Regression Into Early Adulthood: A Data-Driven Perspective of NIH Longitudinal Pediatric DTI Study

Yasheng Chen¹, Hongtu Zhu², Hongyu An², Dinggang Shen², Weili Lin²

Most of the current brain developmental studies model growth trajectory with a global parametric model such as nonlinear polynomials. These approaches may neglect subtle local temporal features in the data and the physiological meanings of the derived high order nonlinear polynomial terms may be elusive. To overcome these limitations, we proposed a powerful approach to model brain growth for large-scale longitudinal datasets from NIH pediatric DTI brain developmental study. Through the combination of the greater flexibility of the free-knot B-spline fitting with quasi-least squares longitudinal analysis, we are able to delineate the complex process of brain growth from newborns to early adulthood into a series of linear spans so that growth velocity based physiological inference can be made.