In this brief, we have proposed a design strategy for an energy-efficient circuit/architecture to detect the onset of epileptic seizures with high efficacy. The architecture consists of two stages. The first stage is a low complexity Coastline parameter algorithm that consumes very low energy per computation. The second stage is a more efficacious wavelet-based algorithm (discrete wavelet transform-quasi-averaging) that consumes relatively higher energy and is powered ON only if determined by the low-complexity first stage. Using this proposed strategy, we achieve significant reduction in the energy consumption of the circuit by avoiding redundant computations, thereby increasing the longevity of the battery.
We also observe that it leads to an improvement in efficacy. The two algorithms are user-programmable to compensate for the intersubject variations of neural signals. We use in vivo neural recordings from large animals (rats) to test the functionality of the system and calculate efficacy, subjected to minimum delay in detection. The system is simulated using 65-nm bulk-Si technology library. The simulated results show 32% energy savings (compared with a single-stage wavelet-based algorithm), consuming an average of 31.2 nJ/computation. The results also show a 12% increase in efficacy.