Diagnostic value of combined transcranial magnetic stimulation (TMS) and electroencephalography (EEG) in epilepsy.

Abstract

Purpose: The main aim of the project is to estimate the value of combined TMS-EEG responses and EEG to increase the sensitivity and/or specificity of the routine EEG in the diagnosis of newly onset epilepsies. Methods: The project is a combined cross-sectional and longitudinal study involving 60 patients recruited from the First Seizure Clinic at Guy's and St Thomas Hospital NHS Foundation Trust who have had their first presumed epileptic seizure. All the participants had a sleep-deprived EEG (baseline EEG) followed by a combined TMS and EEG study (TMS-EEG). The EEG responses to TMS were visually analysed, looking for two different types of TMS-evoked responses or late responses: The delayed responses were assessed in the unprocessed EEG and the repetitive responses (RRs) after averaging the EEG signals synchronized with the TMS pulse. The late responses were compared between epileptic and non-epileptic patients, looking for responses associated with epilepsy. In patients where the baseline EEG was normal, the additional diagnostic value provided by TMS-EEG was estimated by their ability to predict the final diagnosis based on the clinical history and other tests. A quantitative analysis was performed to compare the power ratio in different frequency bands between epilepsy and no epilepsy cohorts and to select epilepsy-associated variables to generate a machine learning-based classification model for epilepsy prediction. Results: In patients with normal baseline EEG, abnormal TMS-EEG evoked responses (late responses) had no statistically significant association with the presence of epilepsy (Fisher’s exact test, p=0.063), but the late responses correctly classified as epilepsy the 36% of patients with a false-negative baseline EEG. The combined presence of late responses and interictal epileptiform discharges (IEDs) in TMS-EEG records has a higher sensitivity (74%) but lower specificity (85%) than baseline EEG alone. The grand average power-ratio differences between epilepsy and no-epilepsy cohorts were not statistically significant. The epilepsy-associated variables selected for machine learning-based classification were predominantly in the alpha-theta and gamma frequency ranges when TMS activation was present and, in the beta-gamma range with Sham. The TMS support vector machine (SVM)-classifier’s disease prediction over an independent cohort had a sensitivity of 83%. Conclusions: The TMS-EEG significantly increased the sensitivity of the baseline EEG and correctly classified as epilepsy approximately one-third of the patients with a false negative baseline EEG and a final clinical diagnosis of epilepsy. TMS stimulation modified the spectral and topographic properties of the epilepsy-associated variables used for disease detection with machine learning linear regression algorithms. The performance of the TMS SVM-classifier in the training cohort has a high sensitivity, high specificity and low misclassification rate. The TMS SVM-classifier performed better than the Sham as an epilepsy disease prediction model in an independent TMS-EEG cohort. The TMS SVM-classifier has a promising value for disease prediction in TMS-EEG datasets.

Date of Award	1 Feb 2024
Original language	English
Awarding Institution	King's College London
Supervisor	Mark Richardson (Supervisor)