Epileptic seizure detection - an AR model based algorithm for implantable device

The algorithm of epileptic seizure is at the core of any implantable device aimed to treat the symptoms of this disorder. A training free (on line) epileptic seizure detection algorithm for implantable device utilizing Autoregressive (AR) model parameters is developed and studied. Pre-recorded (off line) epileptic seizure data are used to estimate the internal parameters of an AR model prior and following the seizure Principle Component Analysis (PCA) is used for reducing the dimension of the problem while allowing only the salient features representing the seizure onset to be saved into the implantable device. The implantable device estimates the AR model parameter in real time and compares the saved features of seizure onset with feature from the incoming signals using cosine similarity. In order to guarantee an efficient on line signal processing, Weighted Least Square Estimation (WLSE) model is utilized. Simulation result shows that the proposed method has average 96.6% detection accuracy and 1.2ms latency for the data sets under study. The proposed approach can be extended to multi channel approach using Multi-Variant Autoregressive (MVAR) model which enables seizure foci localization and the sophisticated seizure prediction.