In this study, we investigated SWM structures using DTI. In past investigations, group-averaged DTI has been studied and many common axonal tracts were identified in the DWM regions. The locations and sizes of these structures are reproducible among normal subjects, and they can be clearly identified in the linear-normalized images. The existence of these tracts has been well-documented in numerous anatomical and histological studies (e.g., (Burgel et al., 2006
; Dejerine, 1895
; Flechsig, 1920
)) with their names assigned. In this paper, we extended these efforts to establish atlases of the SWM. This requires identifying common structures in the peripheral regions, where descriptions by previous studies are scarce.
Axons that connect the distal areas of the brain tend to merge with other axons that share similar destinations, forming large bundles in the DWM regions. These prominent WM tracts are major constituents of both the deep and the superficial WM. On the other hand, it has also been documented that there are many cortico-cortical short association fibers that are confined to the SWM regions (Dejerine, 1895
; Meynert, 1872
). For example, the existence of several short association fibers, such as the vertical occipital fascicle and the orbito-frontal fascicle, has been reported in textbooks, although the exact locations have not been defined (Kahle et al., 1986
). The situation is similar for short association fibers that connect adjacent gyri, called the U-fibers of Meynert (Meynert, 1872
). The existence of these fibers has also[S14]
been demonstrated by neuropathological studies (Cervos-Navarro et al., 1994
; Cobb et al., 1950
; Kurachi et al., 1999
), but locations of U-fibers have not been well identified three-dimensionally in the past. The study of common anatomic structures in the SWM has been further hampered by the significant amount of individual variability in the cortex; deciphering the common and individual-specific structural features would be of great importance and requires group analyses.
Normalization approach for the population-averaged atlas and identified blade structures
For the group analysis of the SWM, we adopted linear brain normalization, which effectively adjusts the overall size of the brain, but leaves details of brain structures not aligned among individuals. With this approach, most large tracts in the DWM can be identified, suggesting that their existence and locations are reproducible (). On the other hand, this approach cannot align detailed cortical structures adequately, which leads to significant blurring in population-averaged images. For the SWM, which is situated between the DWM and the cortex, we can assume that only prominent common structures are identified with this approach. Therefore, it is possible that there are more SWM structures that are common in normal populations, but unidentified in the present study.
After the normalization, the probabilistic WM map was created by averaging the FA-thresholded images of normal subjects, followed by the identification of the CSWM by probability thresholding. This approach identified blade-like structures of the CSWM, which have simpler structures than the cortices to which they are associated. After averaging linear-normalized images, even the most reproducible gyri, such as pre- and post central gyri, cannot be well-defined. However, the blades underneath these gyri are well defined ().
From the shape of the CSWM, we have identified nine large blades, the structures of which have a close relationship with the cortical structures, as shown in and . While[S15] there are some blades that have a one-to-one relationship to cortical areas, several blades include multiple cortical areas per blade (one blade ⊃ one or multiple cortical areas). The only exception is the fusiform WM that spans the temporal blade for the anterior part and the occipital blade for the posterior part. However, based on the definition of the Talairach atlas, the fusiform gyrus is known to consist of at least two regions; one in the temporal lobe and the other in the occipital lobe. Therefore, this result is understandable. We[S16] would like to stress that the nine blades were defined by visual inspection. While some blades have clear structural transitions from one blade to the other, the boundaries of several blades (e.g., the parieto-temporal blade and the occipital blade) have a certain degree of arbitrariness in our definition. This type of arbitrariness is often unavoidable in brain atlases because many anatomical structures inherently do not have clear boundaries.
Identification of SWM tracts that interconnect the blades
Using each blade as an ROI for tractography, we searched for WM tracts that interconnect the blades. Most tracts identified in this approach have already been well-characterized by histology and tractography in the past (Basser et al., 2000
; Catani et al., 2002
; Conturo et al., 1999
; Jellison et al., 2004
; Mori et al., 1999
; Mori et al., 2002
; Mori et al., 2005
; Poupon et al., 2000
; Stieltjes et al., 2001
; Wakana et al., 2004
). In addition, we identified four short association fibers and one long association fiber in the population-averaged map and also in individual DTI data. Among these fibers, the fronto-central short association fibers and the central short association fibers are elaborated in the previous studies of human and nonhuman primates (Barbas and Pandya, 1987
; Catani et al., 2002
; Dejerine, 1895
; Leichnetz, 1986
; Pandya and Kuypers, 1969
; Pandya and Vignolo, 1971
). Based on this prior information, they are likely to be real entities. On the other hand, the frontal and the parietal short association fibers and the parieto-temporal long association fibers, which we identified in this study, have not been well-documented previously, even though some of them were depicted rudimentarily in a textbook (Crosby et al., 1962
) without specific names. The frontal short association and the parieto-temporal long association fibers were also not clearly recognizable in two to three individuals (out of 10) tested in this study. Therefore, their character remains disputable.
Limitations of diffusion tensor imaging and interpretation of the results
The anatomical information used in this study is based on diffusion tensor imaging, in which measured water diffusion properties are fitted to a simple 3×3 tensor model[S17]
. This model, in which it is assumed that there is only one dominant fiber population within each pixel, is an oversimplification (Frank, 2001
; Tuch et al., 2002
). This could be an issue, especially for the SWM, in which fiber architectures are expected to be more complicated than the DWM. Therefore, we need to be cautious about the interpretation of the tractography results (-), which is based on fiber orientation information and prone to artifacts. For example, the lack of short association fibers by tractography does not necessarily mean there are no association fibers in such areas. There are two reasons we would expect false negatives. First, with the current image resolution (2.5 mm), we expected a mixture of fibers with multiple populations. In such areas, the FACT algorithm employed in this study provides conservative results (tracking does not penetrate such problematic areas). For[S18]
further investigation of the potentially complicated anatomy of the SWM, probabilistic tractography (Behrens et al., 2003
; Parker et al., 2002
), based on non-tensor models (Frank, 2001
; Tuch et al., 2002
), would be necessary. Second, the imperfect normalization process also leads to misalignment of small short association fibers. Unless short association fibers are the dominant component in the vicinity, their existence may be covered or averaged out by other more dominant fibers. We also cannot completely exclude a possibility that those short association fibers we identified are false positives. Because the exact coordinates of these short association fibers have not been well-characterized by previous histology studies, we cannot validate our findings easily. However, there could be several interesting applications based on our results. First, the fact that these fibers can be found by DTI reproducibly means we can identify the corresponding brain locations in the SWM of the normal population. While the blade structures can identify specific WM regions at a more macroscopic level, these short association fibers could define more pin-pointed areas in the SWM. This allows us to study WM (e.g., T2, FA, and ADC) and gray matter (e.g., cortical thickness) features of the areas associated with these short association fibers. Second, if group analyses show that one of the tractography-defined short association fibers are severely altered in a patient group (e.g., cannot be identified or are significantly smaller), such results strongly suggest differences in axonal architectures in the related areas.
Application of this atlas
The identification of common anatomic features of the SWM allows us to investigate the effects of brain diseases on these structures. For[S19] example, we can apply the atlas to report the location of plaques in MS patients. Using a conventional T1 or T2-weighted image, the location of the lesion could be reported in absolute coordinates (e.g. x, y, z coordinates in the patient frame) in normalized coordinates, but it is often difficult to relate it to specific white matter anatomy. Our SWM atlas adds new anatomical dimensions to evaluate the frequency of lesions for each white matter areas or to relate lesions locations and functional outcomes. As shown in , the atlas can be warped into individual subject data for automated brain parcellation, which allows volume and contrast (e.g., FA, ADC, T2, MTR, etc.) measurements in each area.
In this study, we employed affine transformation for brain normalization, which is a valid but crude operation. For example, the standard deviations of FA values reported in contain not only the real variability of FA of each structure, but also registration errors. This is apparent for such white matter tracts as the fornix, which are small and registration error would lead to significant measurement inaccuracy. This leads to low statistical power as a quantification tool. If significant differences in FA values are found[S20] between control and patient groups, it could be due to consistent anatomical difference between the two groups and consequent difference in the registration quality. Therefore, the proposed tool should be used as an initial screening and careful inspection and interpretation of the results are required. Employment of a non-linear normalization method is an obvious future direction of this research, which is underway. The atlases developed in this study are now available for downloading from our websites for testing.
In conclusion, we developed a population-based atlas of the SWM using diffusion tensor imaging. The SWM was divided into 9 blade-type structures and their association with cortical areas was described. Based on tractography, several inter-blade tracts were identified. This atlas is expected to be a useful tool to systematically labeling SWM regions.