Sleep Stage Classification With Multi-Modal Fusion and Denoising Diffusion Model.

Sleep stage classification plays a crucial role in sleep quality assessment and sleep disorder prevention. Nowadays, many studies have developed algorithms for this purpose, but they still face two challenges. The first is noise in physiological signals from various devices. The second challenge is that most studies simply concatenate multi-modal features without considering their correlations. To this end, we propose a framework, namely Diff-SleepNet, to efficiently classify sleep stages from multi-modal input. This framework begins with a diffusion model with peak signal-to-noise ratio (PNSR) loss function that adaptively filters noise. The filtered signals are then transformed into a multi-view spectrum through data pre-processing. These spectra are processed by a transformer-based backbone to extract multi-modal features. The production is fed into the following multi-scale attention module for robust feature fusion. The sleep stage category is finally determined by a fully connected layer. Our framework is trained and validated on three typical datasets, i.e., SHHS, Sleep-EDF-SC, and Sleep-EDF-X. Experimental results demonstrate that it is effective and has advantages over other peer methods.

Native language advantage in electrical brain responses to speech sound changes in passive and active listening condition.

It is not clear whether the brain can detect changes in native and non-native speech sounds in both unattended and attended conditions, but this information would be important to understand the nature of potential native language advantage in speech perception. We recorded event-related potentials (ERPs) for changes in duration and in Chinese lexical tone in a repeated vowel /a/ in native speakers of Finnish and Chinese in passive and active listening conditions. ERP amplitudes reflecting deviance detection (mismatch negativity; MMN and N2b) and attentional shifts towards changes in speech sounds (P3a and P3b) were investigated. In the passive listening condition, duration changes elicited increased amplitude in the MMN latency window for both standard and deviant sounds in the Finnish speakers compared to the Chinese speakers, but no group differences were observed for P3a. In passive listening to lexical tones, P3a was increased in amplitude for both standard and deviant stimuli in Chinese speakers compared to Finnish speakers, but the groups did not differ in MMN. In active listening, both tone and duration changes elicited N2b and P3b, but the groups differed only in pattern of results for the deviant type. The results thus suggest an overall increased sensitivity to native speech sounds, especially in passive listening, while the mechanisms of change detection and attentional shifting seem to work well for both native and non-native speech sounds in the attentive mode.

Assessing the effectiveness of spatial PCA on SVM-based decoding of EEG data.

Principal component analysis (PCA) has been widely employed for dimensionality reduction prior to multivariate pattern classification (decoding) in EEG research. The goal of the present study was to provide an evaluation of the effectiveness of PCA on decoding accuracy (using support vector machines) across a broad range of experimental paradigms. We evaluated several different PCA variations, including group-based and subject-based component decomposition and the application of Varimax rotation or no rotation. We also varied the numbers of PCs that were retained for the decoding analysis. We evaluated the resulting decoding accuracy for seven common event-related potential components (N170, mismatch negativity, N2pc, P3b, N400, lateralized readiness potential, and error-related negativity). We also examined more challenging decoding tasks, including decoding of face identity, facial expression, stimulus location, and stimulus orientation. The datasets also varied in the number and density of electrode sites. Our findings indicated that none of the PCA approaches consistently improved decoding performance related to no PCA, and the application of PCA frequently reduced decoding performance. Researchers should therefore be cautious about using PCA prior to decoding EEG data from similar experimental paradigms, populations, and recording setups.

Does artistic training affect color perception? A study of ERPs and EROs in experiencing colors of different brightness.

Color is a visual cue that can convey emotions and attract attention, and there is no doubt that brightness is an important element of color differentiation. To examine the impact of art training on color perception, 44 participants were assigned to two groups-one for those with and one for those without art training-in an EEG experiment. While the participants had their electroencephalographic data recorded, they scored their emotional responses to color stimuli of different brightness levels based on the Munsell color system. The behavioral results revealed that in both groups, high-brightness colors were rated more positively than low-brightness colors. Furthermore, event-related potential results for the artist group showed that high-brightness colors enhanced P2 and P3 amplitudes. Moreover, non-artists had longer N2 latency than artists, and there was a significant Group × Brightness interaction separately for the N2 and P3 components. Simple effect analysis showed that N2 and P3 amplitudes were substantially higher for high-brightness stimuli than for lower-brightness stimuli in the artistic group, but this was not the case in the non-artist group. Additionally, evoked event-related oscillation results showed that in both groups, high-brightness stimuli also elicited large delta, theta, and alpha as well as low gamma responses. These results indicate that high-brightness color stimuli elicit more positive emotions and stronger neurological reactions and that artistic training may have a positive effect on top-down visual perception.

Analysis of modulations of mental fatigue on intra-individual variability from single-trial event related potentials.

BACKGROUND: Intra-individual variability (IIV), a measure of variance within an individual's performance, has been demonstrated as metrics of brain responses for neural functionality. However, how mental fatigue modulates IIV remains unclear. Consequently, the development of robust mental fatigue detection methods at the single-trial level is challenging. NEW METHODS: Based on a long-duration flanker task EEG dataset, the modulations of mental fatigue on IIV were explored in terms of response time (RT) and trial-to-trial latency variations of event-related potentials (ERPs). Specifically, latency variations were quantified using residue iteration decomposition (RIDE) to reconstruct latency-corrected ERPs. We compared reconstructed ERPs with raw ERPs by means of temporal principal component analysis (PCA). Furthermore, a single-trial classification pipeline was developed to detect the changes of mental fatigue levels. RESULTS: We found an increased IIV in the RT metric in the fatigue state compared to the alert state. The same sequence of ERPs (N1, P2, N2, P3a, P3b, and slow wave, or SW) was separated from both raw and reconstructed ERPs using PCA, whereas differences between raw and reconstructed ERPs in explained variances for separated ERPs were found owing to IIV. Particularly, a stronger N2 was detected in the fatigue than alert state after RIDE. The single-trial fatigue detection pipeline yielded an acceptable accuracy of 73.3%. COMPARISON WITH EXISTING METHODS: The IIV has been linked to aging and brain disorders, and as an extension, our finding demonstrates IIV as an efficient indicator of mental fatigue. CONCLUSIONS: This study reveals significant modulations of mental fatigue on IIV at the behavioral and neural levels and establishes a robust mental fatigue detection pipeline.

MIHIC: a multiplex IHC histopathological image classification dataset for lung cancer immune microenvironment quantification.

Background: Immunohistochemistry (IHC) is a widely used laboratory technique for cancer diagnosis, which selectively binds specific antibodies to target proteins in tissue samples and then makes the bound proteins visible through chemical staining. Deep learning approaches have the potential to be employed in quantifying tumor immune micro-environment (TIME) in digitized IHC histological slides. However, it lacks of publicly available IHC datasets explicitly collected for the in-depth TIME analysis. Method: In this paper, a notable Multiplex IHC Histopathological Image Classification (MIHIC) dataset is created based on manual annotations by pathologists, which is publicly available for exploring deep learning models to quantify variables associated with the TIME in lung cancer. The MIHIC dataset comprises of totally 309,698 multiplex IHC stained histological image patches, encompassing seven distinct tissue types: Alveoli, Immune cells, Necrosis, Stroma, Tumor, Other and Background. By using the MIHIC dataset, we conduct a series of experiments that utilize both convolutional neural networks (CNNs) and transformer models to benchmark IHC stained histological image classifications. We finally quantify lung cancer immune microenvironment variables by using the top-performing model on tissue microarray (TMA) cores, which are subsequently used to predict patients' survival outcomes. Result: Experiments show that transformer models tend to provide slightly better performances than CNN models in histological image classifications, although both types of models provide the highest accuracy of 0.811 on the testing dataset in MIHIC. The automatically quantified TIME variables, which reflect proportions of immune cells over stroma and tumor over tissue core, show prognostic value for overall survival of lung cancer patients. Conclusion: To the best of our knowledge, MIHIC is the first publicly available lung cancer IHC histopathological dataset that includes images with 12 different IHC stains, meticulously annotated by multiple pathologists across 7 distinct categories. This dataset holds significant potential for researchers to explore novel techniques for quantifying the TIME and advancing our understanding of the interactions between the immune system and tumors.

Global Field Time-Frequency Representation-Based Discriminative Similarity Analysis of Passive Auditory ERPs for Diagnosis of Disorders of Consciousness.

Behavioural diagnosis of patients with disorders of consciousness (DOC) is challenging and prone to inaccuracies. Consequently, there have been increased efforts to develop bedside assessment based on EEG and event-related potentials (ERPs) that are more sensitive to the neural factors supporting conscious awareness. However, individual detection of residual consciousness using these techniques is less established. Here, we hypothesize that the cross-state similarity (defined as the similarity between healthy and impaired conscious states) of passive brain responses to auditory stimuli can index the level of awareness in individual DOC patients. To this end, we introduce the global field time-frequency representation-based discriminative similarity analysis (GFTFR-DSA). This method quantifies the average cross-state similarity index between an individual patient and our constructed healthy templates using the GFTFR as an EEG feature. We demonstrate that the proposed GFTFR feature exhibits superior within-group consistency in 34 healthy controls over traditional EEG features such as temporal waveforms. Second, we observed the GFTFR-based similarity index was significantly higher in patients with a minimally conscious state (MCS, 40 patients) than those with unresponsive wakefulness syndrome (UWS, 54 patients), supporting our hypothesis. Finally, applying a linear support vector machine classifier for individual MCS/UWS classification, the model achieved a balanced accuracy and F1 score of 0.77. Overall, our findings indicate that combining discriminative and interpretable markers, along with automatic machine learning algorithms, is effective for the differential diagnosis in patients with DOC. Importantly, this approach can, in principle, be transferred into any ERP of interest to better inform DOC diagnoses.

Vision Transformers for Computational Histopathology.

Computational histopathology is focused on the automatic analysis of rich phenotypic information contained in gigabyte whole slide images, aiming at providing cancer patients with more accurate diagnosis, prognosis, and treatment recommendations. Nowadays deep learning is the mainstream methodological choice in computational histopathology. Transformer, as the latest technological advance in deep learning, learns feature representations and global dependencies based on self-attention mechanisms, which is increasingly gaining prevalence in this field. This article presents a comprehensive review of state-of-the-art vision transformers that have been explored in histopathological image analysis for classification, segmentation, and survival risk regression applications. We first overview preliminary concepts and components built into vision transformers. Various recent applications including whole slide image classification, histological tissue component segmentation, and survival outcome prediction with tailored transformer architectures are then discussed. We finally discuss key challenges revolving around the use of vision transformers and envisioned future perspectives. We hope that this review could provide an elaborate guideline for readers to explore vision transformers in computational histopathology, such that more advanced techniques assisting in the precise diagnosis and treatment of cancer patients could be developed.

Predicting pathological complete response based on weakly and semi-supervised joint learning from breast cancer MRI.

Neoadjuvant chemotherapy (NAC) is the standard treatment for breast cancer patients. Patients achieving complete pathological response (pCR) after NAC usually have a good prognosis. However, automatic pCR prediction has been a challenging problem due to lacking well annotations in 3D MRI. Thus far, unifying different annotation information to predict the tumor's early response to NAC has not been systematically addressed. This paper proposes a weakly and semi-supervised joint learning method that integrates attentional features from multi-parametric MRI with radiomic features for predicting pCR to NAC in breast cancer patients. The attention-based multi-instance learning (MIL) is first developed to generate informative MRI bag-level features and mine key instances. The mean-teacher framework is then employed to segment tumor regions in a semi-supervised setting for extracting radiomic features. We perform experiments on 442 patients' data and show that our method achieves an AUC value of 0.85 in pCR prediction, which is superior to comparative methods. It is also shown that learning from multi-parametric MRI outperforms that of single-parameter MRI in pCR prediction.

The applied principles of EEG analysis methods in neuroscience and clinical neurology.

Electroencephalography (EEG) is a non-invasive measurement method for brain activity. Due to its safety, high resolution, and hypersensitivity to dynamic changes in brain neural signals, EEG has aroused much interest in scientific research and medical fields. This article reviews the types of EEG signals, multiple EEG signal analysis methods, and the application of relevant methods in the neuroscience field and for diagnosing neurological diseases. First, three types of EEG signals, including time-invariant EEG, accurate event-related EEG, and random event-related EEG, are introduced. Second, five main directions for the methods of EEG analysis, including power spectrum analysis, time-frequency analysis, connectivity analysis, source localization methods, and machine learning methods, are described in the main section, along with different sub-methods and effect evaluations for solving the same problem. Finally, the application scenarios of different EEG analysis methods are emphasized, and the advantages and disadvantages of similar methods are distinguished. This article is expected to assist researchers in selecting suitable EEG analysis methods based on their research objectives, provide references for subsequent research, and summarize current issues and prospects for the future.

Decreased Functional and Structural Connectivity is Associated with Core Symptom Improvement in Children with Autism Spectrum Disorder After Mini-basketball Training Program.

Exercise intervention has been proven helpful to ameliorate core symptoms of Autism Spectrum Disorder (ASD). However, the underlying mechanisms are not fully understood. In this study, we carried out a 12-week mini-basketball training program (MBTP) on ASD children and examined the changes of brain functional and structural networks before and after exercise intervention. We applied individual-based method to construct functional network and structural morphological network, and investigated their alterations following MBTP as well as their associations with the change in core symptom. Structural MRI and resting-state functional MRI data were obtained from 58 ASD children aged 3-12 years (experiment group: n = 32, control group: n = 26). ASD children who received MBTP intervention showed several distinguishable alternations compared to the control without special intervention. These included decreased functional connectivity within the sensorimotor network (SM) and between SM and the salience network, decreased morphological connectivity strength in a cortical-cortical network centered on the left inferior temporal gyrus, and a subcortical-cortical network centered on the left caudate. Particularly, the aforementioned functional and structural changes induced by MBTP were associated with core symptoms of ASD. Our findings suggested that MBTP intervention could be an effective approach to improve core symptoms in ASD children, decrease connectivity in both structure and function networks, and may drive the brain change towards normal-like neuroanatomy.

Denoising brain networks using a fixed mathematical phase change in independent component analysis of magnitude-only fMRI data.

Brain networks extracted by independent component analysis (ICA) from magnitude-only fMRI data are usually denoised using various amplitude-based thresholds. By contrast, spatial source phase (SSP) or the phase information of ICA brain networks extracted from complex-valued fMRI data, has provided a simple yet effective way to perform the denoising using a fixed phase change. In this work, we extend the approach to magnitude-only fMRI data to avoid testing various amplitude thresholds for denoising magnitude maps extracted by ICA, as most studies do not save the complex-valued data. The main idea is to generate a mathematical SSP map for a magnitude map using a mapping framework, and the mapping framework is built using complex-valued fMRI data with a known SSP map. Here we leverage the fact that the phase map derived from phase fMRI data has similar phase information to the SSP map. After verifying the use of the magnitude data of complex-valued fMRI, this framework is generalized to work with magnitude-only data, allowing use of our approach even without the availability of the corresponding phase fMRI datasets. We test the proposed method using both simulated and experimental fMRI data including complex-valued data from University of New Mexico and magnitude-only data from Human Connectome Project. The results provide evidence that the mathematical SSP denoising with a fixed phase change is effective for denoising spatial maps from magnitude-only fMRI data in terms of retaining more BOLD-related activity and fewer unwanted voxels, compared with amplitude-based thresholding. The proposed method provides a unified and efficient SSP approach to denoise ICA brain networks in fMRI data.

Removal of site effects and enhancement of signal using dual projection independent component analysis for pooling multi-site MRI data.

Combining magnetic resonance imaging (MRI) data from multi-site studies is a popular approach for constructing larger datasets to greatly enhance the reliability and reproducibility of neuroscience research. However, the scanner/site variability is a significant confound that complicates the interpretation of the results, so effective and complete removal of the scanner/site variability is necessary to realise the full advantages of pooling multi-site datasets. Independent component analysis (ICA) and general linear model (GLM) based harmonisation methods are the two primary methods used to eliminate scanner/site effects. Unfortunately, there are challenges with both ICA-based and GLM-based harmonisation methods to remove site effects completely when the signals of interest and scanner/site effects-related variables are correlated, which may occur in neuroscience studies. In this study, we propose an effective and powerful harmonisation strategy that implements dual projection (DP) theory based on ICA to remove the scanner/site effects more completely. This method can separate the signal effects correlated with site variables from the identified site effects for removal without losing signals of interest. Both simulations and vivo structural MRI datasets, including a dataset from Autism Brain Imaging Data Exchange II and a travelling subject dataset from the Strategic Research Program for Brain Sciences, were used to test the performance of a DP-based ICA harmonisation method. Results show that DP-based ICA harmonisation has superior performance for removing site effects and enhancing the sensitivity to detect signals of interest as compared with GLM-based and conventional ICA harmonisation methods.

Harmonization of multi-site functional MRI data with dual-projection based ICA model.

Modern neuroimaging studies frequently merge magnetic resonance imaging (MRI) data from multiple sites. A larger and more diverse group of participants can increase the statistical power, enhance the reliability and reproducibility of neuroimaging research, and obtain findings more representative of the general population. However, measurement biases caused by site differences in scanners represent a barrier when pooling data collected from different sites. The existence of site effects can mask biological effects and lead to spurious findings. We recently proposed a powerful denoising strategy that implements dual-projection (DP) theory based on ICA to remove site-related effects from pooled data, demonstrating the method for simulated and in vivo structural MRI data. This study investigates the use of our DP-based ICA denoising method for harmonizing functional MRI (fMRI) data collected from the Autism Brain Imaging Data Exchange II. After frequency-domain and regional homogeneity analyses, two modalities, including amplitude of low frequency fluctuation (ALFF) and regional homogeneity (ReHo), were used to validate our method. The results indicate that DP-based ICA denoising method removes unwanted site effects for both two fMRI modalities, with increases in the significance of the associations between non-imaging variables (age, sex, etc.) and fMRI measures. In conclusion, our DP method can be applied to fMRI data in multi-site studies, enabling more accurate and reliable neuroimaging research findings.

Hydrogel electrodes with conductive and substrate-adhesive layers for noninvasive long-term EEG acquisition.

Noninvasive brain-computer interfaces (BCIs) show great potential in applications including sleep monitoring, fatigue alerts, neurofeedback training, etc. While noninvasive BCIs do not impose any procedural risk to users (as opposed to invasive BCIs), the acquisition of high-quality electroencephalograms (EEGs) in the long term has been challenging due to the limitations of current electrodes. Herein, we developed a semidry double-layer hydrogel electrode that not only records EEG signals at a resolution comparable to that of wet electrodes but is also able to withstand up to 12 h of continuous EEG acquisition. The electrode comprises dual hydrogel layers: a conductive layer that features high conductivity, low skin-contact impedance, and high robustness; and an adhesive layer that can bond to glass or plastic substrates to reduce motion artifacts in wearing conditions. Water retention in the hydrogel is stable, and the measured skin-contact impedance of the hydrogel electrode is comparable to that of wet electrodes (conductive paste) and drastically lower than that of dry electrodes (metal pin). Cytotoxicity and skin irritation tests show that the hydrogel electrode has excellent biocompatibility. Finally, the developed hydrogel electrode was evaluated in both N170 and P300 event-related potential (ERP) tests on human volunteers. The hydrogel electrode captured the expected ERP waveforms in both the N170 and P300 tests, showing similarities in the waveforms generated by wet electrodes. In contrast, dry electrodes fail to detect the triggered potential due to low signal quality. In addition, our hydrogel electrode can acquire EEG for up to 12 h and is ready for recycled use (7-day tests). Altogether, the results suggest that our semidry double-layer hydrogel electrodes are able to detect ERPs in the long term in an easy-to-use fashion, potentially opening up numerous applications in real-life scenarios for noninvasive BCI.

An efficient memory reserving-and-fading strategy for vector quantization based 3D brain segmentation and tumor extraction using an unsupervised deep learning network.

Deep learning networks are state-of-the-art approaches for 3D brain image segmentation, and the radiological characteristics extracted from tumors are of great significance for clinical diagnosis, treatment planning, and treatment outcome evaluation. However, two problems have been the hindering factors in brain image segmentation techniques. One is that deep learning networks require large amounts of manually annotated data. Another issue is the computational efficiency of 3D deep learning networks. In this study, we propose a vector quantization (VQ)-based 3D segmentation method that employs a novel unsupervised 3D deep embedding clustering (3D-DEC) network and an efficiency memory reserving-and-fading strategy. The VQ-based 3D-DEC network is trained on volume data in an unsupervised manner to avoid manual data annotation. The memory reserving-and-fading strategy beefs up model efficiency greatly. The designed methodology makes deep learning-based model feasible for biomedical image segmentation. The experiment is divided into two parts. First, we extensively evaluate the effectiveness and robustness of the proposed model on two authoritative MRI brain tumor databases (i.e., IBSR and BrainWeb). Second, we validate the model using real 3D brain tumor data collected from our institute for clinical practice significance. Results show that our method (without data manual annotation) has superior accuracy (0.74±0.04 Tanimoto coefficient on IBSR, 97.5% TP and 97.7% TN on BrainWeb, and 91% Dice, 88% sensitivity and 87% specificity on real brain data) and remarkable efficiency (speedup ratio is 18.72 on IBSR, 31.16 on BrainWeb, 31.00 on real brain data) compared to the state-of-the-art methods. The results show that our proposed model can address the lacks of manual annotations, and greatly increase computation speedup with competitive segmentation accuracy compared to other state-of-the-art 3D CNN models. Moreover, the proposed model can be used for tumor treatment follow-ups every 6 months, providing critical details for surgical and postoperative treatment by correctly extracting numerical radiomic features of tumors.

Dynamic Community Detection for Brain Functional Networks During Music Listening With Block Component Analysis.

The human brain can be described as a complex network of functional connections between distinct regions, referred to as the brain functional network. Recent studies show that the functional network is a dynamic process and its community structure evolves with time during continuous task performance. Consequently, it is important for the understanding of the human brain to develop dynamic community detection techniques for such time-varying functional networks. Here, we propose a temporal clustering framework based on a set of network generative models and surprisingly it can be linked to Block Component Analysis to detect and track the latent community structure in dynamic functional networks. Specifically, the temporal dynamic networks are represented within a unified three-way tensor framework for simultaneously capturing multiple types of relationships between a set of entities. The multi-linear rank- (Lr, Lr, 1) block term decomposition (BTD) is adopted to fit the network generative model to directly recover underlying community structures with the specific evolution of time from the temporal networks. We apply the proposed method to the study of the reorganization of the dynamic brain networks from electroencephalography (EEG) data recorded during free music listening. We derive several network structures ( Lr communities in each component) with specific temporal patterns (described by BTD components) significantly modulated by musical features, involving subnetworks of frontoparietal, default mode, and sensory-motor networks. The results show that the brain functional network structures are dynamically reorganized and the derived community structures are temporally modulated by the music features. The proposed generative modeling approach can be an effective tool for describing community structures in brain networks that go beyond static methods and detecting the dynamic reconfiguration of modular connectivity elicited by continuously naturalistic tasks.

Compact Artificial Neural Network Based on Task Attention for Individual SSVEP Recognition With Less Calibration.

OBJECTIVE: Recently, artificial neural networks (ANNs) have been proven effective and promising for the steady-state visual evoked potential (SSVEP) target recognition. Nevertheless, they usually have lots of trainable parameters and thus require a significant amount of calibration data, which becomes a major obstacle due to the costly EEG collection procedures. This paper aims to design a compact network that can avoid the over-fitting of the ANNs in the individual SSVEP recognition. METHOD: This study integrates the prior knowledge of SSVEP recognition tasks into the attention neural network design. First, benefiting from the high model interpretability of the attention mechanism, the attention layer is applied to convert the operations in conventional spatial filtering algorithms to the ANN structure, which reduces network connections between layers. Then, the SSVEP signal models and the common weights shared across stimuli are adopted to design constraints, which further condenses the trainable parameters. RESULTS: A simulation study on two widely-used datasets demonstrates the proposed compact ANN structure with proposed constraints effectively eliminates redundant parameters. Compared to existing prominent deep neural network (DNN)-based and correlation analysis (CA)-based recognition algorithms, the proposed method reduces the trainable parameters by more than 90% and 80% respectively, and boosts the individual recognition performance by at least 57% and 7% respectively. CONCLUSION: Incorporating the prior knowledge of task into the ANN can make it more effective and efficient. The proposed ANN has a compact structure with less trainable parameters and thus requires less calibration with the prominent individual SSVEP recognition performance.

Machine Learning Models for Predicting Adverse Pregnancy Outcomes in Pregnant Women with Systemic Lupus Erythematosus.

Predicting adverse outcomes is essential for pregnant women with systemic lupus erythematosus (SLE) to minimize risks. Applying statistical analysis may be limited for the small sample size of childbearing patients, while the informative medical records could be provided. This study aimed to develop predictive models applying machine learning (ML) techniques to explore more information. We performed a retrospective analysis of 51 pregnant women exhibiting SLE, including 288 variables. After correlation analysis and feature selection, six ML models were applied to the filtered dataset. The efficiency of these overall models was evaluated by the Receiver Operating Characteristic Curve. Meanwhile, real-time models with different timespans based on gestation were also explored. Eighteen variables demonstrated statistical differences between the two groups; more than forty variables were screened out by ML variable selection strategies as contributing predictors, while the overlap of variables were the influential indicators testified by the two selection strategies. The Random Forest (RF) algorithm demonstrated the best discrimination ability under the current dataset for overall predictive models regardless of the data missing rate, while Multi-Layer Perceptron models ranked second. Meanwhile, RF achieved best performance when assessing the real-time predictive accuracy of models. ML models could compensate the limitation of statistical methods when the small sample size problem happens along with numerous variables acquired, while RF classifier performed relatively best when applied to such structured medical records.

[Automated detection of sleep-arousal using multi-scale convolution and self-attention mechanism].

In clinical, manually scoring by technician is the major method for sleep arousal detection. This method is time-consuming and subjective. This study aimed to achieve an end-to-end sleep-arousal events detection by constructing a convolutional neural network based on multi-scale convolutional layers and self-attention mechanism, and using 1 min single-channel electroencephalogram (EEG) signals as its input. Compared with the performance of the baseline model, the results of the proposed method showed that the mean area under the precision-recall curve and area under the receiver operating characteristic were both improved by 7%. Furthermore, we also compared the effects of single modality and multi-modality on the performance of the proposed model. The results revealed the power of single-channel EEG signals in automatic sleep arousal detection. However, the simple combination of multi-modality signals may be counterproductive to the improvement of model performance. Finally, we also explored the scalability of the proposed model and transferred the model into the automated sleep staging task in the same dataset. The average accuracy of 73% also suggested the power of the proposed method in task transferring. This study provides a potential solution for the development of portable sleep monitoring and paves a way for the automatic sleep data analysis using the transfer learning method.