Cranial radiotherapy (CRT) is a radiotherapy technique critical to the treatment of primary or metastatic brain tumors, which delivers a prescribed radiation dose uniformly across the whole brain. Monitoring the brain tissue changes caused by CRT during and after treatment is very important to understand the long-term effects induced by CRT among the pediatric brain tumor survivors. Non-invasive diffusion tensor imaging (DTI) not only offers an effective way to character tissue integrity by fractional anisotropy (FA), but also makes it possible to study the brain as a structurally connected network by fiber tractography. In this study, we report the results of a graph theory analysis applied to a structural connectivity network model to investigate the changes induced by CRT among pediatric brain tumor patients.MR scans of 30 medulloblastoma patients (Age at exam 11.2+5.2 years) were used for this study. For each patient, two MR scans were performed, one was Pre-CRT and the other was Post-CRT. Each MR scan consists of anatomic 3D T1 weighted imaging and DTI (30 directions, 2 averages, b=700). For each scan, the anatomic imaging set was processed using Freesurfer [1] to obtain a re-sliced 3D isotropic (256 x 256 x 256, 1 mm resolution) T1 weighted image with 82 anatomic brain structures (7 sub-cortical and 34 cortical for each hemisphere). The eigenvectors and eigenvalues for the DTI data were first calculated using FSL FRMIB Toolbox [2] before being registered to the re-sliced T1 weighted image. To establish a reproducible network graph for each exam, probabilistic fiber tracking was then performed using FSL with 10,000 permutations using the 82 anatomic structures as graph nodes. If the connection between a pair of nodes was non-zero for 80% of all patient’s exams, the connection was considered valid. A total of 311 valid pairs of connections or edges were identified. The connection pathway between two nodes, which was the volume in image space that the connection fibers passed through, was extracted for each valid connection using a previously developed adaptation of the probabilistic fiber tracing technique [3]. The mean FA value of the connection pathway served as the quantitative measure for each edge. The graph network metrics were calculated at the global and nodal levels for each exam using Brain Connectome Toolbox [4] (http://site.google.com/a/brain-connectivity-toolbox.net/bct/Home). For each graph metric, the values of Pre-CRT and Post-CRT were compared across the patients using paired samples T-test. Figure 1 illustrated a typical pathway connecting two brain structures (left caudate and left rotralmiddlefrontal). Figure 2 demonstrated a typical 82 x 82 connection matrix with mean FA values of the connection pathway as the connection strength between nodes. In Table 1, the results of a comparison of Pre-CRT and Post-CRT for two global network metrics (global efficiency and character path) as well as the mean FA values were listed. We found that the two global network metrics were changed significantly in Post-CRT (p=0.007 for global efficiency and p=0.004 for character path) compared to Pre-CRT. Even though the mean FA value was also decreased in Post-CRT, it did not reach statistically significance (p=0.270). At the nodal level, we chose local efficiency and clustering coefficient as our network metrics. Among the 82 nodes, 17 were identified with significantly (p< 0.05) decreased local efficiencies and 14 of those nodes with significantly (p < 0.05) decreased clustering coefficients Post-CRT. These 17 nodes are listed in Table 2. It should be noted that the bilateral basal ganglia (caudate, pallidum and putamen) and thalamus were among those 17 structures.Changes in cerebral white matter due to CRT in pediatric brain tumor patients have been detected using a network model and graph metrics at both the global and regional level. Since the network model in this study combined tissue integrity measures with structural connection information, the network metrics were more sensitive for detecting the tissue response to CRT than the tissue properties alone.