Abstracts

Rationale: In spike correlated EEG-fMRI analysis, EEG recorded simultaneously with fMRI is used to identify interictal spikes, and the spike timing is used to analyze the fMRI data to localize the spikes. However, the EEG data is only the half of the information that can be used to identify spike timings: because fMRI signal changes also reflect spikes, they may be used to aid spike detection. Improved spike detection can improve interictal spike localization, especially in patients who have relatively low amplitude or infrequent spikes. In this research, we present an approach that initially uses the spikes that were manually read from the EEG, and suggests adding or deleting spikes based on the combined EEG and fMRI data. Methods: (1) Algorithm: The software makes use of discrete wavelet transform for characterizing interictal spikes in the EEG, and a multiway partial least squares algorithm for keeping the spike-related EEG and fMRI signal timecourses correlated. Along with the EEG traces, the software displays the likelihood that there is an interictal spike at each EEG sampling time and the fMRI spatial pattern associated with the spikes. The user can add or delete spike timings, and the process is repeated until the user decides that no further changes should be made. The system supports multiple spike morphologies. (2) Testing: Simultaneous EEG-fMRI data from eight localization related epilepsy patients with frequent interictal spikes (ages 17 to 51, average 32.6 years old) were analyzed. The EEG-fMRI data were obtained in a 3T scanner and comprised four runs of six minutes of simultaneous EEG-fMRI acquisitions. The spikes in the EEG traces were initially read by a neurologist or a biomedical engineer experienced in reading MR-contaminated EEG, and standard spike correlated analysis was conducted using the general linear model. Spike reading was modified by the latter individual using information from the interactive spike detection software. To keep the number of spikes the same, one spike was added and one spike was deleted in each run. Finally the standard spike correlated analysis was conducted again and the result was compared with the initial result. Results: Two spike morphologies were found in two out of eight patients, thus ten fMRI activation maps were obtained in total. Out of ten fMRI maps, the activations were increased in nine maps after spike timings were modified with the aid of the interactive spike detection software. Collectively, the t-values of the ten maps were significantly increased (p<0.0005, t=5.87, df=9). The map which did not show a significant increase in strength belonged to a patient who had two spike morphologies. However the activation for one of the morphologies increased. Conclusions: The result shows that the interactive spike detection software can significantly increase the quality of spike detection. By adding data from more patients, we seek to investigate the impact of spike modification on the accuracy of spike localization in interictal EEG-fMRI experiments. This research was supported by NIH R01-NS47605.