Neural Modeling of Fetal Biometric Parameters for Detection of Fetal Abnormality

Abstract

This paper presents an approach of neural modeling for the diagnosis of fetus abnormality using ultrasound (US) images. The proposed algorithm is a hybrid approach wherein image processing methods have been used for preprocessing the image data, and an artificial neural network has been used as a classifier to extract fetus abnormality. Initially, 350 US images were collected in DICOM format directly from the radiologist and were preprocessed to extract the fetal biometric parameters using a morphological operator and a gradient vector flow algorithm. The extracted parameters were labeled as normal and abnormal fetal parameters. The extracted parameters were then applied to a Feed-Forward Back-propagation Neural Network (FFNN) for the training and validation purpose. These neural networks are capable to provide excellent performance in the critical cases especially in the field of pattern recognition. The result found from the proposed FFNNs was in closed confirmation with the real-time results. This modeling will help radiologist to take appropriate decisions in the boundary line cases.