precision of reliability and validity estimates revisited: Distinguishing between clinical and statistical significance of sample size requirements. J. Clin. Exp. Neuropsychol. 2001; 23: 695–700. Benjamin C, Singh J, Prabhu S, Warfield S. Optimization of tractography of the optic radiations. Hum. Brain Mapp. 2014; 35: 683–697. Govindan RM, Chugani HT, Makki MI, Behen ME, Dornbush J, Sood S. Diffusion tensor imaging of brain plasticity after occipital lobectomy. Pediatr. Neurol. 2008; 38: 27–33. Kamada K, Todo T, Morita A, Masutani Y, Aoki S, Ino K, Kawai K, Kirino T. Functional monitoring for visual pathway using real-time visual evoked potentials and optic-radiation tractography. Neurosurgery 2005; 57(1 Suppl.): 121–127. Bassi L, Ricci D, Volzone A, Allsop J, Srinivasan L, Pai A, Ribes C, Ramenghi L, Mercuri E, Mosca F, Edwards A, Cowan F, Rutherford M, Counsell S. Probabilistic diffusion tractography of the optic radiations and visual function in preterm infants at term equivalent age. Brain 2008; 131: 573–582.
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY 34. Ciccarelli O, Parker G, Toosy A, Wheeler-Kingshott C, Barker G, Boulby P, Miller D, Thompson A. From diffusion tractography to quantitative white matter tract measures: A reproducibility study. Neuroimage 2003; 18: 348–359. 35. Taoka T, Sakamoto M, Nakagawa H, Nakase H, Iwasaki S, Takayama K, Taoka K, Hoshida T, Sakaki T, Kichikawa K. Diffusion tensor tractography of the Meyer loop in cases of temporal lobe resection for temporal lobe epilepsy: Correlation between postsurgical visual field defect and anterior limit of Meyer loop on tractography. Am. J. Neuroradiol. 2008; 29: 1329–1334.
36. Tuch DS, Reese TG, Wiegell MR, Makris N, Belliveau JW, Wedeen VJ. High angular resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity. Magn. Reson. Med. 2002; 48: 577–582.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article at the publisher’s web site.
431
NMR Biomed. 2015; 28: 423–431
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
Revised: 19 November 2014,
Accepted: 4 January 2015,
Published online in Wiley Online Library: 23 February 2015
(wileyonlinelibrary.com) DOI: 10.1002/nbm.3266
Tractography of the optic radiation: a repeatability and reproducibility study Michael Dayana*, Sylvia Kreutzera,b and Chris A. Clarka Our main objective was to evaluate the repeatability and reproducibility of optic radiation (OR) reconstruction from diffusion MRI (dMRI) data. 14 adults were scanned twice with the same 60-direction dMRI sequence. Peaks in the diffusion profile were estimated with the single tensor (ST), Q-ball (QSH) and persistent angular structure (PAS) methods. Segmentation of the OR was performed by two experimenters with probabilistic tractography based on a manually drawn region-of-interest (ROI) protocol typically employed for OR segmentation, with both standard and extended sets of ROIs. The repeatability and reproducibility were assessed by calculating the intra-class correlation coefficient (ICC) of intra- and inter-rater experiments, respectively. ICCs were calculated for commonly used dMRI metrics (FA, MD, AD, RD) and anatomical dimensions of the optic radiation (distance from Meyer’s loop to the temporal pole, ML-TP), as well as the Dice similarity coefficient (DSC) between the raters’ OR segmentation. Bland–Altman plots were also calculated to investigate bias and variability in the reproducibility measurements. The OR was successfully reconstructed in all subjects by both raters. The ICC was found to be in the good to excellent range for both repeatability and reproducibility of the dMRI metrics, DSC and ML-TP distance. The Bland– Altman plots did not show any apparent systematic bias for any quantities. Overall, higher ICC values were found for the multi-fiber methods, QSH and PAS, and for the standard set of ROIs. Considering the good to excellent repeatability and reproducibility of all the quantities investigated, these findings support the use of multi-fiber OR reconstruction with a limited number of manually drawn ROIs in clinical applications utilizing either OR microstructure characterization or OR dimensions, as is the case in neurosurgical planning for temporal lobectomy. Copyright © 2015 John Wiley & Sons, Ltd. Additional supporting information may be found in the online version of this article at the publisher’s web site. Keywords: optic radiation; Meyer’s loop; diffusion imaging; tractography; reproducibility; reliability; repeatability
INTRODUCTION
NMR Biomed. 2015; 28: 423–431
* Correspondence to: M. Dayan, Developmental Imaging and Biophysics Section (DIBS), Institute of Child Health, University College London, 30 Guilford Street, London, WC1N 1EH, UK. E-mail: [email protected] a M. Dayan, S. Kreutzer, C. A. Clark UCL Institute of Child Health, London, UK b S. Kreutzer Institute of Neuroscience and Medicine, INM-3, Research Center Jülich, Jülich, Germany Abbreviations used: AD, axial diffusivity; dMRI, diffusion MRI; DSC, Dice similarity coefficient; DT, diffusion tensor; FA, fractional anisotropy; ICC, intra-class correlation coefficient; LGN, lateral geniculate nucleus; MD, mean diffusivity; ML, Meyer’s loop; NEX, bumber of excitations; OR, optic radiation; PAS, persistent angular structure; PICo, probability index of connectivity; QSH, Q-ball spherical harmonics; RD, radial diffusivity; ROI, region-of-interest; SNR, signal-to-noise ratio; ST, single tensor; TP, temporal pole; WM, white matter; RATI _ TPJ _ EXPK, rater I, timepoint J, experiment K.
Copyright © 2015 John Wiley & Sons, Ltd.
423
One set of white matter (WM) tracts of special interest in neuroimaging is the optic radiation (OR), a bundle of fibers connecting the lateral geniculate nucleus (LGN, part of the thalamus) to the primary visual cortex. This WM bundle has been the subject of much focus using neuroimaging, as it is a main component of the visual system. Of crucial importance is the ability to characterize its topology for neurosurgery planning, e.g. for temporal lobectomy (1) to avoid visual field defects in patients with intractable epilepsy, as well as its state when affected by diseases, e.g. in multiple sclerosis (2). WM tracts can be reconstructed in vivo with diffusion magnetic resonance imaging (dMRI) combined with tractography. The motion of water molecules in the brain causes an attenuation of the MRI signal in each voxel of the imaged volume and is constrained by WM fiber microstructure and arrangement (3). From dMRI measurements in several directions, one can estimate the water molecular diffusion profile (4), the peaks of which are interpreted as corresponding to the WM fiber directions. The simplest diffusion model is the diffusion tensor (DT) (5), the diffusion profile of which can be represented by an ellipsoid. Tensor elements are used to provide rotationally invariant metrics characterizing the shape of this ellipsoid and a number of important diffusion metrics can be derived from it. They include fractional anisotropy (FA) and mean diffusivity (MD), as well as axial and radial diffusivities (AD and RD, respectively), the values of which can be interpreted as characterizing the underlying microstructure. In the case of the DT, the ellipsoid can provide only a single
peak and therefore a single fiber-bundle orientation. Model-free methods such as Q-ball spherical harmonics (QSH), first introduced by (6), and persistent angular structure (PAS), described by (7), calculate the diffusion profile directly from the measurements without assumptions on its shape and allow the estimation of multiple bundle directions per voxel. Connecting fiber directions from voxel to voxel to reconstruct entire white matter tracts is the basis of tractography (8,9), which initiates tracts from a ‘seed’ region of interest (ROI) and often relies on other selection and exclusion ROIs to reconstruct specific WM tracts.
M. DAYAN ET AL. Since the introduction of tractography, in vivo anatomical depiction of the OR has been possible. Most studies dedicated to OR tractography have relied on careful ROI positioning by experimenters (10–14,2,15–20). The main drawback of tractography is the difficulty of validating in vivo reconstructions of a particular subject. The gold standard for this purpose is comparison of the subject’s in vivo segmentation with histology, and thus is only possible in animals. As a result, human studies investigating tractography validation rely on known anatomy in the population obtained from dissection and histology studies (21,22). In the case of the OR, typical comparisons involve the distances separating its anterior tip, Meyer’s loop (ML), from well-defined landmarks (10,11,13,15,1,19). However, not only is the accuracy of tractography reconstructions of importance, but also its reliability (how much variation characterizes repeated measurements), which includes reproducibility (how much variation characterizes results from different experimenters) and repeatability (how much variation characterizes results from the same experimenter under the same conditions). These issues are particularly important for translation of OR tractography into the clinical setting. The aim of the present work was to assess the repeatability and reproducibility of an OR tractography method based on manually drawn ROIs by comparing results obtained from the same experimenter and from two different experimenters, respectively. The outcome of this study informs us regarding the usability of OR reconstruction methods similar to the one described, for both research and practical clinical purposes.
Data processing Raw DICOM slices and volumes from the scanner acquisition were sorted and converted into Analyze volumes with a custom script based on the Tractor software (23). dMRI volumes were corrected for eddy currents and small head movements by registration with the first non-diffusion-weighted volume with a 12-parameter affine transformation using FSL (24). The diffusion gradient directions were updated accordingly by applying to them the rigid-body transformation associated with the previous transformation. The brain was then segmented with the BET utility (25), in which parameters were adjusted after iterative visual inspection. The diffusion tensor (DT) coefficients were computed with weighted linear least-squares regression with the Camino software (26) and then used in the calculation of the dMRI metrics, namely FA, MD, AD and RD. The main diffusion directions were estimated with the single tensor (ST) model as well as with QSH and PAS, in order to not only compare single-fiber and multi-fiber methods but also evaluate the importance of the method choice for estimating fiber directions. The maximum order of the spherical harmonic series for QSH was set at 6, while the number of basis functions for the PAS representation was chosen to be 16 (27). Tractography
METHODS Subjects The study took place at Great Ormond Street Hospital, London, UK. This work was granted ethical approval by the local ethics committee. 14 healthy adults without any known medical condition took part in the study. Informed consent was obtained in all subjects before their participation. The cohort included 10 males and 4 females, with an age range from 21–65 years. Imaging
424
Each participant underwent a dMRI protocol on a Siemens Avanto 1.5 T clinical system (Siemens Healthcare, Erlangen, Germany), using a self-shielding gradient set with maximum gradient strength of 40 mT m 1 and standard ‘birdcage’ quadrature headcoil. Echoplanar diffusion-weighted images were acquired for an isotropic set of 60 non-collinear directions, using a weighting factor of 1000 s mm 2, along with three T2 weighted (b = 0) volumes and with a repetition factor (or number of excitations (NEX)) of 2 to increase the signal-to-noise ratio (SNR). 45 contiguous axial slices of thickness 2.5 mm were imaged, using a field of view of 240 × 240 mm 2 and 96 × 96 voxel acquisition matrix, for a final image resolution of 2.5 × 2.5 × 2.5 mm 3. The echo time was 89 ms and the repetition time was 6300 ms. In addition, a T1 -weighted 3D FLASH structural image was acquired using 176 contiguous sagittal slices, a 256 × 224 mm 2 field of view and 1 × 1 × 1 mm 3 image resolution. The echo time in this case was 4.9 ms and the repetition time was 11 ms. Once the data were acquired (timepoint 1), the same protocol was repeated during a second scan session undertaken approximately one week later
wileyonlinelibrary.com/journal/nbm
(timepoint 2). The dMRI sequence was also repeated within the first session and, as a result, the overall scan time for the acquisition was approximately 50 min for the first session and 30 min for the second session.
The tractography algorithm run was the PICo algorithm, as implemented in Camino. This algorithm relies on repeating the tractography process, and at each iteration changing the fiber direction(s) according to the associated uncertainty. The typical output of PICo is a probability map, in which each voxel contains the number of times a tract went through that voxel, divided by the total number of tracking iterations (28). In order to perform tractography for the OR, ROIs were defined on the color-coded FA maps, on which the main diffusion directions were overlaid. As described in a previous work (19), a 16-voxel seed region was placed on one coronal slice, slightly anterolaterally to the LGN, and a waypoint ROI including the striatum sagittale (of which the OR is part) was drawn on another coronal slice. Both ROIs were drawn in regions of high FA, where the main diffusion directions were aligned along the anteroposterior directions (i.e. with high green intensity on the FA color-coded map). Two sets of exclusion ROIs were defined solely according to the main diffusion directions, in order to exclude neighbouring tracts to ML, and both included whole plane exclusion masks in inferior, superior and mid-sagittal sections. A ‘standard’ set of exclusion ROIs (Fig. 1, rows 1 and 2), abbreviated thereafter as ‘std set’, featured ROIs placed laterally to prevent tracking of the acoustic radiation, medially to avoid the generated tracts following the anterior commissure and forceps major, and anteriorly to prevent tracking of bundles continuing along the inferior occipitofrontal fasciculus and uncinate fasciculus. An ‘extended’ set, abbreviated as ‘ext set’, was also defined, as some extra artifactual tracts could remain after the implementation of the ‘std set’ in the tractography process. The ‘ext set’ included an additional medial ROI to remove fibers connecting to the cingulum (Fig. 1, row 3) and lateral ROI to exclude fibers coursing to V5 (Fig. 1, row 4).
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY
Figure 1. Sagittal, coronal and axial slices of a single subject FA map colored according to the principal diffusion direction (green: anterior–posterior, blue: superior–inferior, red: left–right), illustrating the placement of the ROIs used in the tractography analysis. ROIs belonging to the ‘std’ set are shown in orange in rows 1 and 2 and those of the ‘ext’ set are shown in blue in rows 3 and 4. Overview of the ‘std’ set ROIs, which are all (1) whole-plane exclusion ROIs, except for (2) the medial ROI, which does not include the occipital lobe, and overview of the ‘ext set’, which includes an additional (3) medial and (4) lateral ROI. Please note that, in all panels, the position of the sagittal slice is indicated by the thin green cursor lines in the associated coronal and axial slices.
NMR Biomed. 2015; 28: 423–431
gray matter regions, whilst avoiding too high a number of spurious tracts associated with voxels exhibiting high uncertainty in the diffusion directions. The maximum intensity projection of the probabilistic maps generated for each seed voxel was displayed with a linear scale, from red to yellow, and with a
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
425
Once the ROIs were created, the PICo algorithm was run for each fiber reconstruction method (ST, QSH, PAS) without any angular threshold (similarly to (15) and (1)) to account for the high curvature of ML. A relatively low FA threshold, 0.1, was used to allow the tracts to reach the extremities of the OR ending in
M. DAYAN ET AL. threshold of 1 % to eliminate artifactual tracts visually without removing plausible OR bundle reconstructions. Calculation of dMRI metrics The parameters selected to quantify the microstructural characteristics of the segmented OR were the average FA, MD, AD and RD. The computation of their values was achieved using a custom MATLAB® script. It consisted of selecting only voxels with probabilities greater than the chosen PICo threshold of 1 %, which resulted in the creation of a binary mask. This threshold was chosen, as it was considered to correspond to the most visually convincing reconstruction of the OR. The next step was to select the area corresponding to the OR in each of the dMRI metrics volumes (e.g. the FA volume) and keep only voxels with PICo probability greater than 1 %. This step was carried out by a voxel-by-voxel multiplication with the binary mask previously computed. An average over the resulting thresholded area was then calculated and provided the dMRI metric mean corresponding to a 1 % PICo threshold. Neuroanatomical distances of the OR To assess the spatial location of ML and compare the results with reference dissection studies, the distance from its tip to the temporal pole (ML-TP) was measured. The tip of ML was identified by visual inspection of the PICo probability maps displayed with a threshold of 1 %. The main diffusion directions were superimposed on to these maps to confirm that the most anterior voxels forming ML, highlighted on the probability maps, were indeed part of the reconstructed OR. The position of the most anterior voxel belonging to the OR was chosen as the location of the tip of ML. The location of the temporal pole (TP) was evaluated on the color-coded FA maps and chosen as the coronal slice coinciding with the most anterior extent of the temporal lobe – including the TP – and the beginning of the frontal lobe (19). In cases where several coronal plane positions along the anterior–posterior axis fit this criterion, the most anterior position was systematically selected. The ML-TP distance was computed as the difference between the tip of ML and the TP.
at timepoint 1: INTER = ICC(RAT1 _ TP1 _ EXP1, RAT2 _ TP1 _ EXP1). The metrics calculated for reproducibility were as follows: • the Dice similarity coefficient (DSC), which informs us regarding tract shape similarity and is only calculated for reproducibility measures involving two different raters (INTRA_SAMESCAN and INTER), DSC ðEXP1 ; EXP2 Þ ¼
2ðEXP1 voxels∩ EXP2 voxelsÞ EXP1 voxels þ EXP2 voxels
• dMRI metrics (FA, MD, AD, RD), which are often relied upon in clinical research using tractography, and • the ML-TP distance, which plays an important role in neurosurgery planning. It is important to underscore that rater 1 had no prior experience of MRI imaging and was trained during a single day by rater 2, who had substantial experience in dMRI and OR anatomy. Tractography reproducibility was further assessed graphically by generating Bland–Altman plots. These plots represent the difference M1(Si) M2(Si) between two repeated measures M1 and M2 on a subject Si as a function of this measure mean, [M1(Si) + M2(Si)]/2, and feature three constants shown as horizontal lines: a solid line for the mean difference between the repeated measures on all N subjects, 1 N M1 ðSi Þ þ M2 ðSi Þ ∑ N i¼1 2 and dashed lines defined such as to include 95 % of the points, called the 95 % lines of agreement. Such plots can reveal any systematic bias (if the mean difference represented by the solid line is different from 0) and allow us to estimate whether the amount of variability is acceptable, i.e. if the interval of variation within the 95 % lines of agreement is not much larger than the effect to be measured.
RESULTS Tractography reconstruction
Reproducibility analysis
426
To assess the reproducibility of the tractography method, the data obtained at timepoint 1 and timepoint 2 (cf. Imaging subsection) were analyzed by two raters to evaluate three reproducibility measures estimated with the intra-class correlation coefficient (ICC). The ranges ICC < 0.40, 0.40 ≤ ICC < 0.60, 0.60 ≤ ICC < 0.75 and 0.75 ≤ ICC indicate poor, fair, good and excellent reliability, respectively (29). The intra-rater reproducibility (INTRA_SAMESCAN) for data acquired from the same scanning session at timepoint 1 (TP1) was evaluated by comparing two tractography experiments (EXP1 and EXP2) undertaken by rater 1 (RAT1) and separated by a time interval of at least one week: INTRA _ SAMESCAN = ICC(RAT1 _ TP1 _ EXP1, RAT1 _ TP1 _ EXP2). The intra-rater variability including scanning variability (INTRA_RESCAN) was estimated by comparing results obtained by the same rater (RAT1) from data acquired at timepoint 1 and timepoint 2: INTRA _ RESCAN = ICC(RAT1 _ TP1 _ EXP1, RAT1 _ TP2 _ EXP1). Finally, the inter-rater reproducibility was evaluated from the comparison between results obtained by two different raters (RAT1 and RAT2) on the same data acquired
wileyonlinelibrary.com/journal/nbm
An example of probability maps obtained for the ‘std’ and ‘ext’ sets from one subject (with fiber directions estimated with the PAS model and PICo threshold 1 %) is shown in Figure 2. The value of the tractography probability in each voxel is displayed on a linear scale from just above 0 (one track went through that voxel) to 1 (all the tracts generated passed through that voxel), while a grayscale FA map is shown in the background. For the ‘std set’, the probability map color scale was chosen from red to yellow and for the ‘ext set’ from dark blue to cyan. As the ‘std set’ excludes fewer fibers, the associated probability map features more voxels than the probability map of the ‘ext set’. The entire OR could be reconstructed with ML clearly visible in all subjects, irrespective of the method chosen for estimating the fiber directions. Tract shape similarity The DSC values of most repeatability and reproducibility measures (INTRA_SAMESCAN, INTRA_RESCAN, INTER) were in the [0.6, 1] range, corresponding to values ranging from good ([0.6, 0.75]) to excellent ([0.75, 1]) agreement (Fig. 3).
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY
Figure 2. Example of a probabilistic map of the OR with PICo threshold of 1 % obtained for the ‘std’ and ‘ext’ sets, represented in sagittal (left), coronal (middle) and axial (right) planes. The OR reconstruction is displayed using a linear color scale ranging from red to yellow for the ‘std’ set and from dark blue to cyan for the ‘ext’ set. The ‘std’ set resulted in more voxels included in the OR, due to the less stringent exclusion ROIs. The extra voxels tended to be part of tracts coursing towards the cingulum (circled in green) and V5 (circled in orange).
Figure 3. Dice similarity coefficient (DSC) for the INTRA_SAMESCAN (top), INTRA_RESCAN (middle) and INTER (bottom) reproducibility measures and the ST, QSH and PAS reconstruction methods. Results of the left and right hemispheres are shown in the left and right columns, respectively. Values of the DSC associated with the ‘std’ and ‘ext’ mask sets are shown in blue and pink. Brown, orange and green dashed lines indicate threshold values separating fair, good and excellent DSC: 0.4, 0.6 and 0.75, respectively.
NMR Biomed. 2015; 28: 423–431
categorized as ‘good’. For the INTRA_RESCAN measure, PAS and QSH methods provided ‘good’ DSC, while the ST model values were exactly at or just below the threshold for the ‘good’ category. The general trend for DSC values was to be larger for
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
427
For the INTRA_SAMESCAN and INTER measures, excellent DSC were obtained for the PAS and QSH methods, although for the latter the extended set of masks resulted more often in ‘good’ DSC, and DSC for the ST model corresponded to values
M. DAYAN ET AL. multifiber reconstructions (QSH and PAS) – with PAS resulting in the largest DSC for most cases – and larger for the standard set of masks compared with the extended one. Anatomical measurements For a given tractography experiment, the ML-TP distance was found to be the same for each fitting method (ST, QSH and PAS). The ICC for the ML-TP distance was found to be good (0.60 ≤ ICC < 0.75) for all repeatability and reproducibility measures (INTRA_SAMESCAN, INTRA_RESCAN, INTER) in both hemispheres, except in the left hemisphere of the INTRA_SAMESCAN, where it was excellent (0.75 ≤ ICC, Fig. 4).
Figure 4. ML-TP distance ICC for the INTRA_SAMESCAN, INTRA_RESCAN and INTER reproducibility measures. The ICC for left and right hemispheres are indicated in blue and green, respectively.
The ICC was shown to be systematically smaller in the right hemisphere. The Bland–Altman plots did not demonstrate any bias, while the 95 % lines of agreement were around ± 1 voxel (2.5 mm) in the left hemisphere and around ± 2 voxels (5 mm) in the right hemisphere. dMRI metrics For all repeatability and reproducibility measures, all reconstruction methods and both hemispheres, the ICC of all dMRI metrics were in the good ([0.6, 0.75]) to excellent ([0.75, 1]) range (Fig. 5), with the only exception being for the INTRA_SAMESCAN repeatability of FA within the right OR reconstructed with fiber directions estimated with the PAS method. For the INTRA_RESCAN reproducibility, dMRI metrics were good to excellent for ST, mostly excellent for QSH and all excellent for PAS. For QSH and PAS, the standard set of masks tends to give slightly more reproducible results, however the opposite was true for the ST model. For the INTER reproducibility, dMRI metrics were good to excellent for both ST and QSH and mostly excellent for PAS, with the exception already mentioned of FA in the right OR. For the INTER reproducibility, the dMRI metrics ICC were all excellent. A trend between standard and extended mask sets similar to that of the INTRA_SAMESCAN repeatability was observed in the INTRA_RESCAN and INTER measures. The Bland–Altman plot of dMRI metrics from the INTRA_SAMESCAN measure had no apparent systematic bias, all mean differences between raters being close to zero and the 95 % lines of agreement globally similar in absolute value (see Fig. S1 in the Supporting Information). A general trend was for the ST model to have larger variation between raters compared with the QSH and PAS methods: FA differences had 95 % lines of agreement larger than ±0.05 and axial, radial and mean
428
Figure 5. dMRI metrics ICC for the INTRA_SAMESCAN (top), INTRA_RESCAN (middle) and INTER (bottom) reproducibility measures and the PAS (left), QSH (middle) and ST (right) reconstruction methods. For each given pair of reproducibility measure and reconstruction method, the ICC for the left and right hemispheres are shown on the left and right, respectively. Values of the ICC associated with the ‘std’ and ‘ext’ mask sets are shown in blue and pink. Brown, orange and green dashed lines indicate threshold values separating fair, good and excellent ICC: 0.4, 0.6 and 0.75, respectively.
wileyonlinelibrary.com/journal/nbm
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY diffusivities larger than ± 0.5 × 10 3 mm 2 s 1 for the ST model, while these values were smaller for QSH and PAS, more markedly for the latter. The left and right hemispheres presented globally the same pattern. The Bland–Altman plot related to the INTRA_RESCAN and INTER_RESCAN reproducibility measures exhibited the same behaviour, with slightly more variation between raters for the INTRA_RESCAN reproducibility (see Figs. S2 and S3 in the Supporting Information).
DISCUSSION Tractography reconstruction shape and dimensions
NMR Biomed. 2015; 28: 423–431
dMRI metrics extracted from the OR The fact that dMRI metrics demonstrated excellent ICC (0.75 ≤ ICC) for QSH and PAS for the INTRA_SAMESCAN and INTER reliability measures, compared with good ICC (0.60 ≤ ICC < 0.75) for ST, concurred with the DSC findings. Another common characteristic with the shape reliability is that the extended set of masks provided more variability than the standard set, again probably due to the added complexity of positioning the extra exclusion masks. It can also be seen in Figure 5 that FA in general had lower ICC compared with other dMRI metrics. This could be due to the higher sensitivity of FA to change in microstructure, which for this very reason is also the most often used metric in clinical studies based on diffusion imaging. As with the DSC results, the INTRA_RESCAN reliability measure is generally lower than both the INTRA_SAMESCAN and INTER measures. This was counterintuitive, as more variation was expected when comparing different raters. Since the influence of subject re-scanning is assumed to be minimal, this difference was thought to arise from the multitude of landmarks to consider when positioning ROIs and the inexperience of rater 1, who was given a single day of training. Consequently, results were more similar between the trainer and trainee than within the own trainee experiments. It can be inferred that, although a single training day was enough to obtain reproducible results, it is highly likely that higher reproducibility could be obtained with a longer training period. More than a day of training therefore seems recommended. The Bland–Altman plots illustrated the highest reliability gained from using multi-fiber reconstruction methods. The ST model had the most spread 95 % lines of confidence for all reliability measures, although in absolute value these were around 0.03 for FA, ± 0.5 × 10 3 mm 2 s 1 for MD and RD, and ± 0.1 × 10 3 mm 2 s 1 for AD. These values are small, but still larger than the effects that can be found as significant in clinical studies. This is, however, not an issue, since the Bland–Altman plots demonstrated that no systematic bias existed in either reproducibility or repeatability and, as a result, the importance of these variations would decrease with the size of the sample studied, such that it should be possible to detect effects even an order of magnitude smaller than these values with a large cohort.
CONCLUSION We demonstrated in this work that, despite being challenging to segment, the OR could be reconstructed from ROI-based tractography methods, similar to the most common methods
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
429
Tractography reconstruction of the optic radiation is notoriously difficult, due to the high curvature of ML and the high variability of individual trajectories. Tractography methods of the OR based on ROI locations can differ subsequently, notably due to the choice of seed ROI (30). In this work, positioning the seed ROI not exactly in the LGN but lateral to it avoided GM voxels where FA is low and thus where diffusion directions are associated with high uncertainty. Importantly, this was of particular interest for estimating the subsequent segmentation repeatability and reproducibility, as many studies have employed a similar seed placement (11,31,15,32–35,17). Selecting a single coronal slice and defining an ROI spanning the entire cross-section of the OR enabled us to recover both its medial and lateral aspects. The visual inspection of the tracts and topological measurements from surrounding anatomical landmarks demonstrated that such a seed ROI was effective for reconstructing the general shape of the OR – including ML – when combined with a waypoint ROI in the sagittal stratum and specific exclusion masks, whatever the method used to estimate the fiber directions, be it ST, QSH or PAS. The standard set of exclusion masks was relatively simple to define, while the extended set required more attention given to the surrounding structures. This was reflected in the results, as discussed subsequently, for which the extended set of masks consistently resulted in higher variation in repeatability and reproducibility. The DSC results demonstrated that there was generally good repeatability and reproducibility, although the ST model was at the boundary between moderate and good agreement for the INTRA_RESCAN reliability measure. The ST only allows for one fiber direction reconstructed per voxel and therefore imposes more restriction during the tracking process. A change in the ROIs would thus have more effect with the ST model compared with multi-fiber methods and so provides a possible explanation for these findings. The good to excellent reliability of the OR DSC concurred with visual inspection, according to which all tracts appeared correctly segmented, with the same anterior tip position found for all fiber direction methods used (ML-TP was the same for ST, QSH and PAS). The ML-TP distance was categorized as excellent or close to excellent (0.75 ≤ ICC) for all reliability measures in the left hemisphere, while it was categorized as good (0.60 ≤ ICC < 0.75) in the right hemisphere. This result was probably caused by an asymmetry in the OR itself (19), likely to have a slightly different fiber configuration – perhaps more complex with a higher level of voxels containing multiple fibers in the right OR. A similar trend was demonstrated in the Bland– Altman plots, where higher variability was found between raters in the right OR (± 2 voxels, i.e. 5 mm) compared with the left (± 1 voxel, i.e. 2.5 mm), although no bias was apparent in either case. The amount of variability in extent is consistent with a slice
position uncertainty of ± 1 voxel for defining both the TP position and ML position. As mentioned above, visual inspection ascertained that all different methods provided a satisfactory reconstruction of the OR and that the most anterior tip of ML was at the same location for each of them. The outputs of all these methods were therefore globally comparable and thus all deemed acceptable for conventional OR tractography. We did not investigate multiple-tensor fitting, though it is available in Camino, as this method requires us to detect multi-fiber population voxels, is unstable when fitting more than two fiber directions (36) and is therefore outperformed by most existing multi-fiber algorithms.
M. DAYAN ET AL. reported in the OR literature, with good to excellent repeatability and reproducibility as assessed from its shape, dimensions and averaged dMRI metrics. Reliability was shown to be enhanced when the complexity of the set of ROIs drawn was reduced and when tractography was performed on data obtained with multi-fiber reconstruction algorithms rather than the ST model. These results are encouraging and further support the use of OR tractography methods based on manually drawn ROIs, for studies investigating microstructural changes with dMRI metrics as well as for neurosurgery planning, in which knowledge of the ML-TP distance is important. The fact that the ML-TP distance was not found to differ between the ST model and the multi-fiber methods suggests that OR tractography with the simpler ST model is adequate when only a limited number of diffusion directions is available for dMRI acquisition.
16. 17. 18.
19.
20.
Acknowledgements This work was supported by the UK Medical Research Council (G0300117) and Engineering and Physical Sciences Research Council (EP/C536851/1). CAC acknowledges European Union Grant FP7-ICT-2009-C 238292.
REFERENCES
22.
23.
430
1. Winston G, Mancini L, Stretton J, Ashmore J, Symms M, Duncan J, Yousry T. Diffusion tensor imaging tractography of the optic radiation for epilepsy surgical planning: A comparison of two methods. Epilepsy Res. 2011; 97: 124–132. 2. Reich D, Smith S, Gordon-Lipkin E, Ozturk A, Caffo B, Balcer L, Calabresi P. Damage to the optic radiation in multiple sclerosis is associated with retinal injury and visual disability. Arch. Neurol. 2009; 66: 998–1006. 3. Beaulieu C. The basis of anisotropic water diffusion in the nervous system – a technical review. NMR Biomed. 2002; 15: 435–455. 4. Yablonskiy D, Sukstanskii A. Theoretical models of the diffusion weighted MR signal. NMR Biomed. 2010; 23: 661–681. 5. Basser PJ, Mattiello J, Le Bihan D. Estimation of the effective selfdiffusion tensor from the NMR spin echo. J. Magn. Reson., Ser. B 1994; 103: 247–254. 6. Tuch D. Q-ball imaging. Magn. Reson. Med. 2004; 52: 1358–1372. 7. Jansons K, Alexander D. Persistent angular structure: New insights from diffusion magnetic resonance imaging data. Inverse Probl. 2003; 19: 1031–1046. 8. Mori S, Crain BJ, Chacko VP, VanZijl PCM. Three-dimensional tracking of axonal projections in the brain by magnetic resonance imaging. Ann. Neurol. 1999; 45: 265–269. 9. Lazar M. Mapping brain anatomical connectivity using white matter tractography. NMR Biomed. 2010; 23: 821–835. 10. Yamamoto T, Yamada K, Nishimura T, Kinoshita S. Tractography to depict three layers of visual field trajectories to the calcarine gyri. Am. J. Ophthalmol. 2005; 140: 781–785. 11. Nilsson D, Starck G, Ljungberg M, Ribbelin S, Jönsson L, Malmgren K, Rydenhag B. Intersubject variability in the anterior extent of the optic radiation assessed by tractography. Epilepsy Res. 2007; 77: 11–16. 12. Xie S, Gong GL, Xiao JX, Ye JT, Liu HH, Gan XL, Jiang ZT, Jiang XX. Underdevelopment of optic radiation in children with amblyopia: A tractography study. Am. J. Ophthalmol. 2007; 143: 642–646. 13. Sherbondy AJ, Dougherty RF, Napel S, Wandell BA. Identifying the human optic radiation using diffusion imaging and fiber tractography. J. Vis. 2008; 8: 1–11. 14. Chen X, Weigel D, Ganslandt O, Buchfelder M, Nimsky C. Prediction of visual field deficits by diffusion tensor imaging in temporal lobe epilepsy surgery. Neuroimage 2009; 45: 286–297. 15. Yogarajah M, Focke NK, Bonelli S, Cercignani M, Acheson J, Parker GJM, Alexander DC, McEvoy AW, Symms MR, Koepp MJ, Duncan JS. Defining Meyer’s loop-temporal lobe resections, visual field
wileyonlinelibrary.com/journal/nbm
21.
24.
25. 26.
27.
28.
29.
30. 31. 32.
33.
deficits and diffusion tensor tractography. Brain 2009; 132: 1656–1668. Hofer S, Karaus A, Frahm J. Reconstruction and dissection of the entire human visual pathway using diffusion tensor MRI. Front. Neuroanat. 2010; 4: 1–7. White M, Zhang Y. Three-tesla diffusion tensor imaging of Meyer’s loop by tractography, color-coded fractional anisotropy maps, and eigenvectors. Clin. Imaging 2010; 34: 413–417. Wu W, Rigolo L, O’Donnell L, Norton I, Shriver S, Golby A. Visual pathway study using in vivo diffusion tensor imaging tractography to complement classic anatomy. Neurosurgery 2012; 70: 145–156 (Discussion 156). Dayan M, Muñoz M, Jentschke S, Chadwick M, Cooper J, Riney K, Vargha-Khadem F, Clark C. Optic radiation structure and anatomy in the normally developing brain determined using diffusion MRI and tractography. Brain Struct. Funct. 2015; 220: 291–306. Kuhnt D, Bauer M, Sommer J, Merhof D, Nimsky C. Optic radiation fiber tractography in glioma patients based on high angular resolution diffusion imaging with compressed sensing compared with diffusion tensor imaging – initial experience. PLoS One 2013; 8(7). Seehaus A, Roebroeck A, Chiry O, Kim DS, Ronen I, Bratzke H, Goebel R, Galuske R. Histological validation of DW–MRI tractography in human postmortem tissue. Cereb. Cortex 2013; 23: 442–450. Assaf Y, Alexander DC, Jones DK, Bizzi A, Behrens TEJ, Clark CA, Cohen Y, Dyrby TB, Huppi PS, Knoesche TR, LeBihan D, Parker GJM, Poupon C. The CONNECT project: Combining macro- and microstructure. Neuroimage 2013; 80: 273–282. ISSN 1053-8119; http:// dx.doi.org/10.1016/j.neuroimage.2013.05.055 Clayden JD, Jentschke S, Muñoz M, Cooper JM, Chadwick MJ, Banks T, Clark CA, Vargha-Khadem F. Normative development of white matter tracts: similarities and differences in relation to age, gender, and intelligence. Cereb. Cortex 2012; 22: 1738–1747. Smith S, Jenkinson M, Woolrich M, Beckmann C, Behrens T, Johansen-Berg H, Bannister P, DeLuca M, Drobnjak I, Flitney D, Niazy R, Saunders J, Vickers J, Zhang Y, DeStefano N, Brady J, Matthews P. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 2004; 23: S208–S219. Smith SM. Fast robust automated brain extraction. Hum. Brain Mapp. 2002; 17: 143–155. Cook PA, Bai Y, Gilani NS, Seunarine KK, Hall MG, Parker GJ, Alexander DC. Camino: Open-source diffusion–MRI reconstruction and processing. In 14th Scientific Meeting of the International Society for Magnetic Resonance in Medicine. Seattle, USA; 2006, p. 2759. Sweet A, Alexander D. Reduced Encoding Persistent Angular Structure. In 14th Scientific Meeting of the International Society for Magnetic Resonance in Medicine. Stockholm, Sweden; 2010. Parker GJ, Haroon HA, Wheeler-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 2003; 18: 242–254. Cicchetti D. The precision of reliability and validity estimates revisited: Distinguishing between clinical and statistical significance of sample size requirements. J. Clin. Exp. Neuropsychol. 2001; 23: 695–700. Benjamin C, Singh J, Prabhu S, Warfield S. Optimization of tractography of the optic radiations. Hum. Brain Mapp. 2014; 35: 683–697. Govindan RM, Chugani HT, Makki MI, Behen ME, Dornbush J, Sood S. Diffusion tensor imaging of brain plasticity after occipital lobectomy. Pediatr. Neurol. 2008; 38: 27–33. Kamada K, Todo T, Morita A, Masutani Y, Aoki S, Ino K, Kawai K, Kirino T. Functional monitoring for visual pathway using real-time visual evoked potentials and optic-radiation tractography. Neurosurgery 2005; 57(1 Suppl.): 121–127. Bassi L, Ricci D, Volzone A, Allsop J, Srinivasan L, Pai A, Ribes C, Ramenghi L, Mercuri E, Mosca F, Edwards A, Cowan F, Rutherford M, Counsell S. Probabilistic diffusion tractography of the optic radiations and visual function in preterm infants at term equivalent age. Brain 2008; 131: 573–582.
Copyright © 2015 John Wiley & Sons, Ltd.
NMR Biomed. 2015; 28: 423–431
TRACTOGRAPHY OF OPTIC RADIATION: A REPRODUCIBILITY STUDY 34. Ciccarelli O, Parker G, Toosy A, Wheeler-Kingshott C, Barker G, Boulby P, Miller D, Thompson A. From diffusion tractography to quantitative white matter tract measures: A reproducibility study. Neuroimage 2003; 18: 348–359. 35. Taoka T, Sakamoto M, Nakagawa H, Nakase H, Iwasaki S, Takayama K, Taoka K, Hoshida T, Sakaki T, Kichikawa K. Diffusion tensor tractography of the Meyer loop in cases of temporal lobe resection for temporal lobe epilepsy: Correlation between postsurgical visual field defect and anterior limit of Meyer loop on tractography. Am. J. Neuroradiol. 2008; 29: 1329–1334.
36. Tuch DS, Reese TG, Wiegell MR, Makris N, Belliveau JW, Wedeen VJ. High angular resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity. Magn. Reson. Med. 2002; 48: 577–582.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article at the publisher’s web site.
431
NMR Biomed. 2015; 28: 423–431
Copyright © 2015 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
