Spatial Position Estimation Method for 3D Ultrasound Reconstruction Based on Hybrid Transfomers

3D ultrasound reconstruction has important application value in clinical diagnosis. To obtain the spatial relationships between ultrasound images in reconstruction tasks, especially the contextual information in long-range sequences, this work proposes a spatial position estimation method for 3D ultrasound image reconstruction based on a hybrid transformer encoder. Inspired by the transformer sequence processing capability, we propose a joint local and global information encoding approach to improve the reconstruction accuracy. The proposed method consists of a backbone network for extracting local information of ultrasound images, and a transformer encoder for performing regression tasks on the global sequence. The experiment results indicated that our method can effectively reduce the cumulative error in the reconstruction task, especially in long sequences reconstruction.