Abstracts

Rationale: Interictal epileptiform discharges (IEDs) are an important biomarker for localization of focal epilepsy, especially in patients that undergo chronic intracranial monitoring. Manual detection of these pathophysiological events is cumbersome, but is still superior to current rule-based approaches in most automated algorithms. We sought to develop an unsupervised machine-learning algorithm for the improved automated detection and localization of IEDs based on spatiotemporal pattern recognition. Methods: We decomposed 24 hours of intracranial EEG signals into basis functions and activation vectors using non-negative matrix factorization (NNMF). Conceptually, NNMF decomposes data into additive constituent representations that can be seen as learned building blocks of the signal. Thresholding the derived activation vector and the basis function of interest detected IEDs in time and space (specific electrodes), respectively. We used convolutive NNMF, a refined algorithm, to add a temporal dimension to basis functions. Results: The receiver operating characteristics of our algorithm are close to the gold standard of human visual-based detection and superior to currently available alternative automated approaches (93% sensitivity and 97% specificity). When inspecting representations used by the NNMF algorithm to detect abnormal activity, we found that it attributed weights to electrodes that reflected their frequency of involvement in IEDs (R2 = 0.90, p < 0.001). Projection of the derived electrode weights onto individual 3D-surface reconstruction accurately summarized the localization of IEDs into a single map. Adding a temporal window to the NNMF bases allowed for visualization of the archetypal propagation network of these epileptiform discharges. This added valuable detection of subtle intricacies of broader epileptic networks including distant propagation. Conclusions: By drawing from hours of recordings at a time and finding the common feature of all detected IEDs across the circadian cycle, this method statistically synthesizes innumerable data and makes it intelligible to the clinician in the form of a single localization map representing the archetypal IED for a specific patient. Unsupervised machine-learning offers a powerful approach towards automated identification of recurrent pathological neurophysiological signals, which may have important implications for precise, quantitative, and individualized evaluation of focal epilepsy. Funding: Funding was provided by the Curci Foundation.