Fetal head biometrics measurements using convolutional neural network and mid-point ellipse drawing algorithm

Abstract

During pregnancy, it is considered to be necessary to monitor and measure fetal development. There are different fetal biometrics among which the fetal head biometrics are useful for finding out the fetus age, diagnosing malformations, and lowering the fetal mortality rate. Accurate measurement of fetal biometry is difficult because of a variety of factors such as the ultrasound image's low quality, inter-and intra-observer variability, and human error as a result of manual estimation. As a result of these issues, new-born babies are born with defects. To track and test fetal head biometrics, several automated and semi-automated methods have been proposed. However, most of the techniques are inaccurate, and the procedure is time-consuming. This work presents the creation of a new process for the segmentation of two-dimensional ultrasound images of fetal skulls based on convolution neural network combination U-Net architecture (UNet-C) and finding the circumference of the fetal head using the Midpoint ellipse drawing algorithm. A new strategy is developed based on U-Net for pre-processing, drop-out, evaluation of different activation layers, activation function, data augmentation, loss function and depth of the network. The computational results reveal the feasibility of the proposal in the correct segmentation of fetal skulls and head circumference measurements and achieved an accuracy of 98.55%. Validation loss and overall loss is calculated to be 0.287 and 0.0390. Validation accuracy and overall accuracy is calculated to be 0.9891 and 0.9840. The mean difference of the proposed method is between − 1.68 and 1.10 and the mean absolute difference is − 1.5 to 0.97. The proposed method is used to detect fetal head and measure circumference of head (HC), bi-parietal diameter (BPD) and occipito frontal diameter (OFD) of a fetus.