Periodogram pattern feature-based seizure detection algorithm using optimized hybrid model of MLP and Ant Colony

Seizure detection with EEG analysis in the form of offline is an application of brain signal processing. In this paper, a seizure detection algorithm using pattern features of periodogram and an optimized hybrid model based on Ant Colony Optimization (ACO) and Multi-Layer Perceptron (MLP) neural network is proposed. In this algorithm, with introducing a novel feature extraction method, a feature vector based on periodogram is extracted. For this purpose, we define a novel color map of periodogram pattern based on windowing method and modified periodogram. Statistical and frequency features are computed for each one of five rhythms in EEG signals. These rhythms and features are extracted by using a novel scenario that is called Periodogram Pattern Feature (PPF). In this scenario, at first by making a color pattern of periodogram as an estimation of the spectral density, for each signal, a Time-Frequency layout is created. This pattern is divided to some epochs for decomposing the brain rhythms. Also, to optimize the feature vector, ACO algorithm in combination with MLP is used. At last, a perfect set of features with 5 valid items has provided. Moreover, seizure from non-seizure signal has been classified with accuracy of 94.4% using the final optimized features.