Towards efficient implementation of MLP-ANN classifier on the FPGA-based embedded system

Abstract

Epilepsy is a widely known neurological disease, causing atypical brain activity such as seizures. Apparently, it is critical to analyse possible triggers as well as establish appropriate medical treatment based on the seizure type. In general, diagnosis of epilepsy is usually conducted by applying various state-of-the-art algorithms to extract useful information from electroencephalography (EEG) signals. For effective real-time epilepsy diagnosis, the biomedical device integrated with the best performing machine learning algorithm is needed. Multilayer Perceptron (feed-forward) Artificial Neural Network (MLP-ANN) has proven to be a much better choice over traditional supervised machine learning algorithms in achieving high accuracy diagnosis. Challenges exist in handling extremely long data processing time in the case of many features in EEG signals plus complex MLP-ANN structure and computations. This research proposes the 5-12-3 MLP-ANN configuration to classify different types of epileptic seizures using Field Programmable Gate Array (FPGA) as a real-time embedded system.