Methods

Twenty six adolescents (15 ± 3 years) who sustained a SRC (3–24 months before imaging) and 24 controls (12 DTI+DSC, 12 DTI only, 15 ± 3 years) were enrolled in the study. Pediatric SRC subjects were referred by a pediatric neurologist and included if they self-reported cognitive, behavioral, or emotional symptoms. Conventional 3D T1 weighted (MPRAGE, repetition time (TR) and echo time (TE) = 1950 msec and 2.26 msec, 1 mm slice thickness, field of view (FOV; 230 x 256 mm2) and DSC-PWI (SE-EPI, TR/TE = 2580/32 ms, flip angle = 90º, 5 mm slice thickness, FOV = 128 x 128, and 50 measures) were acquired using a 3.0T Siemens Tim Trio MR scanner equipped with a 12 channel receive-only head coil. For the DSC-PWI acquisition, gadolinium contrast was administered intravenously (0.1 mmol/kg). Relative CBF maps were generated using Olea Sphere (Olea Medical, Cambridge, MA, USA) using a Bayesian probabilistic estimation algorithm with automatic arterial input function selection. DTI maps (FA, AD, RD, MD) were generated using Camino. Two pediatric templates were created using ANTs (http://stnava.github.io/ANTs/), one from the 3D T1 weighted images and another from the DTI FA images. All T1 and FA images were registered to these common templates using ANTs. Additionally DSC images were registered to the subject’s T1 image using ANTs, the DSC maps DSC maps were transformed into the common T1 template space using the warp files generated by the two registrations steps above. Similarly DTI maps were transformed into the common FA template using the warp file from the FA registration. The OASIS-TRT-20 joint fusion atlas from the Mindboggle dataset⁷ was registered to our T1 pediatric template; standard cortical regions from the atlas were then automatically applied to all subjects DSC maps. The JHU-ICBM white matter atlas was registered to our FA pediatric template; standard DTI tracts were then automatically applied to all subjects DTI maps. Manual regions of interest, corresponding to the same cortical regions and DTI tracts that were automatically defined above, were drawn on unmodified individual DSC and DTI maps.

Results

While generally of very high quality, automatically defined ROIs had some errors on the edge definition of structures, increasing variance across patients and reducing sensitivity. Manual versus automatically defined ROIs showed the same trends (Fig 1 reduced CBF and Fig 2 reduced FA) but more significant results. Automatic and manual ROIs showed hypoperfusion (reduced rCBF) in all regions, with manual ROIs showing significant changes in the occipital and temporal gray matter (Fig 1, p < 0.05) and likely reflect a diffuse reduction of metabolism or neuronal loss. Manual ROIs showed reduced FA in the genu, inferior longitudinal fasciculus (ILF), and splenium (Fig 2). Analysis of splenium showed FA changes due to both increased AD and RD.

Conclusion

While the latest techniques in coregistration and automated image labeling generally show very good results, small errors at the edge of structures increase variance within groups and reduce sensitivity. Automatically defined ROIs showed the same trends as manually drawn ROIs but with less regions reaching significance. In order to maximize sensitivity manually drawing entire anatomic regions still gives the best results, and may be required to detect changes on an individual level. Significant group changes were detected in both DTI and CBF, in particular reduced FA and increased AD, RD and MD in the corpus callosum and ILF, and decreased CBF in the occipital and temporal lobes. These findings may reflect long term changes to the neurovascular unit following concussion, which likely contributes to posttraumatic headaches and long term cognitive and behavior deficits.

Acknowledgements

No acknowledgement found.