Fig. 5: The DCGAN model outperforms a diffusion-based model.
From: Reconstructing lost BOLD signal in individual participants using deep machine learning
Box-and-whisker plots are shown, with the center line indicating the median, box limits indicating upper and lower quartiles, whiskers indicating 1.5 times the interquartile range, and plus signs indicating outliers. DCGAN-reconstructed time series is more similar to the original ones than the diffusion-reconstructed time series, in both the GSP data set (t(19) = 9.15, p < 0.001) and the HCP data set (t(19) = 176.20, p < 0.001). The DCGAN model also outperforms the diffusion model when reconstructing functional connectivity (FC) maps, in both the GSP (t(19) = 27.62, p < 0.001) and HCP (t(19) = 53.50, p < 0.001) data sets. All statistical tests were two-sided and corrected for multiple comparisons. Source data are provided as a Source Data file. *p ≤ 0.001.
