With the development of machine learning techniques, more and more classification models have been designed for seizure detection. The creation of these models has dramatically improved the convenience of epilepsy detection and made seizure labeling automation possible. However, many of the current researches in this field use EEG datasets with small data volumes and are mainly designed for scientific purposes, which do not have a good performance of actual medical data. Besides, most models require complex time-frequency domain transformation and feature extraction process, which result in low classification speed and makes it difficult to achieve real-time monitoring. Moreover, the excessive complexity also means higher power consumption, so most of these models cannot be implemented with wearable EEG devices.
This thesis proposed a new seizure detection algorithm based on the bidirectional long short-term memory(BiLSTM) technique. The seizure detection function is achieved using time-domain features and LSTM networks. The preprocessing steps of this model are simple, and the complexity is low. Thus its operation speed is significantly improved compared to other traditional models. Also, this model is developed and tested based on TUH EEG corpus, which is an open-access dataset. Therefore, the results are directly comparable to others in the literature.

Outstanding seizure detection algorithms using electroencephalogram (EEG) recordings have been developed over the past decade. These works mainly focus on best of class performance, which leads to computationally heavy solutions. This limits the applicability of these detection algorithms for hardware implementations such as field­programmable gate arrays (FPGAs). Which in turn limits its use in real­world applications such as warning systems or neural­stimulation systems. In this work, a convolutional neural network (CNN) is trained and its properties reduced to minimize its footprint on hardware. The network is trained based on 1423 EEG recording sessions of 637 different patients sourced from the TUSZ epilepsy database. The input EEG data is processed by removing artifacts, then applying a short­term Fourier transform (STFT) and normalizing and quantizing this to 1 byte values. The neural network is optimized by reducing the input space, this is done by reducing the number of channels, time, and frequencies used. The CNN itself is reduced by quantizing the neural network and reducing its size. Based on these results two ANNs are selected and implemented on an FPGA, one is optimized for accuracy and one for network size. The resulting networks have a sensitivity of 72.26% and 70.00% and an MCC­score of 0.711 and 0.591. The first FPGA implementations consume 0.296 W and 0.562 W at 50000 detections per second.