Real-time seizure detection system using multiple single-neuron recordings

Approximately 20% of people diagnosed with epilepsy cannot be treated effectively. Consequently, there exists a significant need for alternative types of treatment. To aid in the effort of solving this problem, we developed a prototype system to detect changes in neural activity prior to the onset of a seizure. This system can be used as warning device or as part of a large system to terminate seizures in their initial stages via drug administration or nerve stimulation. The detection algorithm used data collected from intracranial electrodes. The waveforms were filtered and amplified to identify single neuron action potentials. The time of occurrence of each action potential for each neuron was then passed to a preprocessor algorithm that summed the data into 50 ms time bins. Sliding windows consisting of 128 bins for each neuron were cross-correlated. The results were summed and the variance of the cross-correlation was used as a measure of global neuron correlation. The algorithm was implemented in a PC board and tested in rats treated with pentylenetetrazol (PTZ) a known seizure inducing drug. The system was 100% effective at detecting seizures approximately 4.6 seconds before seizure onset and had a false positive rate of 0.3%.