Of the three measurement schemes established for diffuse fluorescence tomography (DFT), the time-domain scheme is well known to provide the richest information about the distribution of the targeting fluorophore in living tissues. However, the explicit use of the full time-resolved data usually leads to a considerably lengthy time for image reconstruction, limiting its applications to three-dimensional or small-volume imaging. To cope with the adversity, we propose herein a computationally efficient scheme for DFT image reconstruction where the time-dependent photon density is expanded to a Fourier-series and calculated by solving the independent frequency-domain diffusion equations at multiple sampling frequencies with the support of a combined multicore CPU-based coarse-grain and multithread GPU-based fine-grain parallelization strategy. With such a parallelized Fourier-series truncated diffusion approximation, both the time- and frequency-domain inversion procedures are developed and validated for their effectiveness and accuracy using simulative and phantom experiments. The results show that the proposed method can generate reconstructions comparable to the explicit time-domain scheme, with significantly reduced computational time.