Estimating the Statistical Distribution of Parameters for Activated Brain in functional PET Images

In many biomedical applications, randomization (or resampling) methods enable us to estimate the probability density function (pdf) of parameters under null hypothesis (H) using limited number of measurements. Null hypothesis stands for the conditions where there is no effect of interest in the measurements (pure noise and disturbances), and H0 is the alternative hypothesis (reverse status). In this paper, a new procedure is developed for estimating the pdf under H1, when boostrap is only applicable under H01. The total pdf of parameter under study is first estimated from real world measurements. Then the pdf under H is estimated using surrogate data which is generated by bootstrap. Making use of these pdf's and total probability law, we derive an estimate of pdf under H10. Whereas the proposed method is applicable to many signal processing and biomedical applications, we demonstrated its validity using simulated data. The method is also applied on a collection of PET functional images, which are processed for localization of brain active areas. The pdf of "activation height" under H is estimated using the proposed method. The resulted estimation of pdf is consistent with the map of activation regions.

کلیدواژه ها:

pdf estimation ، ROC ، Hypothesis testing ، PET image analysis