Performance Analysis of Different Learning Algorithms of Feed Forward Neural Network Regarding Fetal Abnormality Detection

Abstract

Ultrasound imaging is one of the safest and most effective method generally used for the diagnosis of fetal growth. The precise assessment of fetal growth at the time of pregnancy is tough task but ultrasound imaging have improved this vital aspect of Obstetrics and Gynecology. In this paper performance of different learning algorithms of Feed forward neural network based on back-propagation algorithm are analyzed and compared. Basically detection of fetal abnormality using neural network is a hybrid method, in which biometric parameters are extracted and measured from segmentation techniques. Then extracted value of biometric parameters are applied on neural network for detect the fetus status. The artificial neural network (ANN) model is applied for the better diagnosis and effective classification purpose. ANN model are design to discriminate normal and abnormal fetus based on the 2-D US images. In this paper, feed forward back-propagation neural network using Levenberg- Marquardt, Bayesian Regularization and Scaled Conjugate Gradient algorithms are analyzed and used for diagnosis and classification of fetal growth. Perfor- mance of these methods are compared and evaluated based on desired output and mean square error. Results found from the Bayesian based neural networks, are in closed confirmation with the real time results. This modeling will help radiologist to take appropriate decision in the boundary line cases.