A MMSE Framework Combined with Cross Validation to Sonographic Parameters Calibration for Fetal Weight Estimation

Autor: Jia-hao Cao, 曹家豪
Rok vydání: 2011
Druh dokumentu: 學位論文 ; thesis
Popis: 99
Accurately estimating the size of fetus is a important indicator to help clinicians to decide the best mode for babies delivery. Currently most estimation models are based on ultrasound measurement of fetal and maternal physiological parameters. Few of them discussed the relationship between the prenatal and postnatal parameters and estimated fetal weight. In this study, correlation analysis of the differences and associated features between the prenatal and postnatal ultrasound parameters was invested. The parameters were transformed via a discriminative spatial transform model, then a progressive sub-minimum mean square error method is applied for the error compensation. Without changing the original ANN estimation model, the adjusted parameters are able to improve the accuracy of estimated fetal weight. The primarily experiments were conducted based on 50 fetal data collected from an Educational Hospital in Southern Taiwan. The experimental results showed that after compensating the BPD, OFD, AC, FL fetal parameters using proposed MMSE framework, the neural network based fetal weight estimation model achieved a better performance(MAE = 158.53±109.16g) than the original ones(MAE = 178.14±108.55g). The results encourage us that the proposed compensation model is suitable for increasing the correctness of fetal weight estimation.
Databáze: Networked Digital Library of Theses & Dissertations