Research on Abdominal Ultrasound Image Segmentation Method Based on Deep Learning

In the medical field, abdominal ultrasound examination can observe abdominal organs and lesions in real time and non-invasively, precisely dividing ultrasonic images is challenging due to low contrast and sound interference. This paper proposes a new network that combines convolution and self-attention mechanisms based on deep learning. It learns global dependencies by introducing a transformer, and uses a cross-channel attention (multi-head cross-attention) mechanism for multi-scale feature fusion. By introducing the transformer, the network can effectively learn the global features of the image and establish global dependencies. Multi-head cross-attention can make full use of the semantic information of the encoder and the high-resolution features of the decoder to achieve more accurate feature fusion and positioning. The method was verified on the public abdominal ultrasound data set. Compared with the U-Net network, the IoU and Dice coefficients were improved, showing a more accurate segmentation effect. This work shows that this method can significantly improve the performance of ultrasound image segmentation and is valuable for ultrasound medical image analysis.