Summary: | 碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 101 === Maternal serum screening is the common laboratory technique used for preliminary prenatal screening of Down’s syndrome. Among the established analysis methods, the likelihood ratio (LR) model proposed by Cuckle et al.(1987) is the most commonly used screening method. The basic assumption underlying the LR model is that the joint distribution of the logarithmic maternal serum data is multivariate normal distribution, however, the logarithmic transformation may be unnecessary and it may reduce model’s prediction capability. In this thesis, we propose several nonlinear adjustment methods for Down’s syndrome screening, which include nonlinear adjustments for linear quadratic discriminant analysis,nonlinear adjustments for quadratic discriminant analysis and nonlinear adjustments for likelihood ratio model underlying the Gamma distribution(NALRG). It is found that the established LR model is a special member of nonlinear adjustments for quadratic discriminant analysis. The simulation results show that all the new methods outperform the established one. In addition, NALRG has much better prediction performance than LR model.
|