Eigen-Seizure for signal core modeling using PCA-RSF and Circular Mapping Distance classifier in automated epileptic Seizure detector

The computerized analysis of the epileptic seizure EEG signal is essential approach for pediatric patients monitoring and presurgical decision making. In this paper, an offline automated epileptic seizure detector algorithm is proposed. After preprocessing stage, a training seizure subset is arranged. This layout is employed a novel hybrid method based on Principle Component Analysis (PCA) and Real Schur Form (RSF) using Implicit QR Iterations. With this iterative algorithm in Eigen-Space, the seizure eigenvectors based on the optimal eigenvalues are represented. These orthonormal basis functions are called Eigen-Seizure for modeling the signal core and expanding the EEG time series in the Seizure-Space. Meanwhile, a novel detector that is called Circular Mapping Distance (CMD) classifier is presented. In this scenario, by projecting the signals into the Seizure-Space, a trigonometric mapping onto unit circle is applied on the projected seizure signals, the projected original and reconstructed test signals. Then, by computing three distance-based similarity measures between static and rotational mapped signals based on the givens rotation technique, a Minkowski 2-D pattern adaptation is applied to find the maximum correlation and to select the optimal weights for signal descriptor. Ultimately, by defining a framework of decision making methodology, the CMD classifier finally recognizes the seizure and non-seizure classes with the averaged accuracy rate of 92.38%, precision rate of 94.74% and false detection rate of 0.022 h
^-1
on different folds of dataset.