Epilepsy is one of the most common brain disorders and affect people of all ages.
Resective surgery is currently the most effective overall treatment for patients whose
seizures cannot be controlled by medications. Seizure spread network with secondary
epileptogenesis are thought to be responsible for a substantial portion of surgical
failures. However, there is still considerable risk of surgical failures for lacking
of priori knowledge. Cortico-cortical evoked potentials (CCEP) offer the possibility
of understanding connectivity within seizure spread networks to know how seizure evolves
in the brain as it measures directly the intracranial electric signals. This study
is one of the first works to investigate effective seizure spread network modeling
using CCEP signals. The previous unsupervised brain network connectivity problem was
converted into a classical supervised sparse representation problem for the first
time. In particular, we developed an effective network modeling framework using sparse
representation of over-determined features extracted from extensively designed experiments
to predict real seizure spread network for each individual patient. The experimental
results on five patients achieved prediction accuracy of about 70%, which indicates
that it is possible to predict seizure spread network from stimulated CCEP networks.
The developed CCEP signal analysis and network modeling approaches are promising to
understand network mechanisms of epileptogenesis and have a potential to render clinicians
better epilepsy surgical decisions in the future.