This Phase I project will produce PC-base software capable of detecting first level feature (FLF) of clinical utility, such as spikes, K-complexes, and artifacts , from routine EEG's,long-term monitoring , and sleep studies. The goal following Phase II sill be software capable of identifying third level features (TLF), corresponding to clinical diagnostic features. In Phase I, a panel of clinical experts will define FLF and score recordings from our available database. The database will be split into training and tests sets. EEG time domain and frequency domain parameters from the training set will be calculated and entered into the first layer of an artificial neural network (ANN). Automatic classification of FLF in the test set will be compared to expert scoring. A successful completion of Phase I will be 90 percent correct automatic. During Phase II, FLF will be entered into a second level ANN. The resulting second level features (SLF) will be compared with expert classification of patient states, such as sleep, arousal, and seizures. FLF and SLF will then be combined into TLF using syntactical analysis and adaptive segmentation to match expert clinical classifications. TLF represent clinical diagnostic features, such as a focal lesion, epileptiform, or fragmented sleep.
Not available.
