In recent years, more and more studies have shown that USCT (Ultrasound Computed Tomography) technology has a good application prospect in the early diagnosis of breast cancer. The sound speed image of USCT reveals the sound speed distribution of the tissue. As the sound speeds of normal tissues and tumors have different distributions, the sound speed image can help identify tumors more accurately. The ray-based reconstruction method is commonly used for sound speed image reconstruction. The method mainly includes two processes, forward path matrix calculation and reverse parameter optimization. Due to the huge amount of data of USCT, it is time-consuming to reconstruct the sound speed image on CPU. Considering that both the forward path matrix calculation and the reverse parameter optimization process can be divided into different independent calculation processes, this paper utilizes the parallel computing capabilities of GPU to reconstruct the sound speed image. Multi-thread parallel calculations are performed for the calculation of each emit-receive path, the calculation of the grid gradient, and the construction of the inverse of the Hessian matrix. The experimental data are captured by the USCT system developed by the Medical Ultrasound Laboratory of Huazhong University of Science and Technology for verification. The results show no substantial difference between the sound speed image of female breasts reconstructed on GPU and CPU. When the number of emit array elements are 256, 512, and 1024, the speedup ratio is 47.27, 69.93, and 111.56 respectively.

USCT;,sound speed image,parallel processing,GPU