0309 Evaluation of Wake, REM, and NREM State Classifiers using Noninvasive Piezoelectric Sensors for Rodents

Abstract Introduction Assessment of sleep states in rodents by EEG and EMG requires significant resources, limiting sleep research, especially when the study requires many animals. While sleep in mammals is typically defined by brain state, it also affects other physiological states including breathing and other movements which can be measured noninvasively. We have previously demonstrated the efficacy of piezoelectric sensors coupled with appropriate hardware and software to automatically classify sleep with accuracy comparable to EEG/EMG. Methods In this study, we extend our piezoelectric system, coupled with passive infrared (IR) sensors, to classify wake, REM, and NREM. EEG/EMG from 24 mice were acquired simultaneously in a PiezoSleep cage (www.sigsoln.com), and an 8x8 pixel IR sensor mounted above the cage. The EEG/EMG data were scored in 4s intervals over 24 hours, and used to train and test several classifiers based on features extracted from the piezoelectric and IR sensor signals. The features assessed the regularity of the respiratory signal, energy levels, and body movement. The features were applied to linear and decision-tree classifiers for training and testing. Results All classifiers using only features from the piezoelectric signals achieved over a 90% agreement with human scored EEG/EMG for a 2-state sleep-wake classification. For a 3-state classification (wake, NREM, REM) the random forest decision tree classifiers performed best. Using only piezo features, a recall of 95, 95, and 46% was achieved for wake, NREM, and REM states respectively, with a precision of 95, 90, and 78%. When features from the IR sensor were also included, the recall increased to 96, 96, and 53% with a precision of 96, 91, and 87%. Conclusion Results suggest the utility of piezo sensors for classifying REM, NREM and wake noninvasively in rodents, especially with on-going improvements in hardware and software that should further improve performance. Support (If Any) NIH/NINDS Grant R44NS083218 and Grant Agreement KSTC-184-512-13-158 from the Kentucky Cabinet for Economic Development.