Patient perceptions of physical activity after patent foramen ovale (PFO)-associated stroke and transcatheter closure: A qualitative study.

Bears are an unexplored and interesting model for studying therapeutic hypometabolism and the role of sleep in that adaptation. In the hypometabolic state of hibernation black bears have been found to suppress metabolic rate to 25% of normal basal rates, while their core body temperatures (T) on average decrease only 5.5°C during hibernation. But, their Ts do not stay at a stable level, but vary in multiday body temperature cycles in the range of 30–35°C [,]. Animal studies typically classify sleep into stages of non-REM sleep (NREM), rapid eye movement sleep (REM) and Wake [], with NREM characterized by domination of medium to large amplitude delta waves caused by synchronization of neuronal firing across cortex, REM dominated by low amplitude alpha waves and presence of bipolar EOG spikes and lack of EMG activity, and Wake with irregular high-frequency, low-amplitude EEG patterns and random EMG activity. Studies in marine mammals have also defined a stage of Drowsiness [,]. Human studies additionally classify NREM further into sleep stages N1, N2 and N3 according to the American Academy for Sleep Medicine (AASM) standards [,]. With the exception of certain hibernating dwarf lemur species [], EEG based sleep studies have previously only been published for smaller hibernators that go into torpor through a period of sleep but remain at too low temperatures during deep torpor to record brain activity that can be used for classification of sleep stages. Thus sleep stages can typically only be classified in these hibernators during the episodic arousal episodes when animals return to euthermic levels of T for one or two days [–]. While functionality of the increased deltawave at early arousal caused much controversy among these studies as to its function or even if it can be classified as sleep, in contrast, bears in hibernation do not show torpor-arousal cycles and remain at high enough body temperatures to be responsive to disturbance at any time, and because their brain temperatures do not drop below the low 30 °C [,], their vigilance states can be scored throughout hibernation. Presumed without data, bears are often said to be “asleep” during hibernation, but in this context it refers to the hibernating state (“winter dormancy”). No sleep studies based on brain wave activity existed in a large hibernator before we started collecting a very extensive (>3500 days) repository of polysomnographic data in American black bears () kept in outdoor enclosures in Fairbanks, Alaska. Using telemetry implants we recorded global cortical EEG, EOG, EMG, ECG, T, breathing and respirometry (while dens were closed) and in some of the recordings blood pressure, using telemetry implants. In contrast to small hibernators, there is need to analyze sleep architecture though the whole 5–6 month hibernation period and transitions in and out of the hibernation season to understand role of sleep in the hypometabolism of hibernating bears. Through the years of our study, our polysomnographic recordings produced too much data to score manually, and at the time data were collected reliable methods to automatically score these data in their entirety were not available. A broad range of machine learning techniques have become available for sleep classification more recently, and a number of them have been reviewed [–]. The majority of the studies aimed at verifying machine learning based sleep classification have been performed on human data and standard laboratory animals, e.g., mice and rats. Previous studies have not verified machine learning based sleep classification of EEG recordings from hibernating animals. Thus, before any automatic scoring technique is applied to analyze these data on a larger scale, the integrity of the analysis must be verified. The variations in T of hibernators has the potential to affect the frequency distribution of the EEG signals [], and since machine learning techniques are typically based on a first step of convolution to extract the frequency distributions of the signals [], there could be potential problems with correct detection if the machine learning routines are trained on data recorded at different T‘s than occurred during the analyzed recording. For the present study we selected two different machine learning based automatic sleep qualifiers based on their previous successful verifications on animal data: (1) open source Somnotate, a probabilistic classifier based on a combination of linear discriminant analysis with a hidden Markov model that has previously been tested on mouse data [], and (2) Somnivore, which applies proprietary assisted machine learning algorithms that have been successfully tested on a wide range of human data and animal data from mice, rats and pigeons []. The purpose of the present study is to evaluate the integrity of automated sleep scoring using these two applications on data from American black bears () in and out of hibernation. Due to the dissimilarity in how the t