Figure 4.

Non-linear FC-to-SC data completion. An iterative procedure can be used to perform resting-state simulations of an MFM model starting from a randomly guessed structural connectome SC* and progressively modify this SC* to make it compatible with a known target FC_emp. A, Starting from an initial random SC*₍₀₎ matrix, there is no correlation between the target FC_emp and the generated FC*₍₀₎ matrix. However, by adjusting the weights of the used SC* through the algorithm of Table 2, SC* gradually develops a richer organization, leading to an increase of the correlation between FC* and FC_emp (violet dashed line) and in parallel, of the correlation between SC* and SC_emp (violet solid line), as shown here for a representative subject within the SC_emp + FC_emp subset. The algorithm stops when the correlation between FC* and the input target FC_emp reaches a desired quality threshold (here 0.7 after 2000 iterations) and the SC* at the last iteration is used as virtual surrogate SC_MFM. B, The boxplot shows the distribution of correlation between SC_emp and SC_MFM for all subjects in the SC_emp + FC_emp ADNI subset and the healthy ageing dataset. C, The correlation between SC_emp and SC_MFM can vary using different random initial connectomes SC*₍₀₎. Here, we show a boxplot of the percent dispersions of the correlation values obtained for different initial conditions around the median correlation value. The fact that these dispersions lie within a narrow interval of ±2.5% indicates that the expected performance is robust against changes of the initial conditions. See Extended Data Figure 4-1 for linear FC-to-SC completion.