Purpose/Significance To propose an automated ultrasound image analysis method to assist in determining the stability of carotid artery plaques and improve evaluation efficiency and accuracy. Method/Process A two-stage calcified tissue segmentation model based on carotid plaque ultrasound image features, named TRDU-Net, is proposed. The model uses U-Net as the backbone network, incorporating dual residual convolution structures in both the encoding and decoding paths. Additionally, a dilated convolution bridge is introduced in the network’s bottleneck to enhance the model’s focus on specific regions, the positional information is extracted accurately, and segmentation precision is improved. Result/Conclusion The TRDU-Net model achieves a Dice coefficient of 75.32% and an IoU coefficient of 62.83% on the test dataset. It demonstrates superior segmentation performence and accuracy compared to other methods, validating its effectiveness in the task of carotid plaque calcified tissue segmentation.