Abstracts

Rationale: [Aim 1] We aimed to clarify the causal relationship between cortical resection involving naming-related high gamma activation sites and objectively-measured neuropsychological performance following epilepsy surgery. [Aim 2] We generated a machine learning-based prediction model identifying electrode sites, which, if resected, would lead to a postoperative decline in language function.

Methods: We prospectively recruited 65 patients with drug-resistant focal epilepsy who underwent preoperative neuropsychological assessment and intracranial EEG (iEEG) recording as part of our presurgical evaluation at Detroit Medical Center in Detroit between 2009 and 2019. The Clinical Evaluation of Language Fundamentals (CELF) evaluated the baseline and postoperative language function. We assigned patients to undergo auditory and picture naming tasks. Time-frequency analysis determined the spatiotemporal characteristics of naming-related amplitude modulations, including high gamma augmentation (HGA) at 70-110 Hz. We surgically removed the presumed epileptogenic zone based on the extent of iEEG and MRI abnormalities while maximally preserving the eloquent areas defined by electrical stimulation mapping (ESM). [Aim 1] Multivariate linear regression analysis determined whether HGA-based mapping would predict postoperative language performance independently of epilepsy and neuroimaging data available preoperatively. [Aim 2] We generated the machine learning-based atlas visualizing the sites, which, if resected, would lead to a postoperative decline in CELF-4-based language function. To validate our model, we computed the relative risk of language impairment resulting from virtual resection of ESM-defined language sites compared to that of the others.

Results: [Aim 1] The multivariate regression model incorporating auditory naming-related HGA predicted the postoperative changes in Core Language Score (CLS) on CELF with r² of 0.37 (p = 0.015) and in Expressive Language Index (ELI) with r² of 0.32 (p = 0.047). Independently of the effects of epilepsy and neuroimaging profiles, higher HGA at the resected language-dominant hemispheric area predicted a more severe postoperative decline in CLS (p = 0.004) and ELI (p = 0.012). Conversely, the model incorporating picture naming-related HGA predicted the change in Receptive Language Index (RLI) with r² of 0.50 (p < 0.001). Higher HGA independently predicted a more severe postoperative decline in RLI (p = 0.03). Ancillary regression analysis indicated that naming-related low gamma augmentation as well as alpha/beta attenuation likewise independently predicted a more severe CLS decline. [Aim 2] The machine learning-based prediction model, referred to as the boosted tree ensemble model, suggested that naming-related HGA strongly contributed to the improved prediction of patients showing a >5-point CLS decline (reflecting the lower 25 percentile among patients). We generated the machine learning-based atlas visualizing sites, which, if resected, would lead to such a CLS decline (Figure 1). The auditory naming-based model predicted patients who developed the CLS decline with an accuracy of 0.80 after five-fold cross-validation. The model indicated that virtual resection of an ESM-defined language site, compared to that of sites outside, would have increased the relative risk of the CLS decline by 5.28 (95%CI: 3.47 to 8.02; Figure 2).

Conclusions: Naming-related spectral responses predict objectively-measured neuropsychological outcome after epilepsy surgery. We have provided our prediction model, which will indicate the postoperative language function of future patients.

Funding: Please list any funding that was received in support of this abstract.: NIH grant NS64033.