Automatic Measurement of Fetal Head Biometry from Ultrasound Images Using Deep Neural Networks

Gestational age (GA) is an illustrative indicator of fetal growth. The GA is estimated through biometric parameters, including the head circumference (HC) and Biparietal diameter (BPD). This paper proposes a deep learning-based method for automatic measurement of the BPD and HC based on the segmentation of the fetal head from ultrasound images. We utilized an efficient convolutional network architecture, named multi-feature pyramid Unet (MFP-Unet) previously proposed for left ventricle segmentation from echocardiography images. The proposed network tackles the main drawback of U-net, which ignores the contribution of all semantic strengths in the segmentation procedure. To train and evaluate MFP-Unet, a set of ultrasound images was used from the HC18 challenge dataset (fetal head circumference challenge). To evaluate the quantitative accuracy of anatomy segmentation, several metrics, including Dice similarity coefficient (DSC) and Hausdorff distance (HD) were employed. MFP-Unet achieved a DSC of 0.95 and HD of 4.5, which indicates the performance of the MFP-Unet algorithm in segmenting fetal ultrasound images for automatic measurement of biparietal diameter (BPD) and head circumference (HC).