Figure 3
Classification in acute stage and relation with behavior: (A) Reversibility was computed by calculating the average of the difference between the time-shifted correlation matrices for the forward and reversed time series. Mean FC was calculated from the mean of the upper triangle of the FC. (B) The classification between controls and patients at the acute stage showed the reversibility in the source space as the highest accuracy (mean = 79%). The right part of the panel shows the comparison between the input and the output signal of the autoencoder. (C) The distinction between stroke patients with low and high lesion volume indicated that the highest accuracy was given by the reversibility in the latent space (mean = 73%). Same description of the autoencoder as in panel (B) was presented in panel (C). (D) 2-dimension projection of the latent representation obtained in the controls’ vs patients’ latent space (left) and the patients with low vs high lesion volume latent space (right). Asterisks represent the mean of each group. (E) All the metrics used for the classification approach were related with each of the 9 behavioral domains recovery values (score after 1 year minus score after 2 weeks, divided the 2 weeks score). Asterisks represent which of the relations were significant. (F) For each behavioral ___domain, the corresponding metric with the highest association was represented indicating the respective color. Red represents the latent space metric while blue represents the source space.
