Search

Older adults with dementia have a high risk of developing drug-induced parkinsonism; however, formal clinical gait assessments are too infrequent to capture fluctuations in their gait. Camera-based human pose estimation and tracking provides a means to frequently monitor gait in nonclinical settings. In this study, 2160 walking bouts from 49 participants were recorded using a ceiling-mounted camera. Recorded color videos were processed using AlphaPose to obtain 2D joint trajectories of the participant as they were walking down a hallway of the unit. A subset of 324 walking bouts from 14 participants were annotated with clinical scores of parkinsonism on the Unified Parkinson's Disease Rating Scale (UPDRS)-gait scale. Linear, random forest, and ordinal logistic regression models were evaluated for regression to UPDRS-gait scores using engineered 2D gait features calculated from the AlphaPose joint trajectories. Additionally, spatial temporal graph convolutional networks (ST-GCNs) were trained to predict UPDRS-gait scores from joint trajectories and gait features using a two-stage training scheme (self-supervised pretraining stage on all walks followed by a finetuning stage on labelled walks). All models were trained using leave-one-subject-out cross-validation to simulate testing on previously unseen participants. The macro-averaged F1-score was 0.333 for the best model operating on only gait features and 0.372 for the top ST-GCN model that used both joint trajectories and gait features as input. When accepting predicted scores that were only off by at most 1 point on the UPDRS-gait scale, the accuracy of the model that only used gait features was 82.8%, while the model that also used joint trajectories had an accuracy of 94.2%.Clinical Relevance- The combination of gait features and joint trajectories capture parkinsonian qualities in gait better than either group of data individually.

Objectives: There are several mechanisms for monitoring the quality of care in long-term care (LTC), including the use of quality indicators derived from resident assessments and formal inspections. The LTC inspection process is time and resource-intensive, and there may be opportunities to better target inspections. In this study, we aimed to examine whether quality indicators could predict future inspection performance in LTC homes across Ontario, Canada., Setting and Participants: In total, 594 LTC homes across Ontario., Methods: Using a database compiling detailed inspection reports for the period from 2017 to 2018, we classified each home into 1 of 3 categories (in good standing, needing improvement, needing significant improvement). Machine learning techniques were used to examine whether publicly available Resident Assessment Instrument‒Minimum Data Set quality indicators for the period 2016‒2017 could predict facility classification based on inspection results., Results: After running a wide range of models, only a weak relationship was found between quality indicators and future inspection performance. The best-performing model was able to achieve a classification accuracy of 40.1%. Feature analysis was performed on the final model to identify which quality indicators were most indicative of predicted poor performance. Experiencing worsened pain, restraint use, and worsened pressure ulcers were correlated with homes predicted as needing significant improvement. Counterintuitively, improved physical functioning had an inverse relationship with homes predicted as being in good standing., Conclusions and Implications: Most quality indicators are poor predictors of inspection performance. Further work is required to explore the limited relationship between these 2 measures of LTC quality, and to identify other quality measures that may be useful as predictors of facilities facing difficulty in meeting quality standards., (Copyright © 2019 AMDA – The Society for Post-Acute and Long-Term Care Medicine. Published by Elsevier Inc. All rights reserved.)

Background: Sleep apnea is a respiratory disorder characterized by an intermittent reduction (hypopnea) or cessation (apnea) of breathing during sleep. Depending on the presence of a breathing effort, sleep apnea is divided into obstructive sleep apnea (OSA) and central sleep apnea (CSA) based on the different pathologies involved. If the majority of apneas in a person are obstructive, they will be diagnosed as OSA or otherwise as CSA. In addition, as it is challenging and highly controversial to divide hypopneas into central or obstructive, the decision about sleep apnea type (OSA vs CSA) is made based on apneas only. Choosing the appropriate treatment relies on distinguishing between obstructive apnea (OA) and central apnea (CA)., Objective: The objective of this study was to develop a noncontact method to distinguish between OAs and CAs., Methods: Five different computer vision-based algorithms were used to process infrared (IR) video data to track and analyze body movements to differentiate different types of apnea (OA vs CA). In the first two methods, supervised classifiers were trained to process optical flow information. In the remaining three methods, a convolutional neural network (CNN) was designed to extract distinctive features from optical flow and to distinguish OA from CA., Results: Overnight sleeping data of 42 participants (mean age 53, SD 15 years; mean BMI 30, SD 7 kg/m 2 ; 27 men and 15 women; mean number of OA 16, SD 30; mean number of CA 3, SD 7; mean apnea-hypopnea index 27, SD 31 events/hour; mean sleep duration 5 hours, SD 1 hour) were collected for this study. The test and train data were recorded in two separate laboratory rooms. The best-performing model (3D-CNN) obtained 95% accuracy and an F 1 score of 89% in differentiating OA vs CA., Conclusions: In this study, the first vision-based method was developed that differentiates apnea types (OA vs CA). The developed algorithm tracks and analyses chest and abdominal movements captured via an IR video camera. Unlike previously developed approaches, this method does not require any attachment to a user that could potentially alter the sleeping condition., (©Sina Akbarian, Nasim Montazeri Ghahjaverestan, Azadeh Yadollahi, Babak Taati. Originally published in the Journal of Medical Internet Research (http://www.jmir.org), 22.05.2020.)

Sleep apnea is a common chronic respiratory disorder which occurs due to the repetitive complete or partial cessations of breathing during sleep. The gold standard assessment of sleep apnea requires full night polysomnography in a sleep laboratory which is expensive, time consuming, and inconvenient. Hence, there is an urgent need for a convenient, robust and wearable monitoring device for screening of sleep apnea. A simple and convenient accelerometer-based portable system is presented to estimate the severity of sleep apnea by analyzing tracheal movements. Respiratory related movements were recorded over the suprasternal notch using a 3D accelerometer. Twenty-one physiological features (7 features, 3 accelerometer channels) were extracted. Performance of three different deep learning models - convolutional neural network, recurrent neural network, and their combination - were evaluated for estimating the apnea hypopnea index (AHI). The estimated AHI is compared to the gold standard polysomnography. In 3-fold cross-validation experiments with 20 participants (9 female, age=47.8±18.0 years, BMI=30.8±4.8, AHI=22.2±21.8 events/hr), we achieved a correlation coefficient between gold standard and estimated values (r-value = 0.84). The proposed system is an accurate, convenient, and portable device suitable for home sleep apnea screening.

This study aimed to extract respiratory signal from tracheal sound recordings during cardio-pulmonary exercise test (CPET), which is the state-of-the-art diagnosis of cardiovascular and respiratory diseases. However, the availability of CPET is limited due to its expense. Physiological signal analysis is a convenient alternative to measure clinical parameters. We collected data from 30 healthy adults and applied wavelet transform thresholding (WTT), empirical mode decomposition (EMD), and Synchrosqueezing transform filtering (SST) to de-noise the raw data. Signal to noise ratio (SNR) was calculated as a quantitative measure of signal quality. We observed that SST yielded the highest SNR and introduced lowest signal distortion by visual-auditory inspection. Respiratory rate was then determined using the signal extracted by SST. Discrepancy of respiratory rate determined from the signal and the gold standard CPET was within 1.2 breaths per minute. In conclusion, SST effectively removed noises in tracheal sound recorded during CPET and provided analyzable respiratory signal for clinical parameter determination.

Sleep apnea is a highly prevalent and underdiagnosed sleep disorder characterized by repeated intermittent interruptions to breathing. Sleep apnea severity is measured with the apnea-hypopnea index (AHI), defined as the number of apnea or hypopnea events per hour of sleep. We hypothesize that respiratory related motion features extracted from infrared video can be used to reliably estimate AHI. The 3 feature variables chosen for apneic event estimation, and separately for sleep versus awake estimation, were: the estimated respiratory rate, the magnitude of respiratory movement, and the amount of movements. Leave-one-person-out cross validation on data from 19 participants was used to train and test a random forest binary classifier to detect apneas and hypopneas. Linear regression of the number of estimated events over estimated sleep duration and the total duration of estimated apneic events over estimated sleep duration was used to estimate AHI. Sleeping versus awake segments was estimated with mean ± standard deviation accuracy of 76.0% ± 17.7%. AHI was estimated with correlation coefficient of 0.76 (p <; 0.01) to the clinical gold standard AHI. Accuracy of 78.9% was achieved for classifying AHI ≥ 15, with sensitivity of 70.0%, specificity of 88.9%, and precision of 87.5%. Motion features extracted from infrared video are concluded to be suitable for estimation of AHI.

An efficient and reliable method to detect drowsiness can reduce accidents and injuries related to drowsy driving. However, existing systems for detecting drowsiness are often of low-resolution, expensive, and dependent on external parameters. Therefore, the goal of this study is to develop a high-resolution and efficient drowsiness detection algorithm using relatively less noisy sleep study data. To this end, we recorded electroencephalogram (EEG) from 53 subjects during a sleep study and leveraged the EEG frequency band changes at sleep onset to develop a model for drowsiness detection. The model devised herein provided a likelihood of wakefulness for 3-s signal segments. By choosing appropriate thresholds of the model output, we have identified three clusters that represent wakefulness, drowsiness, and, sleep. The proposed scheme has been validated using arousals which are cases of alertness and deep sleep segments, cluster quality evaluation metrics, graphical, and statistical analyses. The results presented in this work suggest that spectral properties of EEG can be utilized for high-resolution drowsiness detection in sleep study. Upon its successful validation in a driving study, the proposed model can lead to the development of an efficient drowsy driving monitoring system.

Facial expressions of pain are important in assessing individuals with dementia and severe communicative limitations. Though frontal views of the face are assumed to allow for the most valid and reliable observational assessments, the impact of viewing angle is unknown. We video-recorded older adults with and without dementia using cameras capturing different observational angles (e.g., front vs. profile view) both during a physiotherapy examination designed to identify painful areas and during a baseline period. Facial responses were coded using the fine-grained Facial Action Coding System, as well as a systematic clinical observation method. Coding was conducted separately for panoramic (incorporating left, right, and front views), and a profile view of the face. Untrained observers also judged the videos in a laboratory setting. Trained coder reliability was satisfactory for both the profile and panoramic view. Untrained observer judgments from a profile view were substantially more accurate compared to the front view and accounted for more variance in differentiating non-painful from painful situations. The findings add specificity to the communications models of pain (clarifying factors influencing observers' ability to decode pain messages). Perhaps more importantly, the findings have implications for the development of computer vision algorithms and vision technologies designed to monitor and interpret facial expressions in a pain context. That is, the performance of such automated systems is heavily influenced by how reliably these human annotations could be provided and, hence, evaluation of human observers' reliability, from multiple angles of observation, has implications for machine learning development efforts.

Introduction: Technological solutions for quantifying Parkinson's disease (PD) symptoms may provide an objective means to track response to treatment, including side effects such as levodopa-induced dyskinesia. Vision-based systems are advantageous as they do not require physical contact with the body and have minimal instrumentation compared to wearables. We have developed a vision-based system to quantify a change in dyskinesia as reported by patients using 2D videos of clinical assessments during acute levodopa infusions., Methods: Nine participants with PD completed a total of 16 levodopa infusions, where they were asked to report important changes in dyskinesia (i.e. onset and remission). Participants were simultaneously rated using the UDysRS Part III (from video recordings analyzed post-hoc). Body joint positions and movements were tracked using a state-of-the-art deep learning pose estimation algorithm applied to the videos. 416 features (e.g. kinematics, frequency distribution) were extracted to characterize movements. The sensitivity and specificity of each feature to patient-reported changes in dyskinesia severity was computed and compared with physician-rated results., Results: Features achieved similar or superior performance to the UDysRS for detecting the onset and remission of dyskinesia. The best AUC for detecting onset of dyskinesia was 0.822 and for remission of dyskinesia was 0.958, compared to 0.826 and 0.802 for the UDysRS., Conclusions: Video-based features may provide an objective means of quantifying the severity of levodopa-induced dyskinesia, and have responsiveness as good or better than the clinically-rated UDysRS. The results demonstrate encouraging evidence for future integration of video-based technology into clinical research and eventually clinical practice., (Copyright © 2018 Elsevier Ltd. All rights reserved.)

Agitation and aggression are among the most challenging symptoms of dementia. Agitated persons with dementia can harm themselves, their caregivers, or other patients in a care facility. Automatic detection of agitation would be useful to alert caregivers so that appropriate interventions can be performed. The building blocks in the automatic detection of agitation and aggression are appropriate sensing platforms and generalized predictive models. In this article, we perform a systematic review of studies that use different types of sensors to detect agitation and aggression in persons with dementia. We conclude that actigraphy shows some evidence of correlation with incidences of agitation and aggression; however, multimodal sensing has not been fully evaluated for this purpose. Based on this systematic review, we provide guidelines and recommendations for future research directions in this field., (Copyright © 2018 the Alzheimer's Association. Published by Elsevier Inc. All rights reserved.)

Robotic stroke rehabilitation therapy can greatly increase the efficiency of therapy delivery. However, when left unsupervised, users often compensate for limitations in affected muscles and joints by recruiting unaffected muscles and joints, leading to undesirable rehabilitation outcomes. This paper aims to develop a computer vision system that augments robotic stroke rehabilitation therapy by automatically detecting such compensatory motions. Nine stroke survivors and ten healthy adults participated in this study. All participants completed scripted motions using a table-top rehabilitation robot. The healthy participants also simulated three types of compensatory motions. The 3-D trajectories of upper body joint positions tracked over time were used for multiclass classification of postures. A support vector machine (SVM) classifier detected lean-forward compensation from healthy participants with excellent accuracy (AUC = 0.98, F1 = 0.82), followed by trunk-rotation compensation (AUC = 0.77, F1 = 0.57). Shoulder-elevation compensation was not well detected (AUC = 0.66, F1 = 0.07). A recurrent neural network (RNN) classifier, which encodes the temporal dependency of video frames, obtained similar results. In contrast, F1-scores in stroke survivors were low for all three compensations while using RNN: lean-forward compensation (AUC = 0.77, F1 = 0.17), trunk-rotation compensation (AUC = 0.81, F1 = 0.27), and shoulder-elevation compensation (AUC = 0.27, F1 = 0.07). The result was similar while using SVM. To improve detection accuracy for stroke survivors, future work should focus on predefining the range of motion, direct camera placement, delivering exercise intensity tantamount to that of real stroke therapies, adjusting seat height, and recording full therapy sessions.

Because falls are funny, YouTube and other video sharing sites contain a large repository of real-life falls. We propose extracting gait and balance information from these videos to help us better understand some of the factors that contribute to falls. Proof-of-concept is explored in a single video containing multiple (n=14) falls/non-falls in the presence of an unexpected obstacle. The analysis explores: computing spatiotemporal parameters of gait in a video captured from an arbitrary viewpoint; the relationship between parameters of gait from the last few steps before the obstacle and falling vs. not falling; and the predictive capacity of a multivariate model in predicting a fall in the presence of an unexpected obstacle. Homography transformations correct the perspective projection distortion and allow for the consistent tracking of gait parameters as an individual walks in an arbitrary direction in the scene. A synthetic top view allows for computing the average stride length and a synthetic side view allows for measuring up and down motions of the head. In leave-one-out cross-validation, we were able to correctly predict whether a person would fall or not in 11 out of the 14 cases (78.6%), just by looking at the average stride length and the range of vertical head motion during the 1-4 most recent steps prior to reaching the obstacle.

Levodopa is the gold standard therapy for Parkinson's disease (PD), but its prolonged usage leads to additional motor complications, namely levodopa-induced dyskinesia (LID). To assess LID and adjust drug regimens for optimal relief, patients attend regular clinic visits. However, the intermittent nature of these visits can fail to capture important changes in a person's condition. With the recent emergence of deep learning achieving impressive results in a wide array of fields including computer vision, there is an opportunity for video analysis to be used for automated assessment of LID. Deep learning for pose estimation was studied as a viable means of extracting body movements from PD assessment videos. Movement features were computed from joint trajectories. Results show that features derived from vision-based analysis have moderate to good correlation with clinician ratings of dyskinesia severity. This study presents the first application of deep learning to video analysis in PD, and demonstrates promise for future development of a non-contact system for objective PD assessment.

This paper presents a study to evaluate the concurrent validity of the Microsoft Kinect for Windows v2 for measuring the spatiotemporal parameters of gait. Twenty healthy adults performed several sequences of walks across a GAITRite mat under three different conditions: usual pace, fast pace, and dual task. Each walking sequence was simultaneously captured with two Kinect for Windows v2 and the GAITRite system. An automated algorithm was employed to extract various spatiotemporal features including stance time, step length, step time and gait velocity from the recorded Kinect v2 sequences. Accuracy in terms of reliability, concurrent validity and limits of agreement was examined for each gait feature under different walking conditions. The 95% Bland-Altman limits of agreement were narrow enough for the Kinect v2 to be a valid tool for measuring all reported spatiotemporal parameters of gait in all three conditions. An excellent intraclass correlation coefficient (ICC2, 1) ranging from 0.9 to 0.98 was observed for all gait measures across different walking conditions. The inter trial reliability of all gait parameters were shown to be strong for all walking types (ICC3, 1 > 0.73). The results of this study suggest that the Kinect for Windows v2 has the capacity to measure selected spatiotemporal gait parameters for healthy adults., (Copyright © 2016 IPEM. Published by Elsevier Ltd. All rights reserved.)

This study uses machine learning methods to distinguish between healthy and pathological gait. Examples of multi-dimensional pathological and normal gait sequences were collected from post-stroke and healthy individuals in a real clinical setting and with two Kinect sensors. The trajectories of rotational angle and global velocity of selected body joints (hips, spine, shoulders, neck, knees and ankles) over time formed the gait sequences. The combination of k nearest neighbor (kNN) and dynamic time warping (DTW) was used for classification. Leave one subject out cross validation was implemented to evaluate the performance of the binary classifier in terms of F1-score in the original feature space, and also in a reduced dimensional feature space using PCA. The pair of k = 1 in kNN and the warping window size 25% of gait sequences in DTW achieved maximum F1-score. Using PCA, pathological gait sequences were discriminated from healthy sequences with the F1-score = 96%.

A non-contact vision-based system is presented for continuous respiratory rate monitoring. The system identifies feature points in a video feed and tracks them over time. Two methods are presented for comparison - a method which uses principal component analysis (PCA) and a simple averaging approach. These methods condense the feature point trajectories into a compact set of representative signals. The signal which most closely resembles an expected respiratory trace is selected based on spectral analysis. System performance is assessed by comparing the estimated respiratory rate to the rate determined via inductance plethysmogram. The system was evaluated on 5 participants in 4 simulated sleep scenarios. Accuracies of within 1 breath/minute were achieved for more than 97% of the recorded time in all scenarios. The proposed system is accurate, cost-effective, and simple, making it a suitable candidate for at-home installation.

Falling in the home is one of the major challenges to independent living among older adults. The associated costs, coupled with a rapidly growing elderly population, are placing a burden on healthcare systems worldwide that will swiftly become unbearable. To facilitate expeditious emergency care, we have developed an artificially intelligent camera-based system that automatically detects if a person within the field-of-view has fallen. The system addresses concerns raised in earlier work and the requirements of a widely deployable in-home solution. The presented prototype utilizes a consumer-grade camera modified with a wide-angle lens. Machine learning techniques applied to carefully engineered features allow the system to classify falls at high accuracy while maintaining invariance to lighting, environment and the presence of multiple moving objects. This paper describes the system, outlines the algorithms used and presents empirical validation of its effectiveness.