Seismic tomography solves high-dimensional optimization problems to image subsurface structures of Earth. In this talk, we propose to use random batch methods to construct the gradient used for iterations in seismic tomography. Specifically, we use the frozen Gaussian approximation to compute seismic wave propagation, and then construct stochastic gradients by random batch methods. We also provide a rigorous convergence analysis for the inverse problems of acoustic wave propagation arising from seismic tomography.