High-throughput visual assessment of sleep stages in mice using machine learning

Visual detection of seizures in mice using supervised machine learning

Seizures are caused by abnormally synchronous brain activity that can result in changes in muscle tone, such as twitching, stiffness, limpness, or rhythmic jerking. These behavioral manifestations are clear on visual inspection and the most widely used seizure scoring systems in preclinical models, such as the Racine scale in rodents, use these behavioral patterns in semiquantitative seizure intensity scores. However, visual inspection is time-consuming, low-throughput, and partially subjective, and there is a need for rigorously quantitative approaches that are scalable. In this study, we used supervised machine learning approaches to develop automated classifiers to predict seizure severity directly from noninvasive video data. Using the PTZ-induced seizure model in mice, we trained video-only classifiers to predict ictal events, combined these events to predict an univariate seizure intensity for a recording session, as well as time-varying seizure intensity scores. Our results show, for the first time, that seizure events and overall intensity can be rigorously quantified directly from overhead video of mice in a standard open field using supervised approaches. These results enable high-throughput, noninvasive, and standardized seizure scoring for downstream applications such as neurogenetics and therapeutic discovery.

SlumberNet: deep learning classification of sleep stages using residual neural networks

Sleep research is fundamental to understanding health and well-being, as proper sleep is essential for maintaining optimal physiological function. Here we present SlumberNet, a novel deep learning model based on residual network (ResNet) architecture, designed to classify sleep states in mice using electroencephalogram (EEG) and electromyogram (EMG) signals. Our model was trained and tested on data from mice undergoing baseline sleep, sleep deprivation, and recovery sleep, enabling it to handle a wide range of sleep conditions. Employing k-fold cross-validation and data augmentation techniques, SlumberNet achieved high levels of overall performance (accuracy = 97%; F1 score = 96%) in predicting sleep stages and showed robust performance even with a small and diverse training dataset. Comparison of SlumberNet's performance to manual sleep stage classification revealed a significant reduction in analysis time (~ 50 × faster), without sacrificing accuracy. Our study showcases the potential of deep learning to facilitate sleep research by providing a more efficient, accurate, and scalable method for sleep stage classification. Our work with SlumberNet further demonstrates the power of deep learning in mouse sleep research.

Sleep scoring in rodents: Criteria, automatic approaches and outstanding issues

There is nothing we spend as much time on in our lives as we do sleeping, which makes it even more surprising that we currently do not know why we need to sleep. Most of the research addressing this question is performed in rodents to allow for invasive, mechanistic approaches. However, in contrast to human sleep, we currently do not have shared and agreed upon standards on sleep states in rodents. In this article, we present an overview on sleep stages in humans and rodents and a historical perspective on the development of automatic sleep scoring systems in rodents. Further, we highlight specific issues in rodent sleep that also call into question some of the standards used in human sleep research.

Behavioral phenotyping based on physical inactivity can predict sleep in female rats before, during, and after sleep disruption

BACKGROUND

A noninvasive method that can accurately quantify sleep before, during, and after sleep disruption (SD) has not been validated in female rats across their estrous cycle. In female rats, we hypothesized that the duration of physical inactivity (PIA) required to predict sleep would 1) change with the differences in baseline sleep between the circadian and estrous cycle phases and 2) predict sleep and the change in sleep (Δsleep) before, during, and after SD independent of circadian and estrous cycle phase.

NEW METHODS

EEG, EMG, physical activity and estrous cycle phase were measured in female Sprague-Dawley rats before, during, and after SD. Sleep was determined by two methods [EEG/EMG and a duration of continuous PIA (i.e., PIA criterion)]. Reliability between the methods was tested with a previously validated criterion (40 s). Sensitivity analyses and criterion-related validity analyses for sleep during SD and recovery were conducted across multiple PIA criteria (10 s-120 s). Predictability between the two methods and Δsleep was calculated.

RESULTS/COMPARISON WITH EXISTING METHODS

Three criteria (10 s, 20 s, 30 s) predicted baseline sleep independent of circadian and estrous cycle phase. Sleep during SD and recovery were predicted by two criteria (30 s and 10 s). Δsleep between study periods was not reliably predicted by a single PIA criterion.

CONCLUSION

PIA predicted sleep independent of estrous cycle phase in female rats. However, the specific criterion was dependent upon the study period (before, during, and after SD) and circadian phase. Thus, prior work validating a PIA criterion in male rodents is not applicable to the female rat.

Highly Accurate and Precise Determination of Mouse Mass Using Computer Vision

Changes in body mass are a key indicator of health and disease in humans and model organisms. Animal body mass is routinely monitored in husbandry and preclinical studies. In rodent studies, the current best method requires manually weighing the animal on a balance which has at least two consequences. First, direct handling of the animal induces stress and can have confounding effects on studies. Second, the acquired mass is static and not amenable to continuous assessment, and rapid mass changes can be missed. A noninvasive and continuous method of monitoring animal mass would have utility in multiple areas of biomedical research. Here, we test the feasibility of determining mouse body mass using video data. We combine computer vision methods with statistical modeling to demonstrate the feasibility of our approach. Our methods determine mouse mass with 4.8% error across highly genetically diverse mouse strains, with varied coat colors and mass. This error is low enough to replace manual weighing with image-based assessment in most mouse studies. We conclude that visual determination of rodent mass using video enables noninvasive and continuous monitoring and can improve animal welfare and preclinical studies.

A convolutional neural network to characterize mouse hindlimb foot strikes during voluntary wheel running

Voluntary wheel running (VWR) is widely used to study how exercise impacts a variety of physiologies and pathologies in rodents. The primary activity readout of VWR is aggregated wheel turns over a given time interval (most often, days). Given the typical running frequency of mice (∼4 Hz) and the intermittency of voluntary running, aggregate wheel turn counts, therefore, provide minimal insight into the heterogeneity of voluntary activity. To overcome this limitation, we developed a six-layer convolutional neural network (CNN) to determine the hindlimb foot strike frequency of mice exposed to VWR. Aged female C57BL/6 mice (22 months, n = 6) were first exposed to wireless angled running wheels for 2 h/d, 5 days/wk for 3 weeks with all VWR activities recorded at 30 frames/s. To validate the CNN, we manually classified foot strikes within 4800 1-s videos (800 randomly chosen for each mouse) and converted those values to frequency. Upon iterative optimization of model architecture and training on a subset of classified videos (4400), the CNN model achieved an overall training set accuracy of 94%. Once trained, the CNN was validated on the remaining 400 videos (accuracy: 81%). We then applied transfer learning to the CNN to predict the foot strike frequency of young adult female C57BL6 mice (4 months, n = 6) whose activity and gait differed from old mice during VWR (accuracy: 68%). In summary, we have developed a novel quantitative tool that non-invasively characterizes VWR activity at a much greater resolution than was previously accessible. This enhanced resolution holds potential to overcome a primary barrier to relating intermittent and heterogeneous VWR activity to induced physiological responses.

Pupil Dynamics-derived Sleep Stage Classification of a Head-fixed Mouse Using a Recurrent Neural Network

The standard method for sleep state classification is thresholding the amplitudes of electroencephalography (EEG) and electromyography (EMG) data, followed by manual correction by an expert. Although popular, this method has some shortcomings: (1) the time-consuming manual correction by human experts is sometimes a bottleneck hindering sleep studies, (2) EEG electrodes on the skull interfere with wide-field imaging of the cortical activity of a head-fixed mouse under a microscope, (3) invasive surgery to fix the electrodes on the thin mouse skull risks brain tissue injury, and (4) metal electrodes for EEG and EMG recording are difficult to apply to some experimental apparatus such as that for functional magnetic resonance imaging. To overcome these shortcomings, we propose a pupil dynamics-based vigilance state classification method for a head-fixed mouse using a long short-term memory (LSTM) model, a variant of a recurrent neural network, for multi-class labeling of NREM, REM, and WAKE states. For supervisory hypnography, EEG and EMG recording were performed on head-fixed mice. This setup was combined with left eye pupillometry using a USB camera and a markerless tracking toolbox, DeepLabCut. Our open-source LSTM model with feature inputs of pupil diameter, pupil location, pupil velocity, and eyelid opening for 10 s at a 10 Hz sampling rate achieved vigilance state estimation with a higher classification performance (macro F1 score, 0.77; accuracy, 86%) than a feed-forward neural network. Findings from a diverse range of pupillary dynamics implied possible subdivision of the vigilance states defined by EEG and EMG. Pupil dynamics-based hypnography can expand the scope of alternatives for sleep stage scoring of head-fixed mice.

Synaptic dysfunction connects autism spectrum disorder and sleep disturbances: A perspective from studies in model organisms

Sleep disturbances (SD) accompany many neurodevelopmental disorders, suggesting SD is a transdiagnostic process that can account for behavioral deficits and influence underlying neuropathogenesis. Autism Spectrum Disorder (ASD) comprises a complex set of neurodevelopmental conditions characterized by challenges in social interaction, communication, and restricted, repetitive behaviors. Diagnosis of ASD is based primarily on behavioral criteria, and there are no drugs that target core symptoms. Among the co-occurring conditions associated with ASD, SD are one of the most prevalent. SD often arises before the onset of other ASD symptoms. Sleep interventions improve not only sleep but also daytime behaviors in children with ASD. Here, we examine sleep phenotypes in multiple model systems relevant to ASD, e.g., mice, zebrafish, fruit flies and worms. Given the functions of sleep in promoting brain connectivity, neural plasticity, emotional regulation and social behavior, all of which are of critical importance in ASD pathogenesis, we propose that synaptic dysfunction is a major mechanism that connects ASD and SD. Common molecular targets in this interplay that are involved in synaptic function might be a novel avenue for therapy of individuals with ASD experiencing SD. Such therapy would be expected to improve not only sleep but also other ASD symptoms.