Subject-dependent Model Research Articles

Acquisition and Analysis of Facial Electromyographic Signals for Emotion Recognition.

The objective of the article is to recognize users' emotions by classifying facial electromyographic (EMG) signals. A biomedical signal amplifier, equipped with eight active electrodes positioned in accordance with the Facial Action Coding System, was used to record the EMG signals. These signals were registered during a procedure where users acted out various emotions: joy, sadness, surprise, disgust, anger, fear, and neutral. Recordings were made for 16 users. The mean power of the EMG signals formed the feature set. We utilized these features to train and evaluate various classifiers. In the subject-dependent model, the average classification accuracies were 96.3% for KNN, 94.9% for SVM with a linear kernel, 94.6% for SVM with a cubic kernel, and 93.8% for LDA. In the subject-independent model, the classification results varied depending on the tested user, ranging from 91.4% to 48.6% for the KNN classifier, with an average accuracy of 67.5%. The SVM with a cubic kernel performed slightly worse, achieving an average accuracy of 59.1%, followed by the SVM with a linear kernel at 53.9%, and the LDA classifier at 41.2%. Additionally, the study identified the most effective electrodes for distinguishing between pairs of emotions.

Transfer Learning for Efficient Intent Prediction in Lower-Limb Prosthetics: A Strategy for Limited Datasets.

This paper presents a transfer learning method to enhance locomotion intent prediction in novel transfemoral amputee subjects, particularly in data-sparse scenarios. Transfer learning is done with three pre-trained models trained on separate datasets: transfemoral amputees, able-bodied individuals, and a mixed dataset of both groups. Each model is subsequently fine-tuned using data from a new transfemoral amputee subject. While subject-dependent models, trained and tested using individual user data, can achieve the least error rate, they require extensive training datasets. In contrast, our transfer learning approach yields comparable error rates while requiring significantly less data. This highlights the benefit of using preexisting, pre-trained features when data is scarce. As anticipated, the performance of transfer learning improves as more data from the subject is made available. We also explore the performance of the intent prediction system under various sensor configurations. We identify that a combination of a thigh inertial measurement unit and load cell offers a practical and efficient choice for sensor setup. These findings underscore the potential of transfer learning as a powerful tool for enhancing intent prediction accuracy for new transfemoral amputee subjects, even under data-limited conditions.

Prediction of instantaneous perceived effort during outdoor running using accelerometry and machine learning.

The rate of perceived effort (RPE) is a subjective scale widely used for defining training loads. However, the subjective nature of the metric might lead to an inaccurate representation of the imposed metabolic/mechanical exercise demands. Therefore, this study aimed to predict the rate of perceived exertions during running using biomechanical parameters extracted from a commercially available running smartwatch. Forty-three recreational runners performed a simulated 5-km race on a track, providing their RPE from a Borg scale (6-20) every 400 m. Running distance, heart rate, foot contact time, cadence, stride length, and vertical oscillation were extracted from a running smartwatch (Garmin 735XT). Machine learning regression models were trained to predict the RPE at every 5 s of the 5-km race using subject-independent (leave-one-out), as well as a subject-dependent regression method. The subject-dependent method was tested using 5%, 10%, or 20% of the runner's data in the training set while using the remaining data for testing. The average root-mean-square error (RMSE) in predicting the RPE using the subject-independent method was 1.8 ± 0.8 RPE points (range 0.6-4.1; relative RMSE ~ 12 ± 6%) across the entire 5-km race. However, the error from subject-dependent models was reduced to 1.00 ± 0.31, 0.66 ± 0.20 and 0.45 ± 0.13 RPE points when using 5%, 10%, and 20% of data for training, respectively (average relative RMSE < 7%). All types of predictions underestimated the maximal RPE in ~ 1 RPE point. These results suggest that the data accessible from commercial smartwatches can be used to predict perceived exertion, opening new venues to improve training workload monitoring.

Unobtrusive Refractive Power Monitoring: Using EOG to Detect Blurred Vision.

The rise in population and aging has led to a significant increase in the number of individuals affected by common causes of vision loss. Early diagnosis and treatment are crucial to avoid the consequences of visual impairment. However, in early stages, many visual problems are making it difficult to detect. Visual adaptation can compensate for several visual deficits with adaptive eye movements. These adaptive eye movements may serve as indicators of vision loss. In this work, we investigate the association between eye movement and blurred vision. By using Electrooculography (EOG) to record eye movements, we propose a new tracking model to identify the deterioration of refractive power. We verify the technical feasibility of this method by designing a blurred vision simulation experiment. Six sets of prescription lenses and a pair of flat lenses were used to create different levels of blurring effects. We analyzed binocular movements through EOG signals and performed a seven-class classification using the ResNet18 architecture. The results revealed an average classification accuracy of 94.7% in the subject-dependent model. However, the subject-independent model presented poor performance, with the highest accuracy reaching only 34.5%. Therefore, the potential of an EOG-based visual quality monitoring system is proven. Furthermore, our experimental design provides a novel approach to assessing blurred vision.

Bathroom activities monitoring for older adults by a wrist-mounted accelerometer using a hybrid deep learning model.

Monitoring activities of daily life (ADLs) allows to evaluate health conditions for older adults. However, there are still a limited number of studies on bathroom activities monitoring using a wrist-mounted accelerometer. To fill this gap, in this study, researchers collected data from 15 older adults wearing a wrist-mounted accelerometer. Six bathroom activities, i.e., dressing, undressing, brushing teeth, using toilet, washing face, and washing hands, were investigated. In total, 49.4-hour data for bathroom activities were collected. A hybrid convolutional neural network (CNN) is introduced for bathroom activity recognition. This hybrid CNN model is developed using both hand-crafted and CNN-based features as input. The proposed hybrid CNN model is compared to four machine learning models, i.e., Multilayer Perceptron (MLP), Support Vector Machines (SVM), K-nearest Neighbors (KNN), and Decision Trees (DT), and a conventional CNN model. Based on the classification results of leave-one-subject-out cross-validation (LOSO), the hybrid CNN model outperformed the other models. The hybrid CNN model is also tested based on a transfer learning method. As a calibration step based on LOSO, the transfer learning method additionally trains the model with an example of each activity from the test subject. The transfer learning method obtained better classification performance than LOSO. With transfer learning, the f1-score for using toilet was improved from 0.7784 to 0.8437. This study proposes a deep learning model fusing hand-crafted features and CNN-based features. Besides, the transfer learning method offers a way to build subject-dependent models to improve the classification performance.Clinical relevance -This provides a model that helps monitoring older adults' bathroom activities using a single wrist-mounted accelerometer.

Fear Recognition for Women Using a Reduced Set of Physiological Signals.

Emotion recognition is benefitting from the latest research into physiological monitoring and wireless communications, among other remarkable achievements. These technologies can indeed provide solutions to protect vulnerable people in scenarios such as personal assaults, the abuse of children or the elderly, gender violence or sexual aggression. Cyberphysical systems using smart sensors, artificial intelligence and wearable and inconspicuous devices can serve as bodyguards to detect these risky situations (through fear-related emotion detection) and automatically trigger a protection protocol. As expected, these systems should be trained and customized for each user to ensure the best possible performance, which undoubtedly requires a gender perspective. This paper presents a specialized fear recognition system for women based on a reduced set of physiological signals. The architecture proposed is characterized by the usage of three physiological sensors, lightweight binary classification and the conjunction of linear (temporal and frequency) and non-linear features. Moreover, a binary fear mapping strategy between dimensional and discrete emotional information based on emotional self-report data is implemented to avoid emotional bias. The architecture is evaluated using a public multi-modal physiological dataset with two approaches (subject-dependent and subject-independent models) focusing on the female participants. As a result, the proposal outperforms the state-of-the-art in fear recognition, achieving a recognition rate of up to 96.33% for the subject-dependent model.

MindLink-Eumpy: An Open-Source Python Toolbox for Multimodal Emotion Recognition.

Emotion recognition plays an important role in intelligent human–computer interaction, but the related research still faces the problems of low accuracy and subject dependence. In this paper, an open-source software toolbox called MindLink-Eumpy is developed to recognize emotions by integrating electroencephalogram (EEG) and facial expression information. MindLink-Eumpy first applies a series of tools to automatically obtain physiological data from subjects and then analyzes the obtained facial expression data and EEG data, respectively, and finally fuses the two different signals at a decision level. In the detection of facial expressions, the algorithm used by MindLink-Eumpy is a multitask convolutional neural network (CNN) based on transfer learning technique. In the detection of EEG, MindLink-Eumpy provides two algorithms, including a subject-dependent model based on support vector machine (SVM) and a subject-independent model based on long short-term memory network (LSTM). In the decision-level fusion, weight enumerator and AdaBoost technique are applied to combine the predictions of SVM and CNN. We conducted two offline experiments on the Database for Emotion Analysis Using Physiological Signals (DEAP) dataset and the Multimodal Database for Affect Recognition and Implicit Tagging (MAHNOB-HCI) dataset, respectively, and conducted an online experiment on 15 healthy subjects. The results show that multimodal methods outperform single-modal methods in both offline and online experiments. In the subject-dependent condition, the multimodal method achieved an accuracy of 71.00% in the valence dimension and an accuracy of 72.14% in the arousal dimension. In the subject-independent condition, the LSTM-based method achieved an accuracy of 78.56% in the valence dimension and an accuracy of 77.22% in the arousal dimension. The feasibility and efficiency of MindLink-Eumpy for emotion recognition is thus demonstrated.

Machine Learning Model Comparisons of User Independent &amp; Dependent Intent Recognition Systems for Powered Prostheses

Developing intelligent prosthetic controllers to recognize user intent across users is a challenge. Machine learning algorithms present an opportunity to develop methods for predicting user's locomotion mode. Currently, linear discriminant analysis (LDA) offers the standard solution in the state-of-the-art for subject dependent models and has been used in the development of subject independent applications. However, the performance of subject independent models differ radically from their dependent counterpart. Furthermore, most of the studies limit the evaluation to a fixed terrain with individual stair height and ramp inclination. In this study, we investigated the use of the XGBoost algorithm for developing a subject independent model across 8 individuals with transfemoral amputation. We evaluated the performance of XGBoost across different stair heights and inclination angles and found that it generalizes well across preset conditions. Our findings suggest that XGBoost offers a potential benefit for both subject independent and subject dependent algorithms outperforming LDA and NN (DEP SS Error: 2.93% ± 0.49%, DEP TS Error: 7.03% ± 0.74%, IND SS Error: 10.12% ± 3.16%, and IND TS Error: 15.78% ± 2.39%)(p <; 0.05). We were also able to show that with the inclusion of extra sensors the model performance could continually be improved in both user dependent and independent models (p <; 0.05). Our study provides valuable information for future intent recognition systems to make them more reliable across different users and common community ambulation modes.

BPNN-Based Real-Time Recognition of Locomotion Modes for an Active Pelvis Orthosis with Different Assistive Strategies

Real-time human intent recognition is important for controlling low-limb wearable robots. In this paper, to achieve continuous and precise recognition results on different terrains, we propose a real-time training and recognition method for six locomotion modes including standing, level ground walking, ramp ascending, ramp descending, stair ascending and stair descending. A locomotion recognition system is designed for the real-time recognition purpose with an embedded BPNN-based algorithm. A wearable powered orthosis integrated with this system and two inertial measurement units is used as the experimental setup to evaluate the performance of the designed method while providing hip assistance. Experiments including on-board training and real-time recognition parts are carried out on three able-bodied subjects. The overall recognition accuracies of six locomotion modes based on subject-dependent models are 98.43% and 98.03% respectively, with the wearable orthosis in two different assistance strategies. The cost time of recognition decision delivered to the orthosis is about 0.9[Formula: see text]ms. Experimental results show an effective and promising performance of the proposed method to realize real-time training and recognition for future control of low-limb wearable robots assisting users on different terrains.

Subject-Independent Brain-Computer Interfaces Based on Deep Convolutional Neural Networks.

For a brain-computer interface (BCI) system, a calibration procedure is required for each individual user before he/she can use the BCI. This procedure requires approximately 20-30 min to collect enough data to build a reliable decoder. It is, therefore, an interesting topic to build a calibration-free, or subject-independent, BCI. In this article, we construct a large motor imagery (MI)-based electroencephalography (EEG) database and propose a subject-independent framework based on deep convolutional neural networks (CNNs). The database is composed of 54 subjects performing the left- and right-hand MI on two different days, resulting in 21 600 trials for the MI task. In our framework, we formulated the discriminative feature representation as a combination of the spectral-spatial input embedding the diversity of the EEG signals, as well as a feature representation learned from the CNN through a fusion technique that integrates a variety of discriminative brain signal patterns. To generate spectral-spatial inputs, we first consider the discriminative frequency bands in an information-theoretic observation model that measures the power of the features in two classes. From discriminative frequency bands, spectral-spatial inputs that include the unique characteristics of brain signal patterns are generated and then transformed into a covariance matrix as the input to the CNN. In the process of feature representations, spectral-spatial inputs are individually trained through the CNN and then combined by a concatenation fusion technique. In this article, we demonstrate that the classification accuracy of our subject-independent (or calibration-free) model outperforms that of subject-dependent models using various methods [common spatial pattern (CSP), common spatiospectral pattern (CSSP), filter bank CSP (FBCSP), and Bayesian spatio-spectral filter optimization (BSSFO)].

Densely Feature Fusion Based on Convolutional Neural Networks for Motor Imagery EEG Classification

Electroencephalogram (EEG) signals have been used in the Brain-computer interface (BCI) technology to implement direct communication between the human body and the outside world, which has important application prospects in the fields of cognitive science and medical rehabilitation. In recent years, deep learning technology has achieved remarkable results in the BCI system, especially the using of convolutional neural networks (CNNs) frameworks for the identification and analysis of motor imagery signals. However, practical applications are limited by the complex process of data representation, and the end-to-end method will deteriorate the recognition results. In this paper, we propose a densely feature fusion convolutional neural networks (DFFN). Combining the morphological information of EEG signals, we propose two data representation methods with low complexity, then design and optimize the densely feature fusion network framework for this form of inputs. DFFN considers the correlation between adjacent layers and cross layer features, which reduces the information loss in the process of convolutional operation and considers the local and global characteristics of the network. The simulation results showed that our network improve classification results by 5% in the BCI competition IV-2a data set compare to the ordinary CNNs framework. In order to verify the practical application of the densely feature fusion network framework, we train an adaptive global model method. The results of average classification are close to the baseline approach of the subject-dependent model and better than others.

A Real-Time Lift Detection Strategy for a Hip Exoskeleton.

Repetitive lifting of heavy loads increases the risk of back pain and even lumbar vertebral injuries to workers. Active exoskeletons can help workers lift loads by providing power assistance, and therefore reduce the moment and force applied on L5/S1 joint of human body when performing lifting tasks. However, most existing active exoskeletons for lifting assistance are unable to automatically detect user's lift movement, which limits the wide application of active exoskeletons in factories. In this paper, we propose a simple but effective lift detection strategy for exoskeleton control. This strategy uses only exoskeleton integrated sensors, without any extra sensors to capture human motion intentions. This makes the lift detection system more practical for applications in manufacturing environments. Seven healthy subjects participated in this research. Three different sessions were carried out, two for training and one for testing the algorithm. In the two training sessions, subjects were asked to wear a hip exoskeleton, controlled in transparent mode, and perform repetitive lifting and a locomotion circuit; lifting was executed with different techniques. The collected data were used to train the lift detection model. In the testing session, the exoskeleton was controlled in order to deliver torque to assist the lifting action, based on the lift detection made by the trained algorithm. The across-subject average accuracy of lift detection during online test was 97.97 ± 1.39% with subject-dependent model. Offline, the algorithm was trained with data acquired from all subjects to verify its performance for subject-independent detection, and an accuracy of 97.48 ± 1.53% was achieved. In addition, timeliness of the algorithm was quantitatively evaluated and the time delay was <160 ms across different lifting speeds. Surface electromyography was also measured to assess the efficacy of the exoskeleton in assisting subjects in performing load lifting tasks. These results validate the promise of applying the proposed lift detection strategy for exoskeleton control aiming at lift assistance.

Factor Analysis for Finding Invariant Neural Descriptors of Human Emotions

A major challenge in decoding human emotions from electroencephalogram (EEG) data is finding representations that are invariant to inter- and intrasubject differences. Most of the previous studies are focused in building an individual discrimination model for every subject (subject dependent model). Building subject-independent models is a harder problem due to the high data variability between different subjects and different experiments with the same subject. This paper explores, for the first time, the Factor Analysis as an efficient technique to extract temporal and spatial EEG features suitable to build brain-computer interface for decoding human emotions across various subjects. Our findings show that early waves (temporal window of 200–400 ms after the stimulus onset) carry more information about the valence of the emotion. Also, spatial location of features, with a stronger impact on the emotional valence, occurs in the parietal and occipital regions of the brain. All discrimination models (NN, SVM, kNN, and RF) demonstrate better discrimination rate of the positive valence. These results match closely experimental psychology hypothesis that, during early periods after the stimulus presentation, the brain response—to images with highly positive valence—is stronger.

Multiple order model migration and optimal model selection for online glucose prediction in Type 1 diabetes

In order to address the problem of insufficient available modelling data for glucose prediction, as well as modelling burden, a model migration method was developed in a previous work to quickly transfer an old model to a new subject by using a simple parameter adjustment. However, this method, which is referred to as first order model migration (FOMM), only considers a single order for each exogenous input, and may not produce an optimal model structure for accurate prediction. To overcome this problem, a multiple order model migration (MOMM) algorithm is proposed in this study. For different numbers of modelling samples, including glucose and two exogenous inputs (meal and insulin), the optimal modelling method may be different, and therefore must be properly determined for each modelling scenario. First, the optimal model order is determined for each input and a multiple order prediction model is used. Then, a MOMM algorithm is developed based on particle swarm optimization (PSO) to simultaneously revise multiple parameters. The multiple order parameters of each input in the old model are quickly customized so that the revised model can be used for new subjects with desirable prediction accuracy. In particular, the influence of the number of modelling samples is analysed to check the applicability of different methods; this analysis determines the appropriate selection guidelines for the optimal model in response to different data sizes. The proposed method was evaluated using thirty in silico subjects and clinical data from seven individuals with type 1 diabetes mellitus (DM). Overall, the MOMM algorithm presented superior results when the time period for collecting the samples was larger than 10 h (50 samples). In particular, the size of the modelling samples was separated into three different regions by evaluating the glucose prediction performance and the comparison between different algorithms for both in silico and clinical subjects. In Region I, the FOMM method achieves the best performance. In Region II, the MOMM method should be used and the prediction accuracy is superior in Region II in general. With enough samples (Region III), the subject-dependent model (SM) algorithm can be chosen. The MOMM algorithm is demonstrated to be able to transfer models for new subjects with improved model structure. This article is protected by copyright. All rights reserved.

Real-Time Subject-Independent Pattern Classification of Overt and Covert Movements from fNIRS Signals

Recently, studies have reported the use of Near Infrared Spectroscopy (NIRS) for developing Brain–Computer Interface (BCI) by applying online pattern classification of brain states from subject-specific fNIRS signals. The purpose of the present study was to develop and test a real-time method for subject-specific and subject-independent classification of multi-channel fNIRS signals using support-vector machines (SVM), so as to determine its feasibility as an online neurofeedback system. Towards this goal, we used left versus right hand movement execution and movement imagery as study paradigms in a series of experiments. In the first two experiments, activations in the motor cortex during movement execution and movement imagery were used to develop subject-dependent models that obtained high classification accuracies thereby indicating the robustness of our classification method. In the third experiment, a generalized classifier-model was developed from the first two experimental data, which was then applied for subject-independent neurofeedback training. Application of this method in new participants showed mean classification accuracy of 63% for movement imagery tasks and 80% for movement execution tasks. These results, and their corresponding offline analysis reported in this study demonstrate that SVM based real-time subject-independent classification of fNIRS signals is feasible. This method has important applications in the field of hemodynamic BCIs, and neuro-rehabilitation where patients can be trained to learn spatio-temporal patterns of healthy brain activity.

Acoustic reflex attenuation in phon loudness measurements

Equal Loudness-level Contours, ELC(f, L), represent the sound pressure level in dB SPL of tones at frequency f that are perceived by normal-hearing listeners as equally loud as a 1 kHz tone at L dB SPL. Loudness is defined relatively as L phons (Fletcher & Munson, 1933). ELC measurements by Lydolf and Moller (1997) included in the current ISO standard (Suzuki & Takeshima, JASA vol. 116, 2004), show systematic increases in ELC growth rate with loudness above 60 phons and below 1 kHz, which suggests middle-ear attenuation by the acoustic reflex (AR). A steady-state ELC model was assembled including known mechanisms: (1) middle ear transmission modified by a head-related-transfer-function, (2) compressive cochlear amplification (CA) for signaling loudness, (3) a negative feedback model for AR attenuation by CA inputs exceeding AR threshold, and (4) attenuation of pressure-field stimuli by trans-eardrum static pressure. Model parameters were calculated from ELC data using minimum-square-error estimation. AR attenuation below 1 kHz depends on AR attenuation at the 1 kHz loudness reference frequency, but predicted ELCs are relatively insensitive to it. An earlier psychophysical study of AR attenuation, including 1 kHz, is consistent with subject-dependent model predictions (Rabinowitz & Goldstein, JASA vol. 54, 1973; Rabinowitz, 1977). [NIH-Funded.]

Rapid Model Identification for Online Glucose Prediction of New Subjects With Type 1 Diabetes Using Model Migration Method

Online prediction of glucose concentration is of importance for blood glucose control in diabetes. For conventional modeling methods, model identification has to be repeated with sufficient data collected for each subject. This may cause repetitive cost and burden for patients and clinician and requires a lot of modeling efforts. Here, a rapid model development strategy is proposed using the idea of model migration for new subjects. First, a base model is obtained which can be empirically identified from any subject or constructed by priori knowledge. Then parameters of inputs in the base model are properly revised based on a small amount of data from new subjects. These issues are investigated by developing auto-regressive models with exogenous inputs (ARX) based on thirty in silico subjects. Some important issues relating to model adjustment performance are also checked, referring to the data used for model parameter adjustment and the interaction of two inputs, etc. The rapid modeling method is compared with subject-dependent models developed based on a large number of data with respect to on-line short-term (30min) glucose prediction accuracy.

Real-time EEG-based happiness detection system.

We propose to use real-time EEG signal to classify happy and unhappy emotions elicited by pictures and classical music. We use PSD as a feature and SVM as a classifier. The average accuracies of subject-dependent model and subject-independent model are approximately 75.62% and 65.12%, respectively. Considering each pair of channels, temporal pair of channels (T7 and T8) gives a better result than the other area. Considering different frequency bands, high-frequency bands (Beta and Gamma) give a better result than low-frequency bands. Considering different time durations for emotion elicitation, that result from 30 seconds does not have significant difference compared with the result from 60 seconds. From all of these results, we implement real-time EEG-based happiness detection system using only one pair of channels. Furthermore, we develop games based on the happiness detection system to help user recognize and control the happiness.

EEG-Based Subject- and Session-independent Drowsiness Detection: An Unsupervised Approach

Monitoring and prediction of changes in the human cognitive states, such as alertness and drowsiness, using physiological signals are very important for driver's safety. Typically, physiological studies on real-time detection of drowsiness usually use the same model for all subjects. However, the relatively large individual variability in EEG dynamics relating to loss of alertness implies that for many subjects, group statistics may not be useful to accurately predict changes in cognitive states. Researchers have attempted to build subject-dependent models based on his/her pilot data to account for individual variability. Such approaches cannot account for the cross-session variability in EEG dynamics, which may cause problems due to various reasons including electrode displacements, environmental noises, and skin-electrode impedance. Hence, we propose an unsupervised subject- and session-independent approach for detection departure from alertness in this study. Experimental results showed that the EEG power in the alpha-band (as well as in the theta-band) is highly correlated with changes in the subject's cognitive state with respect to drowsiness as reflected through his driving performance. This approach being an unsupervised and session-independent one could be used to develop a useful system for noninvasive monitoring of the cognitive state of human operators in attention-critical settings.