Although seizure detection algorithms are widely used to localize seizure onset on intracranial EEG in epilepsy patients, relatively few studies focus on seizure activity beyond the seizure onset zone to direct treatment of surgical patients with epilepsy. To address this gap, we develop and compare fully automated deep learning algorithms to detect seizure activity on single channels, effectively quantifying spread when deployed across multiple channels. Across 275 seizures in 71 patients, we discover that the extent of seizure spread across the brain and the timing of seizure spread between temporal lobe regions is associated with both surgical outcomes and the brain's structural connectivity between temporal lobes. Finally, we uncover a hierarchical structure of seizure spread patterns highlighting the relationship between clusters of seizures. Collectively, these findings underscore the broad utility in quantifying seizure activity past seizure onset to identify novel mechanisms of seizure evolution and its relationship to potential seizure freedom.