Ultrasound Placenta Image Segmentation Based on TransCDnet

The placenta is essential for fetal growth, making its health assessment critical during pregnancy. Ultrasound, despite being the primary diagnostic tool for placental evaluation, presents challenges like noise, low contrast, and artifacts, complicating image segmentation. This paper introduces an advanced TransCDnet model tailored for ultrasound segmentation hurdles, enhancing feature extraction through ResNet, Deformable convolution, and Depthwise separable convolution. These technologies allow dynamic receptive field adjustments and efficient multi-scale detail capture. The decoder leverages depthwise separable convolution, optimizing local and global context processing, thus boosting image detail reconstruction and accurate target segmentation. Tested on a dataset from the Women's and Children's Health Hospital of Wuhan City, the TransCDnet model demonstrated superior placental segmentation performance, achieving a mean_dice of 0.873 and a miou of 0.771. This study not only advances medical image segmentation technology but also significantly contributes to prenatal diagnostics, offering a robust tool for placental health evaluation in clinical settings.