No abstract

Detection, characterization and classification of patterns within time series from electrophysiological signals have been a challenge for neuroscientists due to their complexity and variability. Here, we aimed to use graph theory to characterize and classify waveforms within biological signals using maxcliques as a feature for a deep learning method. We implemented a compact and easy to visualize algorithm and interface in Python. This software uses time series as input. We applied the maxclique graph operator in order to obtain further graph parameters. We extracted features of the time series by processing all graph parameters through K-means, one of the simplest unsupervised machine learning algorithms. As proof of principle, we analyzed integrated electrical activity of XII nerve to identify waveforms. Our results show that the use of maxcliques allows identification of two distinct types of waveforms that match expert classification. We propose that our method can be a useful tool to characterize and classify other electrophysiological signals in a short time and objectively. Reducing the classification time improves efficiency for further analysis in order to compare between treatments or conditions, e.g., pharmacological trials, injuries, or neurodegenerative diseases.

No abstract

Background: Sleep disruption in elderly has been associated with an increased risk of cognitive impairment and its transition into Alzheimer’s disease (AD). High arousal indices (AIs) during sleep may serve as an early-stage biomarker of cognitive impairment non-dementia (CIND). Objective: Using full-night polysomnography (PSG), we investigated whether CIND is related to different AIs between NREM and REM sleep stages. Methods: Fourteen older adults voluntarily participated in this population-based study that included Mini-Mental State Examination, Neuropsi battery, Katz Index of Independence in Activities of Daily Living, and single-night PSG. Subjects were divided into two groups (n = 7 each) according to their results in Neuropsi memory and attention subtests: cognitively unimpaired (CU), with normal results; and CIND, with –2.5 standard deviations in memory and/or attention subtests. AIs per hour of sleep during N1, N2, N3, and REM stages were obtained and correlated with Neuropsi total score (NTS). Results: AI (REM)  was significantly higher in CU group than in CIND group. For the total sample, a positive correlation between AI (REM)  and NTS was found (r = 0.68, p = 0.006), which remained significant when controlling for the effect of age and education. In CIND group, the AI (N2)  was significantly higher than the AI (REM) . Conclusion: In CIND older adults, this attenuation of normal arousal mechanisms in REM sleep are dissociated from the relative excess of arousals observed in stage N2. We propose as probable etiology an early hypoactivity at the locus coeruleus noradrenergic system, associated to its early pathological damage, present in the AD continuum.