Direction-guided and multi-scale feature screening for fetal head–pubic symphysis segmentation and angle of progression calculation

Abstract

Transperineal ultrasound imaging in the mid-sagittal plane can potentially enable the objective quantification of the level of fetal head descent in the birth canal by measuring the angle of progression (AoP). Specifically, the AoP is defined as the angle between a straight line drawn along the longitudinal axis of the pubic symphysis and a line drawn from the inferior edge of the pubic symphysis to the leading edge of the fetal head. However, the process of outlining contours and measuring AoP on ultrasound images is complex and tedious, often leading to misjudgments by physicians. In response to this challenge, we propose the fetal head–pubic symphysis segmentation network (FH-PSSNet) for automatic AoP measurement. The FH-PSSNet model is based on an encoder–decoder framework, incorporating a dual attention module, a multi-scale feature screening module and a direction guidance block. The encoder extracts the relevant features and learns the global feature correlation through a dual attention module, while the decoder utilizes skip connections to preserve spatial information. Additionally, a multi-scale feature screening module with attention mechanism is proposed to extract global contextual information, enhancing object localization. Within the decoder, a dual-branch structure generates both direction information and initial segmentation features. Finally, a direction guidance block corrects the initial segmentation features to refine the segmentation results. Experiments on two datasets demonstrate the effectiveness of the proposed method. Compared to existing approaches, our model showcases competitive performance. The FH-PSSNet significantly improves automatic segmentation and AoP measurement, reducing errors and aiding sonographers in clinical settings.