Abstract
This paper describes a classification method is presented using an empirical Rule-base System to detect the occurrences of Partial Seizures from Epilepsy data, which can be implemented in any embedded system as a wearable detection system. The system distinguishes between 'Normal' and 'Seizure' state using on-the-fly calculated features representing the statistical measures for specifically filtered signals from the raw data. It was noticed that for a large number of cases, the seizure waveforms manifest higher energy components during the seizure episodes as compared to the normal brain activity in specific bands of frequencies. Same is also true in the reverse fashion for a separate band of frequency that changes the energy levels from higher to lower when a patient goes from Normal to a Seizure state. This fact has been exploited in this paper and filter has been developed to isolate the seizure band. The rule base has been developed on the calculated measures for the filtered signal from this band-filter and classification is performed on the basis of certain empirical thresholds. Since the complexity of calculations have been deliberately kept quite low, the algorithm is highly suitable for implementation in a small micro-controller environment with near-real-time operation. This gives an enhanced advantage over the existing EEG based seizure detection systems due to their complex pattern classification methodologies. Based on the presented technique, a wearable ubiquitous system can be easily developed with applications in personal healthcare and clinical usage. In this case, the users are not necessarily restricted to the clinical environment in which many devices are connected to the patient externally. The wearable devices allow the user to continue daily activities while being monitored for seizure activities. Once seizure is detected, a number of possible usages can be employed such as alerting the user while driving/holding a baby etc.