Abstract:The area outside the cranial halo in fetal brain ultrasound images contains a lot of irrelevant information and has fuzzy boundaries, which is not conducive to the classification or recognition tasks of ultrasound images. Traditional segmentation methods rely on the subjective judgment of doctors, which is not only time-consuming and labor-intensive but also heavily dependent on the operator's experience, and the results are often unreliable. To address these problems, this paper proposes a deformable Attention U-net network for automatic segmentation of cranial halo in fetal brain ultrasound images. The improved Attention U-net adds a 3x3 kernel deformable convolution to better learn features, and uses a gated unit instead of MaxPooling to filter the convolution feature map, retaining good features and suppressing inappropriate features. The experimental results on the ultrasound fetal head circumference automatic measurement dataset show that compared with the Attention U-net, the proposed method improves IOU 2.9%, DICE by 1.8%, and accuracy by 0.9%, proving the effectiveness of the algorithm.