This is a revised Phase II application, which proposes to develop a microcomputer-based EEG epilepsy monitoring system for spike and seizure detection using artificial neural networks. During Phase I, the applicants investigated the performance of three different network methods. These included a standard Back-Propagation network, a Time-Delay network, and an Adaptive Resonance Theory network. All were shown to have promise in comparison to currently-available rule-based methods, and a number of areas for improvement were identified. In addition, a modular/dynamic neural network was constructed to address the temporal relationships between spikes and surrounding EEG activity, and the temporal organization within spike trains. The work proposed for Phase II is divided into 4 sub-projects. The neural network methods will continue to be refined under Subproject 1, and implemented in real time using dedicated digital signal processing boards in Subproject 2. Subproject 3 will develop a user interface that will allow the identified segments to be displayed, and will allow operator input into the network training process.The performance of the system will be tested and compared with EEG expert scorings in Subproject 4.
| Ozdamar, O; Kalayci, T (1998) Detection of spikes with artificial neural networks using raw EEG. Comput Biomed Res 31:122-42 |
