Fig. 3: The DCGAN-generated functional connectivity maps are highly similar to the original maps, throughout cortex.
From: Reconstructing lost BOLD signal in individual participants using deep machine learning
The top row shows functional connectivity (FC) maps of seeds from intact BOLD frames in the temporal, lateral frontal, medial frontal, lateral parietal, and occipital cortices. The middle row shows the FC maps of the same seeds extracted from DCGAN-reconstructed BOLD frames. The similarity between the original and reconstructed FC maps is high, with the following correlation coefficients: r = 0.96 for the temporal seed, r = 0.84 for the lateral frontal seed, r = 0.89 for the medial frontal seed, r = 0.84 for the lateral parietal seed, and r = 0.88 for the occipital seed. The bottom row shows diffusion-reconstructed FC maps. While correlations with original FC maps are also high, they are consistently lower than DCGAN correlations: r = 0.84 for the temporal seed, r = 0.80 for the lateral frontal seed, r = 0.42 for the medial frontal seed, r = 0.71 for the lateral parietal seed, and r = 0.83 for the occipital seed.
