Since resting state (rs)-fMRI observed the neuronal activity without external stimuli, various rs-fMRI analysis methods have been proposed. Seed based analysis was initially proposed to assess rs-fMRI connectivity (1,2), and is still used in many studies due to its easy implementation and simple hypothesis.It is known that seed selection at the individual level is preferable to seed selection using common space (e.g. MNI) coordinates(3). However, in studies with a large population, individual seed selection is time-consuming and impractical. In this study, we tested different methods of common seed selection for motor network in a large dataset and demonstrate the feasibility of non-linear registration in seed selection.

Subject & MR: This study uses 149 professional boxers from the professional brain health protocol (PFBHS) data under a protocol approved by the Cleveland Clinic Institutional Review Board. The detailed description of PFHBS is found in the reference(4). Scans were conducted at 3T, and GRE-EPI was used (TR/TE=2.8s/28ms, 30 slices, 132 repetitions) Seed selection: As a baseline, the left M1 seed was selected in MNI (-40,-20, 52) and moved to individual space using linear registration (case 1). In contrast, twenty nine subjects were randomly chosen, and left M1 seed (radius = 6mm) was selected manually using a technique that uses InstaCcor(5) to functionally identify the seed (6) . Twenty nine individual seeds were co-registered to MNI space using linear registration (case2) or a template using non-linear registration (case3 and 4 here) and the most overlapped voxel was chosen as a center of common seed. This common seed with 6 mm radius in a template or MNI is moved to the individual brain using linear (case 2 and 3) and non-linear registration (case 4)(7). Individual CC map from the average signal in the seed was calculated and converted to z-score map, and individual z maps were co-registered to MNI space (case 1, 2, and 3) and a template (case 4). For the demonstration purpose, case 4 was shown in MNI space here. Table 1 summarizes the different registration and analysis space of each case. Group analysis: We have previously reported that motor network connectivity is correlated to the average 10 second finger tapping test score. (8)

Figure 1 visualizes the alignment of manually chosen seeds in MNI space. Twenty nine seeds with 6 mm radius, chosen in individual space are moved to common space using linear and non-linear registration, shown in Fig1. B) and C), respectively. We found that 18 non-linearly registered seeds were overlapped commonly at (-32,-20, 57), while 10 linearly registered seed were overlapped at (-36,-16,-52) among 29 seeds. Figure 2 shows the average motor network of 149 boxers with different seed selections and alignments with z-score, scaled by ±10 (red/blue). Case 4 (non-linear registration) delineates the right M1 anatomical structure best. Group analysis is shown in fig. 3. We found that cases 3 and 4 replicated the previous finding, the significant negative correlation between motor network and left finger tapping score in the left M1 and supplementary motor area (SMA) while case1 and 2 show the correlation only in SMA. The numbers of significant voxels in left M1 are 30 and 64 in case3 and case 4, respectively, again indicating better performance with non-linear registration.

It is uncertain and imprecise that the accurate coordinate of center of primary motor cortex is defined within Broadmann area 4, even for the optimal seed location for rs-fMRI. It is not surprising that the coordinate of M1 seed has been reported differently in literatures.

While seed selection for each individual case is the most desirable, the current study demonstrates that alternative common seed selection methods using non-linear registration works well. This method is not only time-saving but also improves the accuracy of seed selection, and increases the statistical power in group analysis.