Supervised and unsupervised machine learning for automated scoring of sleep-wake and cataplexy in a mouse model of narcolepsy

Tetherless miniaturized point detector device for monitoring cortical surface hemodynamics in mice

Significance

Several miniaturized optical neuroimaging devices for preclinical studies mimicking benchtop instrumentation have been proposed in the past. However, they are generally relatively large, complex, and power-hungry, limiting their usability for long-term measurements in freely moving animals. Further, there is limited research in the development of algorithms to analyze long-term signals.

Aim

We aim to develop a cost-effective, easy-to-use miniaturized intrinsic optical monitoring system (TinyIOMS) that can be reliably used to record spontaneous and stimulus-evoked hemodynamic changes and further cluster brain states based on hemodynamic features.

Approach

We present the design and fabrication of TinyIOMS ( 8    mm × 13    mm × 9    mm 3 , 1.2 g with battery). A standard camera-based widefield system (WFIOS) is used to validate the TinyIOMS signals. Next, TinyIOMS is used to continuously record stimulus-evoked activity and spontaneous activity for 7 h in chronically implanted mice. We further show up to 2 days of intermittent recording from an animal. An unsupervised machine learning algorithm is used to analyze the TinyIOMS signals.

Results

We observed that the TinyIOMS data is comparable to the WFIOS data. Stimulus-evoked activity recorded using the TinyIOMS was distinguishable based on stimulus magnitude. Using TinyIOMS, we successfully achieved 7 h of continuous recording and up to 2 days of intermittent recording in its home cage placed in the animal housing facility, i.e., outside a controlled lab environment. Using an unsupervised machine learning algorithm ( k -means clustering), we observed the grouping of data into two clusters representing asleep and awake states with an accuracy of ∼ 91 % . The same algorithm was then applied to the 2-day-long dataset, where similar clusters emerged.

Conclusions

TinyIOMS can be used for long-term hemodynamic monitoring applications in mice. Results indicate that the device is suitable for measurements in freely moving mice during behavioral studies synchronized with behavioral video monitoring and external stimuli.

MLS-Net: An Automatic Sleep Stage Classifier Utilizing Multimodal Physiological Signals in Mice

Over the past decades, feature-based statistical machine learning and deep neural networks have been extensively utilized for automatic sleep stage classification (ASSC). Feature-based approaches offer clear insights into sleep characteristics and require low computational power but often fail to capture the spatial-temporal context of the data. In contrast, deep neural networks can process raw sleep signals directly and deliver superior performance. However, their overfitting, inconsistent accuracy, and computational cost were the primary drawbacks that limited their end-user acceptance. To address these challenges, we developed a novel neural network model, MLS-Net, which integrates the strengths of neural networks and feature extraction for automated sleep staging in mice. MLS-Net leverages temporal and spectral features from multimodal signals, such as EEG, EMG, and eye movements (EMs), as inputs and incorporates a bidirectional Long Short-Term Memory (bi-LSTM) to effectively capture the spatial-temporal nonlinear characteristics inherent in sleep signals. Our studies demonstrate that MLS-Net achieves an overall classification accuracy of 90.4% and REM state precision of 91.1%, sensitivity of 84.7%, and an F1-Score of 87.5% in mice, outperforming other neural network and feature-based algorithms in our multimodal dataset.

Which are best for successful aging prediction? Bagging, boosting, or simple machine learning algorithms?

BACKGROUND

The worldwide society is currently facing an epidemiological shift due to the significant improvement in life expectancy and increase in the elderly population. This shift requires the public and scientific community to highlight successful aging (SA), as an indicator representing the quality of elderly people's health. SA is a subjective, complex, and multidimensional concept; thus, its meaning or measuring is a difficult task. This study seeks to identify the most affecting factors on SA and fed them as input variables for constructing predictive models using machine learning (ML) algorithms.

METHODS

Data from 1465 adults aged ≥ 60 years who were referred to health centers in Abadan city (Iran) between 2021 and 2022 were collected by interview. First, binary logistic regression (BLR) was used to identify the main factors influencing SA. Second, eight ML algorithms, including adaptive boosting (AdaBoost), bootstrap aggregating (Bagging), eXtreme Gradient Boosting (XG-Boost), random forest (RF), J-48, multilayered perceptron (MLP), Naïve Bayes (NB), and support vector machine (SVM), were trained to predict SA. Finally, their performance was evaluated using metrics derived from the confusion matrix to determine the best model.

RESULTS

The experimental results showed that 44 factors had a meaningful relationship with SA as the output class. In total, the RF algorithm with sensitivity = 0.95 ± 0.01, specificity = 0.94 ± 0.01, accuracy = 0.94 ± 0.005, and F-score = 0.94 ± 0.003 yielded the best performance for predicting SA.

CONCLUSIONS

Compared to other selected ML methods, the effectiveness of the RF as a bagging algorithm in predicting SA was significantly better. Our developed prediction models can provide, gerontologists, geriatric nursing, healthcare administrators, and policymakers with a reliable and responsive tool to improve elderly outcomes.

Pharmacological options for narcolepsy: are they the way forward?

INTRODUCTION

Narcolepsy is an under-recognized, rare neurologic disorder of hypersomnolence that is associated with increased mortality and medical and psychiatric co-morbidities. Narcolepsy exerts a substantial economic burden on patients and society. There is currently no cure, and life-long symptomatic therapy is needed. Available drugs do not modify the disease course.

AREAS COVERED

This manuscript provides an overview of narcolepsy symptoms, diagnosis, pathophysiology, current pharmacotherapies, and emerging treatments. Gaps and unresolved issues in diagnosis and management of narcolepsy are discussed to answer whether pharmacological options are the way forward.

EXPERT OPINION

Diagnostic criteria for narcolepsy (ICSD-3) need revision and greater clarity. Improved recognition of cataplexy and other symptoms through educational outreach, new biomarkers, improved test scoring through artificial intelligence algorithms, and use of machine learning may facilitate earlier diagnosis and treatment. Pharmacological options need improved symptomatic therapy in addition to targeted therapies that address the loss of hypocretin signaling. Optimal narcolepsy care also needs a better understanding of the pathophysiology, recognition of the different phenotypes in narcolepsy, identification of at-risk individuals and early recognition of symptoms, better diagnostic tools, and a database for research and disease monitoring of treatment, side-effects, and comorbidities.

Discriminating rapid eye movement sleep from wakefulness by analyzing high frequencies from single-channel EEG recordings in mice

Rapid eye movement sleep (REMS) is characterized by the appearance of fast, desynchronized rhythms in the cortical electroencephalogram (EEG), similar to wakefulness. The low electromyogram (EMG) amplitude during REMS distinguishes it from wakefulness; therefore, recording EMG signal seems to be imperative for discriminating between the two states. The present study evaluated the high frequency components of the EEG signal from mice (80-500 Hz) to support REMS detection during sleep scoring without an EMG signal and found a strong positive correlation between waking and the average power of 80-120 Hz, 120-200 Hz, 200-350 Hz and 350-500 Hz. A highly negative correlation was observed with REMS. Furthermore, our machine learning approach demonstrated that simple EEG time-series features are enough to discriminate REMS from wakefulness with sensitivity of roughly 98 percent and specificity of around 92 percent. Interestingly, assessing only the higher frequency bands (200-350 Hz as well as 350-500 Hz) gives significantly greater predictive power than assessing only the lower end of the EEG frequency spectrum. This paper proposes an approach that can detect subtle changes in REMS reliably, and future unsupervised sleep-scoring approaches could greatly benefit from it.

Unsupervised machine learning effectively clusters pediatric spastic cerebral palsy patients for determination of optimal responders to selective dorsal rhizotomy

Selective dorsal rhizotomy (SDR) can reduce the spasticity in patients with spastic cerebral palsy (SCP) and thus improve the motor function in these patients, but different levels of improvement in motor function were observed among patients after SDR. The aim of the present study was to subgroup patients and to predict the possible outcome of SDR based on the pre-operational parameters. A hundred and thirty-five pediatric patients diagnosed with SCP who underwent SDR from January 2015 to January 2021 were retrospectively reviewed. Spasticity of lower limbs, the number of target muscles, motor functions, and other clinical parameters were used as input variables for unsupervised machine learning to cluster all included patients. The postoperative motor function change is used to assess the clinical significance of clustering. After the SDR procedure, the spasticity of muscles in all patients was reduced significantly, and the motor function was promoted significantly at the follow-up duration. All patients were categorized into three subgroups by both hierarchical and K-means clustering methods. The three subgroups showed significantly different clinical characteristics except for the age at surgery, and the post-operational motor function change at the last follow-up in these three clusters was different. Three subgroups clustered by two methods could be identified as "best responders", "good responders" and "moderate responders" based on the increasement of motor function after SDR. Clustering results achieved by hierarchical and K-means algorithms showed high consistency in subgrouping the whole group of patients. These results indicated that SDR could relieve the spasticity and promote the motor function of patients with SCP. Unsupervised machine learning methods can effectively and accurately cluster patients into different subgroups suffering from SCP based on pre-operative characteristics. Machine learning can be used for the determination of optimal responders for SDR surgery.

Classifier for the Rapid Simultaneous Determination of Sleep-Wake States and Seizures in Mice

Independent automated scoring of sleep-wake and seizures have recently been achieved; however, the combined scoring of both states has yet to be reported. Mouse models of epilepsy typically demonstrate an abnormal electroencephalographic (EEG) background with significant variability between mice, making combined scoring a more difficult classification problem for manual and automated scoring. Given the extensive EEG variability between epileptic mice, large group sizes are needed for most studies. As large datasets are unwieldy and impractical to score manually, automatic seizure and sleep-wake classification are warranted. To this end, we developed an accurate automated classifier of sleep-wake states, seizures, and the post-ictal state. Our benchmark was a classification accuracy at or above the 93% level of human inter-rater agreement. Given the failure of parametric scoring in the setting of altered baseline EEGs, we adopted a machine-learning approach. We created several multi-layer neural network architectures that were trained on human-scored training data from an extensive repository of continuous recordings of electrocorticogram (ECoG), left and right hippocampal local field potential (HPC-L and HPC-R), and electromyogram (EMG) in the murine intra-amygdala kainic acid model of medial temporal lobe epilepsy. We then compared different network models, finding a bidirectional long short-term memory (BiLSTM) design to show the best performance with validation and test portions of the dataset. The SWISC (sleep-wake and the ictal state classifier) achieved >93% scoring accuracy in all categories for epileptic and non-epileptic mice. Classification performance was principally dependent on hippocampal signals and performed well without EMG. Additionally, performance is within desirable limits for recording montages featuring only ECoG channels, expanding its potential scope. This accurate classifier will allow for rapid combined sleep-wake and seizure scoring in mouse models of epilepsy and other neurologic diseases with varying EEG abnormalities, thereby facilitating rigorous experiments with larger numbers of mice.

Use of Machine Learning in Stroke Rehabilitation: A Narrative Review

A narrative review was conducted of machine learning applications and research in the field of stroke rehabilitation. The machine learning models commonly used in medical research include random forest, logistic regression, and deep neural networks. Convolutional neural networks (CNNs), a type of deep neural network, are typically used for image analysis. Machine learning has been used in stroke rehabilitation to predict recovery of motor function using a large amount of clinical data as input. Recent studies on predicting motor function have trained CNN models using magnetic resonance images as input data together with clinical data to increase the accuracy of motor function prediction models. Additionally, a model interpreting videofluoroscopic swallowing studies was developed and investigated. In the future, we anticipate that machine learning will be actively used to treat stroke patients, such as predicting the occurrence of depression and the recovery of language, cognitive, and sensory function, as well as prescribing appropriate rehabilitation treatments.

An attention-based temporal convolutional network for rodent sleep stage classification across species, mutants and experimental environments with single-channel electroencephalogram

Objective.Sleep perturbation by environment, medical procedure and genetic background is under continuous study in biomedical research. Analyzing brain states in animal models such as rodents relies on categorizing electroencephalogram (EEG) recordings. Traditionally, sleep experts have classified these states by visual inspection of EEG signatures, which is laborious. The heterogeneity of sleep patterns complicates the development of a generalizable solution across different species, genotypes and experimental environments.Approach.To realize a generalizable solution, we proposed a cross-species rodent sleep scoring network called CSSleep, a robust deep-learning model based on single-channel EEG. CSSleep starts with a local time-invariant information learning convolutional neural network. The second module is the global transition rules learning temporal convolutional network (TRTCN), stacked with bidirectional attention-based temporal convolutional network modules. The TRTCN simultaneously captures positive and negative time direction information and highlights relevant in-sequence features. The dataset for model evaluation comprises the single-EEG signatures of four cohorts of 16 mice and 8 rats from three laboratories.Main results.In leave-one-cohort-out cross-validation, our model achieved an accuracy of 91.33%. CSSleep performed well on generalization across experimental environments, mutants and rodent species by using single-channel EEG.Significance.This study aims to promote well-standardized cross-laboratory sleep studies to improve our understanding of sleep. Our source codes and supplementary materials will be disclosed later.

Evaluation of feature projection techniques in object grasp classification using electromyogram signals from different limb positions

A myoelectric prosthesis is manipulated using electromyogram (EMG) signals from the existing muscles for performing the activities of daily living. A feature vector that is formed by concatenating data from many EMG channels may result in a high dimensional space, which may cause prolonged computation time, redundancy, and irrelevant information. We evaluated feature projection techniques, namely principal component analysis (PCA), linear discriminant analysis (LDA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and spectral regression extreme learning machine (SRELM), applied to object grasp classification. These represent feature projections that are combinations of either linear or nonlinear, and supervised or unsupervised types. All pairs of the four types of feature projection with seven types of classifiers were evaluated, with data from six EMG channels and an IMU sensors for nine upper limb positions in the transverse plane. The results showed that SRELM outperformed LDA with supervised feature projections, and t-SNE was superior to PCA with unsupervised feature projections. The classification errors from SRELM and t-SNE paired with the seven classifiers were from 1.50% to 2.65% and from 1.27% to 17.15%, respectively. A one-way ANOVA test revealed no statistically significant difference by classifier type when using the SRELM projection, which is a nonlinear supervised feature projection (p = 0.334). On the other hand, we have to carefully select an appropriate classifier for use with t-SNE, which is a nonlinear unsupervised feature projection. We achieved the lowest classification error 1.27% using t-SNE paired with a k-nearest neighbors classifier. For SRELM, the lowest 1.50% classification error was obtained when paired with a neural network classifier.

Application of Machine Learning to Sleep Stage Classification

Sleep studies are imperative to recapitulate phenotypes associated with sleep loss and uncover mechanisms contributing to psychopathology. Most often, investigators manually classify the polysomnography into vigilance states, which is time-consuming, requires extensive training, and is prone to inter-scorer variability. While many works have successfully developed automated vigilance state classifiers based on multiple EEG channels, we aim to produce an automated and openaccess classifier that can reliably predict vigilance state based on a single cortical electroencephalogram (EEG) from rodents to minimize the disadvantages that accompany tethering small animals via wires to computer programs. Approximately 427 hours of continuously monitored EEG, electromyogram (EMG), and activity were labeled by a domain expert out of 571 hours of total data. Here we evaluate the performance of various machine learning techniques on classifying 10-second epochs into one of three discrete classes: paradoxical, slow-wave, or wake. Our investigations include Decision Trees, Random Forests, Naive Bayes Classifiers, Logistic Regression Classifiers, and Artificial Neural Networks. These methodologies have achieved accuracies ranging from approximately 74% to approximately 96%. Most notably, the Random Forest and the ANN achieved remarkable accuracies of 95.78% and 93.31%, respectively. Here we have shown the potential of various machine learning classifiers to automatically, accurately, and reliably classify vigilance states based on a single EEG reading and a single EMG reading.

Predicting Space Radiation Single Ion Exposure in Rodents: A Machine Learning Approach

This study presents a data-driven machine learning approach to predict individual Galactic Cosmic Radiation (GCR) ion exposure for ⁴He, ¹⁶O, ²⁸Si, ⁴⁸Ti, or ⁵⁶Fe up to 150 mGy, based on Attentional Set-shifting (ATSET) experimental tests. The ATSET assay consists of a series of cognitive performance tasks on irradiated male Wistar rats. The GCR ion doses represent the expected cumulative radiation astronauts may receive during a Mars mission on an individual ion basis. The primary objective is to synthesize and assess predictive models on a per-subject level through Machine Learning (ML) classifiers. The raw cognitive performance data from individual rodent subjects are used as features to train the models and to explore the capabilities of three different ML techniques for elucidating a range of correlations between received radiation on rodents and their performance outcomes. The analysis employs scores of selected input features and different normalization approaches which yield varying degrees of model performance. The current study shows that support vector machine, Gaussian naive Bayes, and random forest models are capable of predicting individual ion exposure using ATSET scores where corresponding Matthews correlation coefficients and F₁ scores reflect model performance exceeding random chance. The study suggests a decremental effect on cognitive performance in rodents due to ≤150 mGy of single ion exposure, inasmuch as the models can discriminate between 0 mGy and any exposure level in the performance score feature space. A number of observations about the utility and limitations in specific normalization routines and evaluation scores are examined as well as best practices for ML with imbalanced datasets observed.

An artificial neural network for automated behavioral state classification in rats

Accurate behavioral state classification is critical for many research applications. Researchers typically rely upon manual identification of behavioral state through visual inspection of electrophysiological signals, but this approach is time intensive and subject to low inter-rater reliability. To overcome these limitations, a diverse set of algorithmic approaches have been put forth to automate the classification process. Recently, novel machine learning approaches have been detailed that produce rapid and highly accurate classifications. These approaches however, are often computationally expensive, require significant expertise to implement, and/or require proprietary software that limits broader adoption. Here we detail a novel artificial neural network that uses electrophysiological features to automatically classify behavioral state in rats with high accuracy, sensitivity, and specificity. Common parameters of interest to sleep scientists, including state-dependent power spectra and homeostatic non-REM slow wave activity, did not significantly differ when using this automated classifier as compared to manual scoring. Flexible options enable researchers to further increase classification accuracy through manual rescoring of a small subset of time intervals with low model prediction certainty or further decrease researcher time by generalizing trained networks across multiple recording days. The algorithm is fully open-source and coded within a popular, and freely available, software platform to increase access to this research tool and provide additional flexibility for future researchers. In sum, we have developed a readily implementable, efficient, and effective approach for automated behavioral state classification in rats.

Boosting automated sleep staging performance in big datasets using population subgrouping

Current approaches to automated sleep staging from the electroencephalogram (EEG) rely on constructing a large labeled training and test corpora by aggregating data from different individuals. However, many of the subjects in the training set may exhibit changes in the EEG that are very different from the subjects in the test set. Training an algorithm on such data without accounting for this diversity can cause underperformance. Moreover, test data may have unexpected sensor misplacement or different instrument noise and spectral responses. This work proposes a novel method to learn relevant individuals based on their similarities effectively. The proposed method embeds all training patients into a shared and robust feature space. Individuals who share strong statistical relationships and are similar based on their EEG signals are clustered in this feature space before being passed to a deep learning framework for classification. Using 994 patient EEGs from the 2018 Physionet Challenge (≈6,561 h of recording), we demonstrate that the clustering approach significantly boosts performance compared to state-of-the-art deep learning approaches. The proposed method improves, on average, a precision score from 0.72 to 0.81, a sensitivity score from 0.74 to 0.82, and a Cohen's Kappa coefficient from 0.64 to 0.75 under 10-fold cross-validation.

Automated scoring of pre-REM sleep in mice with deep learning

Reliable automation of the labor-intensive manual task of scoring animal sleep can facilitate the analysis of long-term sleep studies. In recent years, deep-learning-based systems, which learn optimal features from the data, increased scoring accuracies for the classical sleep stages of Wake, REM, and Non-REM. Meanwhile, it has been recognized that the statistics of transitional stages such as pre-REM, found between Non-REM and REM, may hold additional insight into the physiology of sleep and are now under vivid investigation. We propose a classification system based on a simple neural network architecture that scores the classical stages as well as pre-REM sleep in mice. When restricted to the classical stages, the optimized network showed state-of-the-art classification performance with an out-of-sample F1 score of 0.95 in male C57BL/6J mice. When unrestricted, the network showed lower F1 scores on pre-REM (0.5) compared to the classical stages. The result is comparable to previous attempts to score transitional stages in other species such as transition sleep in rats or N1 sleep in humans. Nevertheless, we observed that the sequence of predictions including pre-REM typically transitioned from Non-REM to REM reflecting sleep dynamics observed by human scorers. Our findings provide further evidence for the difficulty of scoring transitional sleep stages, likely because such stages of sleep are under-represented in typical data sets or show large inter-scorer variability. We further provide our source code and an online platform to run predictions with our trained network.