Fei Jiang, PhD, Assistant Professor, University of Hong Kong
The talk comprises two projects that analyze the data from medical devices. In the first project, we propose a Bayesian model selection based (BMS) change point detection algorithm, which identifies the abnormalities that cause radiation dose changes in the Magnetic Resonance quided Radiation Therapy (MRgRT) device. The BMS method effectively reduces the detection errors through using non-local priors. More importantly, the BMS method successfully identifies the abnormality locations, and hence helps controlling the radiation dose level in time.
In the second project, we propose a functional censored quantile regression (FCQR) model to describe the time-varying relationship between the time to stroke recurrence and continuous blood pressure measurement. The FCQR avoids using proportional hazard as-sumption and facilitates the survival time quantile prediction. We combine the B-spline device with a generalized approximate cross-validation method for the nonparametric parameter estimation. The FCQR is applied to study the functinal relationship between the ambulatory blood pressure (ABP) trajectories and the time to stroke recurrences collected from the largest stroke study in China. The results reinforce the effectiveness of using ABP trajectories for stroke prevention.