Epileptic Electrocorticogram Signal Detections with Patient-Specific Quantized Convolution Neural Network Models on the Edge TPU device

The closed-loop systems with neural stimulation and monitoring can be a promising tool of treatment for intractable epilepsies. Machine learning approaches to build patient-specific seizure detection algorisms has been collecting much attention, but implementation of the algorithms in the resource-constrained closed-loop systems has been challenging. In the present, we employed the Edge TPU device recently introduced by Google LLC to implement and test the patient-specific deep convolutional neural network models for the epileptic electrocorticogram signal detections. The present results suggested that the edge computing devices like the one we used here are useful for the real-time seizure detection, at least in some cases of the epilepsy patients.