Recent advances in graph theory and its application in brain imaging science (i.e., brain connectomes) have shed tremendous light on organizational principles of brain structure and function[1]. However, research utilizing the technique to study the development of infant brain networks longitudinally is still rare. Such work will reveal new information about the postnatal maturation of brain structure and function and may be served as benchmarks against which the atypical development of brain organizations in infants affected by neurodevelopmental disorders can be compared. 14 typically developing infants (corrected gestational age: 34-211 days, 4 females) were imaged up to 3 times during natural sleep in the first 6 months of life, resulting in 21 scans in total. Diffusion data were collected on a Siemens Trio TIM system with a 32-channel head coil using the multiband technique. Imaging parameters are: MB factor of 2, TR/TE=6200/74ms, FOV=184×184, matrix size of 92×92, diffusion directions of 61 with 6 b=0 images and a b-value of 700. The TR, b-value and MB factors were selected as a trade-off between signal-to-noise-ratio, scan time and minimal table vibration from the scanner. Streamline probabilistic tractography based on the outputs of ‘bedpostx’ in FSL was implemented in Camino, with 50 samples in each voxel of the whole brain, achieving approximately 7 million streamlines in each subject. To reconstruct structural brain networks, we assume that patterns of macro-scale region-to-region connections (i.e., brain connectivity pattern) in the brain of infants at different ages will be largely constant. It is the brain connectivity efficacy (defined as the inverse of the mean radial diffusivity (RD)) along the pathway linking the two regions that changes over time[2]. As a result, we first derived connectional matrices of 90 brain regions (AAL atlas) in 14 typical infant brains and then thresholded the averaged connectivity matrix at a network density of 10% to define brain connectivity pattern in this cohort. The weights (i.e., brain connectivity efficacy) of brain networks in each subject were quantified by the mean of 1/RD. We then examined the most critical brain regions (i.e., hubs), the modular structure, as well as the relationships between graph-theoretic metrics and age for insights into developmental changes in brain networks over time.Streamline probabilistic tractography can reliably track even secondary white matter pathways (i.e., AMG-VMPFC pathway) in infants with the age included in our study (Fig.1A). When ranking the top 20% brain regions based on three centrality metrics[3], we found brain hubs in infants from 0 to 6 months mainly located at subcortical, posterior cingulate, precuneus and occipital regions, whereas such hubs lack in frontal and temporal lobes (Fig.1B)[4]. Infant brain networks have modular structure, which is largely symmetrical and can be divided into one module covering occipital and temporal lobes and another covering frontal, sensorimotor and subcortical regions (Fig.1C). Global efficiency of whole brain networks increases with age(R=0.94, P<1.8e-10), but small-worldness decreases over time, similar to that in the previous work on older infants (Fig.1D)[5]. The topological roles of each brain region in structural brain networks of infants vary with time and have divergent trends: Using betweenness centrality (BC), bilateral thalamus and right pre- and post- central gyri have decreasing centrality with age (Fig.1E), even though the averaged connectivity strength (as measured by 1/RD) in these regions increases as the function of time (Fig.1E right column). In contrast, several cortical regions, such as left posterior cingulate gyrus (CINGpost), left middle frontal gyrus (MFG), right angular gyrus (ANGU) and right inferior occipital lobe (IOC) have increased BC with age, indicating their increasingly important roles in infant brain networks (Fig.1E). Our optimized network analysis framework enables us to probe into the spatial and temporal details of brain networks of infants. Our findings also map onto the sequences of early brain myelination[6] and are consistent with previous infant studies using different methods and/or with older age ranges. In sum, graph-theoretic approach on neuroimaging data may serve as a unique and powerful tool for filing our knowledge gap in early brain development[7]. Such a detailed benchmark of unfolding networks will enable probing of mechanistic hypotheses regarding spatial-temporal disruptions of brain development in early-emerging neurodevelopmental disorders such as autism spectrum disorder[8].