IoMT-Based Seizure Detection System Leveraging Edge Machine Learning

As the fast growth of mHealth industry, seizure detection system (SDS) based on electroencephalography (EEG) has become a research hotspot. Cloud-based approach for EEG-based SDS has several shortcomings introduced by the wireless transmission of large amounts of EEG data. In this context, this article proposes an implementation method for EEG-based SDS leveraging edge machine learning. The hardware-friendly seizure detection algorithm (SDA) is presented to meet the requirement of edge devices with limited hardware resources. The feature extractor based on exponential moving average and generalized Hurst exponent (EMA–GHE) is proposed, and the random forest (RF) is employed as a classifier, in which the synthetic minority oversampling technique (SMOTE) is used to address the problem of imbalanced classification. Meanwhile, a field-programmable gate array (FPGA) is used as an edge device, and the software–hardware co-design method is employed to decrease hardware resource utilization. Experimental results show that the proposed method achieves excellent seizure detection performance, with the sensitivity (SE) of 95.2%, specificity (SP) of 99.9%,
${F}1$
score of 0.963, and area under the curve (AUC) of 0.976. The power consumption for the core logic module of the designed SDS is 316 mW with a low response latency of 112 ms. The applicability of designed edge-based SDS in the Internet of Medical Things (IoMT) framework is demonstrated by the constructed validation platform.