Fetal biometric based abnormality detection during prenatal development using deep learning techniques

Abstract

Early in pregnancy, ultrasounds are used to confirm the fetal heartbeat and a uterine pregnancy. Later, ultrasounds screen for fetal growth, placenta location and umbilical cord, as well as the baby's general health and anatomy. Identifying and interpreting fetal standard scan planes during 2-D ultrasound mid-pregnancy examinations are highly complex tasks, which require years of training. Apart from guiding the probe to the correct location, it is equally difficult for a non-expert to identify relevant structures within the image. The procedure requires a sonographer to find the standardized visualization planes with a probe and manually place measurement calipers on the structures of interest. The process is tedious, time consuming, and introduces user variability into the measurements. Automatic image processing can provide tools to help experienced as well as inexperienced operators with these tasks. The proposed method is realized with deep convolutional neural network models to find the region of interest (ROI) of the fetal biometric and organs region in the US image. Based on the ROI, AlexNet, GoogleNet and CNN evaluate the image quality by assessing the goodness of depiction for the key structures of fetal biometrics. In this method both normal and abnormal US data are considered. In addition with that the input sources of the neural network are augmented with the local phase features along with the original US data. These augmented input sources helps to improve the performance of the various Neural Networks. The input sources are trained by AlexNet, GoogleNet and CNN. Then the process of validation is done by performance in proposed Networks for evaluating the accuracy. The performance of proposed work is evaluated with different network configuration. On the dataset of 400 images used in this classification task, proposed work of AlexNet, GoogleNet and CNN achieves accuracy of 90.43%, 88.70%, and 81.25% with reference to expert’s ground truth results respectively.