Hardware-Friendly Random Forest Classification of iEEG Signals for Implantable Seizure Detection

Early and accurate detection of epileptic seizures is an extremely important therapeutic goal due to the severity of complications it can prevent. To this end, a low-power machine learning-based seizure detection implemented on an FPGA is proposed in this paper. Feature extraction is performed using time domain features which exhibit low hardware implementation complexity as well as high classification performance. A comparison between a Random Forest and a linear Support Vector Machine classifier has been conducted leading to the superior performance of the Random Forest. In addition, the hyperparameters of the Random Forest classifier are optimized to reach the best classification performance as well as to maintain the hardware implementation complexity sufficiently low for medical devices implants. The proposed seizure detector is implemented on a Cyclone V FPGA of the ALTERA DE10-standard board and tested on iEEG signals of six patients from the Bern University Hospital. FPGA implementation results demonstrate 100% seizure detection sensitivity as well as better specificity and faster seizure detection compared to recently published works using random forest classification. The FPGA dynamic power consumption is 0.59 mW which is acceptable for low-power implantable devices.