As considered to be less risky, less expensive, and more convenient than radiological examinations, ultrasound has been routinely employed in prenatal exams for the past decades. However, the quality of acquired ultrasound samples, i.e., ultrasound images or videos, and the further diagnosis is crucially depended on the sonographer. At the meantime, there are an extremely limited number of experienced sonographer available for the fetal ultrasound screening. Therefore, to reduce the workload of sonographers, and to promote the quality of fetal ultrasound screening, a deep convolutional neural network based framework is proposed for automatically differentiating five types of fetal head ultrasound standard planes, i.e., Transventricular plane (TV), Transthalamic plane (TT), Transcerebellar plane (TC), Coronal view of eyes (Eyes), Coronal view of nose (Nose), and other non-standard fetal head ultrasound images (Background). A dataset consists of 19928 fetal ultrasound images is applied for the model training and performance evaluation. By combining object detection network, object classification network, and model stacking technique, the proposed framework achieves the state-of-the-art performance with the average accuracy of 89.61% and the average F-1 score of 89.61%.

Ventriculomegaly (VM) is the medical term used to describe enlargement of the lateral ventricles to a level of 10 mm or more, which is the most frequent sign of possible CNS abnormality detected on prenatal ultrasound. In this paper, we aim to evaluate the feasibility of CNN-based DL algorithms predicting the fetal lateral ventricular width from prenatal ultrasound images. The data was collected from 626 pregnant women with gestational age between 22 to 26 weeks. 3456 brain images were picked out from all 49222 stored freeze-frame images. 2304 transventricular (TV) or transthalamic (TT) plane images were further picked out and the brain regions were detected and extracted. 1431 TV-TT planes had known lateral ventricular width. The mean absolute error (MAE) of the predicted lateral ventricular width was 1.01 mm. More than 65% test images had a MAE of less than 1 mm. If we used only the 610 cases with lateral ventricular width less than 15 mm to train and test the model, the MAE was 0.54 mm and more than 82% test images had a MAE of less than 1 mm. We also implemented heat maps to provide evidence that our regression model predicting the lateral ventricular width was based on the anatomical structure of lateral ventricular. The results shown that the regression model can locate the lateral ventricular region of images with large lateral ventricular width successfully and then predict its width based on this region.