Figure 4
From: Optimization of epilepsy surgery through virtual resections on individual structural brain networks
Comparison of various resection strategies for all patients as measured by the normalized EC difference and the normalized difference in I(t0), respectively in panels (A) and (B). (A) The normalized EC difference for a resection is defined as the ratio between the EC difference for the optimal resection and the EC difference for a full resection. Removing all connections resulted in the largest EC difference (100%, black stars). A 10% decrease in the EC difference was accepted (red stars), using simulated annealing to optimize which connections to remove, and thereby sparing connections. This method performed better than the average of 100 random resections (green cross with standard deviation), using the same number of removed connections chosen randomly from the candidate connections. For comparison, the same number of removed connections were also chosen using network measures: edge BC (blue triangles), EC (blue diamonds), degree (blue circles), and BC (blue squares). The three latter measures were based on the property of the connected node outside the SOZ. (B) The normalized I(t0) difference for a resection is defined as the difference between I(t0) before and after the resection, normalized by this difference for a full resection. Similarly to the EC difference, removing all connections (black stars) resulted in 100% decrease of I(t0). The optimal resection given by the surrogate model (red stars) performed marginally better than the resections based on network metrics (blue markers) for some patients. Random resections (green markers) performed the worst. All values correspond to the same resections used in panel A. The error bars were calculated as in Fig. 3. The not seizure-free patients are marked in red in both panels.
