.png "Decision making module output for early seizure detection")
In this Phase II SBIR, we developed methods for automatic detection of epileptic seizures before and immediately after clinical onset using only features derived from scalp EEG. The detection method is patient-specific, using recurrent neural networks and a variety of input features.
The problem is to develop a highly reliable method to detect seizure onsets - ideally before the onset of clinical symptoms - in time to take effective evasive, diagnostic or therapeutic action.
Our preonset detector was successful on 14 of 25 patients in our database. For these 14 patients, 100% of seizures were detected (0.06 false alarms per hour) with a median preonset time of 51 seconds.
Our onset detector was successful in all 25 patients in our database. 100% of seizures were detected (0.023 false alarms per hour) with a median detection time of 4 seconds after onset.
We continue to expand our test database, and see comparable performance over a variety of seizure types.
Project details:
Supported by NIH/NINDS SBIR grants R44NS039214 and R43NS051881.
Principal Investigator: Georgiy R. Minasyan (gminasyan@chattenassociates.com)