An Artificial Intelligence Model for Automated Measurement of the Facial Maxillary Angle

In prenatal ultrasonography, the Facial Maxillary Angle (FMA) is pivotal in screening for fetal facial abnormalities. Manual measurement of the FMA by experienced sonographers is a time-consuming and labor-intensive process. To address this issue, the present study proposes an automatic measurement method for FMA based on deep learning, utilizing a deep learning model trained on 1400 ultrasound images of fetal mid-sagittal planes. Initially, a semantic segmentation network is employed to acquire the key regions, followed by the use of a keypoint detection network to obtain the coordinates of the required points. Finally, the angle is calculated based on these coordinates. The proposed method has achieved promising results. The automatic measurement method yields an average absolute error of 2.88° and an average relative error of 0.037. This result outperforms the junior expert's, demonstrating good accuracy and potential for assisting sonographers.