Autonomous neuro Clustering of Pathologic Oscillations Using Discretized Centroids

When an exhaustive search is impractical in non-clinical environs, people tend to use heuristic methods to expedite their search. Conversely, clinicians trust exhaustive searches rather than heuristics such as clinical decision-support systems (CDSS), particularly when electroencephalography (EEG) sequences are used to search for pathologic oscillations in an epileptogenic brain. This paradigm will only change when a CDSS incorporates intelligence that identifies, learns and reprograms itself without human intervention each time it procures a false positive or false negative resultant. In essence, CDSSs need autonomous intelligence. In the ongoing goal to establish an autonomous machinelearning CDSS for detecting seizures the authors present a further embodiment of their existing neuroClustering system that optimizes calculating the limit of rectangular approximations of the EEG above and below a bisecting line using Riemann sums. The authors plot these approximations against the change of time using a novel K-means clustering approach. In this paper we present a methodology that reduces analyzing 7 days of EGG data, something that takes a human expert many days to accomplish, into a 40-minute task that renders the resulting timestamps of pathologic oscillations in an artificial intelligent-ready array.