Introduction

The human brain undergoes a remarkable development from newborn to adult. There have been many studies that characterized the developmental trajectory of cortex, white matter, and structural/functional connectivity^(1-3). Connectivity-based parcellation of cortex^(4-6) was developed as a data-driven approach to automatically segment sub-regions within cortical area, which provided more detailed information to understand structural and functional characters of the heterogeneous area. However, it is unclear how the parcellation altered during early developed, or different cortical regions experienced different alternation pattern. Therefore, we performed structure connectivity-based parcellation within the primary motor cortex (M1), primary sensory cortex (S1) and insula for subjects in early infancy, late infancy, early toddler, and adult stages.

Methods

Data collection: We recruited 126 normal brains from Baby Connectome Project (BCP) ⁽⁷⁾, including 40 subjects between 0-6 months, 36 subjects between 7-12 months and 50 subjects between 13-24 months, which were collected on 3T Siemens Prisma scanners. T2-weighted images were acquired using a T2wSPC sequence with resolution = 0.8*0.8*0.8 mm³. dMRI data were acquired using a single-shot EPI sequence with resolution = 1.5*1.5*1.5 mm³, 6 b0, 37 diffusion-weighted images at b=1500 and 3000 s/mm², respectively.
Besides, 67 adults (22-31 years old) from S1200 release of Human Connectome Project (HCP) database (https://www.humanconnectome.org/) were included. dMRI data were acquired on 3T Siemens Magnetom scanners using a DW-EPI sequence with resolution = 1.25*1.25*1.25 mm³, 18 b0, and 90 diffusion-weighted images at b=1000, 2000, 3000 s/mm², respectively. Here we selected 6 b0, 37 b1000 s/mm², 37 b3000 s/mm² volumes to match BCP data.
Data processing: The processing pipeline is shown in Figure 1. Firstly, to segment the M1, S1 and insula regions, we performed affine and nonlinear registration between individual T2-weighted images and the DK atlas⁽⁸⁾ (M-CRIB atlas⁽⁹⁾ for BCP data). According to the pipeline proposed by Li et al.⁽¹⁰⁾, we performed probabilistic tractography in FSL⁽¹¹⁾ with each voxel in selected cortical region as seed and weighted the fiber density maps with fractional anisotropy (FA) maps. Connectivity matrix and cross-correlation matrix were then calculated. Subsequently, we used spectral clustering to identify sub-regions within each cortex area based on cross-correlation matrices. The number of clusters was assessed by Dice, cramer V (CV) and normalized mutual information (NMI). Within each sub-region, we performed probabilistic fiber tracking and calculated connectivity pattern based on sum of fiber densities in each region of AAL3 atlas⁽¹²⁾. Besides, we quantified the volume fractions of sub-regions in each cortex and compared them between age groups. Lastly, we calculated difference of volume fractions between each stage and adult to quantify the developmental change.

Results

Based on Dice, CV and NMI shown in Figure 2 and prior parcellation-related studies^{(13, 14)}, we finally chose optimal cluster numbers of four, three, and two for M1, S1, and insula, respectively, for optimal and consistent parcellation across age groups.
The connectivity distributions in Figure 3 showed distinct connectivity patterns for different sub-regions. The connectivity distributions in M1 and S1 experienced gradual change. In insula, connectivity distribution showed little variation through 0-24-month-old but large change from early stage to adult.
Quantitative analysis of the sub-region volume fractions (Figure 4) showed a gradual decrease in sub-region 1 in both left and right M1, and an increase in sub-region 2 in left M1. In S1, only sub-region 3 in right hemisphere showed significant differences from 0-2 years to adult. In insula, significant differences were found in all sub-regions between the BCP data and adult, whereas little significant change were found within the first two years of life.
The absolute deviations between each stage and adult stage were shown in Figure 5. We found considerably higher developmental changes in insula compared to M1 and S1.

Discussion and Conclusion

In this study, we utilized dMRI to investigate structure connectivity-based parcellations of specific cortices and characterized the parcellation-related features during early development. Results revealed that: 1) the sub-regional parcellation of most cortical areas was dynamically changing due to the change of connectivity. Therefore, it is necessary to generate age-specific sub-regional atlases; 2) Parcellation of primary cortices (M1, S1) experienced early development during the first two years of life and became similar to adult, while high-order cortex (insula) experiences ongoing large reshaping of its connectivity after 2 years old. 3) Volume fraction of each sub-region could be a biomarker to assess the development. Insula showed tremendous changes from neonate to adult while M1 and S1 showed steady changes. These results indicated that primary cortex may have developed sufficiently in the neonatal stage while high-order cortex continued to develop from newborn to adult.

Acknowledgements

This work is supported by the National Natural Science Foundation of China (81971606, 82122032), and Science and Technology Department of Zhejiang Province (2022C03057, 202006140)