Deep neural networks for wearable sensor-based activity recognition in Parkinson's disease: investigating generalizability and model complexity.

BACKGROUND: The research gap addressed in this study is the applicability of deep neural network (NN) models on wearable sensor data to recognize different activities performed by patients with Parkinson's Disease (PwPD) and the generalizability of these models to PwPD using labeled healthy data. METHODS: The experiments were carried out utilizing three datasets containing wearable motion sensor readings on common activities of daily living. The collected readings were from two accelerometer sensors. PAMAP2 and MHEALTH are publicly available datasets collected from 10 and 9 healthy, young subjects, respectively. A private dataset of a similar nature collected from 14 PwPD patients was utilized as well. Deep NN models were implemented with varying levels of complexity to investigate the impact of data augmentation, manual axis reorientation, model complexity, and domain adaptation on activity recognition performance. RESULTS: A moderately complex model trained on the augmented PAMAP2 dataset and adapted to the Parkinson domain using domain adaptation achieved the best activity recognition performance with an accuracy of 73.02%, which was significantly higher than the accuracy of 63% reported in previous studies. The model's F1 score of 49.79% significantly improved compared to the best cross-testing of 33.66% F1 score with only data augmentation and 2.88% F1 score without data augmentation or domain adaptation. CONCLUSION: These findings suggest that deep NN models originating on healthy data have the potential to recognize activities performed by PwPD accurately and that data augmentation and domain adaptation can improve the generalizability of models in the healthy-to-PwPD transfer scenario. The simple/moderately complex architectures tested in this study could generalize better to the PwPD domain when trained on a healthy dataset compared to the most complex architectures used. The findings of this study could contribute to the development of accurate wearable-based activity monitoring solutions for PwPD, improving clinical decision-making and patient outcomes based on patient activity levels.

Exploring the feasibility of tensor decomposition for analysis of fNIRS signals: a comparative study with grand averaging method.

The analysis of functional near-infrared spectroscopy (fNIRS) signals has not kept pace with the increased use of fNIRS in the behavioral and brain sciences. The popular grand averaging method collapses the oxygenated hemoglobin data within a predefined time of interest window and across multiple channels within a region of interest, potentially leading to a loss of important temporal and spatial information. On the other hand, the tensor decomposition method can reveal patterns in the data without making prior assumptions of the hemodynamic response and without losing temporal and spatial information. The aim of the current study was to examine whether the tensor decomposition method could identify significant effects and novel patterns compared to the commonly used grand averaging method for fNIRS signal analysis. We used two infant fNIRS datasets and applied tensor decomposition (i.e., canonical polyadic and Tucker decompositions) to analyze the significant differences in the hemodynamic response patterns across conditions. The codes are publicly available on GitHub. Bayesian analyses were performed to understand interaction effects. The results from the tensor decomposition method replicated the findings from the grand averaging method and uncovered additional patterns not detected by the grand averaging method. Our findings demonstrate that tensor decomposition is a feasible alternative method for analyzing fNIRS signals, offering a more comprehensive understanding of the data and its underlying patterns.

Activity Recognition in Parkinson's Patients from Motion Data Using a CNN Model Trained by Healthy Subjects.

Physical activity recognition in patients with Parkinson's Disease (PwPD) is challenging due to the lack of large-enough and good quality motion data for PwPD. A common approach to this obstacle involves the use of models trained on better quality data from healthy patients. Models can struggle to generalize across these domains due to motor complications affecting the movement patterns in PwPD and differences in sensor axes orientations between data. In this paper, we investigated the generalizability of a deep convolutional neural network (CNN) model trained on a young, healthy population to PD, and the role of data augmentation on alleviating sensor position variability. We used two publicly available healthy datasets - PAMAP2 and MHEALTH. Both datasets had sensor placements on the chest, wrist, and ankle with 9 and 10 subjects, respectively. A private PD dataset was utilized as well. The proposed CNN model was trained on PAMAP2 in k-fold cross-validation based on the number of subjects, with and without data augmentation, and tested directly on MHEALTH and PD data. Without data augmentation, the trained model resulted in 48.16% accuracy on MHEALTH and 0% on the PD data when directly applied with no model adaptation techniques. With data augmentation, the accuracies improved to 87.43% and 44.78%, respectively, indicating that the method compensated for the potential sensor placement variations between data. Clinical Relevance- Wearable sensors and machine learning can provide important information about the activity level of PwPD. This information can be used by the treating physician to make appropriate clinical interventions such as rehabilitation to improve quality of life.

Dyskinesia Estimation of Imbalanced Data Using a Deep-Learning Model.

The collection of Parkinson's Disease (PD) time-series data usually results in imbalanced and incomplete datasets due to the geometric distribution of PD complications' sever-ity scores. Consequently, when training deep convolutional models on these datasets, the models suffer from overfitting and lack generalizability to unseen data. In this paper, we investigated a new framework of Conditional Generative Ad-versarial Netuwoks (cGANs) as a solution to improve the extrapolation and generalizability of the regression models in such datasets. We used a real-world PD dataset to esti-mate Dyskinesia severity in patients with PD. The developed cGAN demonstrated significantly better generalizability to unseen data samples than a traditional Convolutional Neural Network with an improvement of 34%. This solution can be applied in similar imbalanced time-series data, especially in the healthcare domain, where balanced and uniformly distributed data samples are not readily available.

The forecast of COVID-19 spread risk at the county level.

The early detection of the coronavirus disease 2019 (COVID-19) outbreak is important to save people's lives and restart the economy quickly and safely. People's social behavior, reflected in their mobility data, plays a major role in spreading the disease. Therefore, we used the daily mobility data aggregated at the county level beside COVID-19 statistics and demographic information for short-term forecasting of COVID-19 outbreaks in the United States. The daily data are fed to a deep learning model based on Long Short-Term Memory (LSTM) to predict the accumulated number of COVID-19 cases in the next two weeks. A significant average correlation was achieved (r=0.83 (p = 0.005)) between the model predicted and actual accumulated cases in the interval from August 1, 2020 until January 22, 2021. The model predictions had r > 0.7 for 87% of the counties across the United States. A lower correlation was reported for the counties with total cases of <1000 during the test interval. The average mean absolute error (MAE) was 605.4 and decreased with a decrease in the total number of cases during the testing interval. The model was able to capture the effect of government responses on COVID-19 cases. Also, it was able to capture the effect of age demographics on the COVID-19 spread. It showed that the average daily cases decreased with a decrease in the retiree percentage and increased with an increase in the young percentage. Lessons learned from this study not only can help with managing the COVID-19 pandemic but also can help with early and effective management of possible future pandemics. The code used for this study was made publicly available on https://github.com/Murtadha44/covid-19-spread-risk. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s40537-021-00491-1.

Dyskinesia estimation during activities of daily living using wearable motion sensors and deep recurrent networks.

Levodopa-induced dyskinesias are abnormal involuntary movements experienced by the majority of persons with Parkinson's disease (PwP) at some point over the course of the disease. Choreiform as the most common phenomenology of levodopa-induced dyskinesias can be managed by adjusting the dose/frequency of PD medication(s) based on a PwP's motor fluctuations over a typical day. We developed a sensor-based assessment system to provide such information. We used movement data collected from the upper and lower extremities of 15 PwPs along with a deep recurrent model to estimate dyskinesia severity as they perform different activities of daily living (ADL). Subjects performed a variety of ADLs during a 4-h period while their dyskinesia severity was rated by the movement disorder experts. The estimated dyskinesia severity scores from our model correlated highly with the expert-rated scores (r = 0.87 (p < 0.001)), which was higher than the performance of linear regression that is commonly used for dyskinesia estimation (r = 0.81 (p < 0.001)). Our model provided consistent performance at different ADLs with minimum r = 0.70 (during walking) to maximum r = 0.84 (drinking) in comparison to linear regression with r = 0.00 (walking) to r = 0.76 (cutting food). These findings suggest that when our model is applied to at-home sensor data, it can provide an accurate picture of changes of dyskinesia severity facilitating effective medication adjustments.

Ensemble deep model for continuous estimation of Unified Parkinson's Disease Rating Scale III.

BACKGROUND: Unified Parkinson Disease Rating Scale-part III (UPDRS III) is part of the standard clinical examination performed to track the severity of Parkinson's disease (PD) motor complications. Wearable technologies could be used to reduce the need for on-site clinical examinations of people with Parkinson's disease (PwP) and provide a reliable and continuous estimation of the severity of PD at home. The reported estimation can be used to successfully adjust the dose and interval of PD medications. METHODS: We developed a novel algorithm for unobtrusive and continuous UPDRS-III estimation at home using two wearable inertial sensors mounted on the wrist and ankle. We used the ensemble of three deep-learning models to detect UPDRS-III-related patterns from a combination of hand-crafted features, raw temporal signals, and their time-frequency representation. Specifically, we used a dual-channel, Long Short-Term Memory (LSTM) for hand-crafted features, 1D Convolutional Neural Network (CNN)-LSTM for raw signals, and 2D CNN-LSTM for time-frequency data. We utilized transfer learning from activity recognition data and proposed a two-stage training for the CNN-LSTM networks to cope with the limited amount of data. RESULTS: The algorithm was evaluated on gyroscope data from 24 PwP as they performed different daily living activities. The estimated UPDRS-III scores had a correlation of [Formula: see text] and a mean absolute error of 5.95 with the clinical examination scores without requiring the patients to perform any specific tasks. CONCLUSION: Our analysis demonstrates the potential of our algorithm for estimating PD severity scores unobtrusively at home. Such an algorithm could provide the required motor-complication measurements without unnecessary clinical visits and help the treating physician provide effective management of the disease.

Detection of Mild Cognitive Impairment and Alzheimer's Disease using Dual-task Gait Assessments and Machine Learning.

OBJECTIVE: Early detection of mild cognitive impairment (MCI) and Alzheimer's disease (AD) can increase access to treatment and assist in advance care planning. However, the development of a diagnostic system that d7oes not heavily depend on cognitive testing is a major challenge. We describe a diagnostic algorithm based solely on gait and machine learning to detect MCI and AD from healthy. METHODS: We collected "single-tasking" gait (walking) and "dual-tasking" gait (walking with cognitive tasks) from 32 healthy, 26 MCI, and 20 AD participants using a computerized walkway. Each participant was assessed with the Montreal Cognitive Assessment (MoCA). A set of gait features (e.g., mean, variance and asymmetry) were extracted. Significant features for three classifications of MCI/healthy, AD/healthy, and AD/MCI were identified. A support vector machine model in a one-vs.-one manner was trained for each classification, and the majority vote of the three models was assigned as healthy, MCI, or AD. RESULTS: The average classification accuracy of 5-fold cross-validation using only the gait features was 78% (77% F1-score), which was plausible when compared with the MoCA score with 83% accuracy (84% F1-score). The performance of healthy vs. MCI or AD was 86% (88% F1-score), which was comparable to 88% accuracy (90% F1-score) with MoCA. CONCLUSION: Our results indicate the potential of machine learning and gait assessments as objective cognitive screening and diagnostic tools. SIGNIFICANCE: Gait-based cognitive screening can be easily adapted into clinical settings and may lead to early identification of cognitive impairment, so that early intervention strategies can be initiated.

Tensor Decomposition of Functional Near-Infrared Spectroscopy (fNIRS) Signals for Pattern Discovery of Cognitive Response in Infants.

Functional near-infrared spectroscopy (fNIRS) provides an effective tool in neuroscience studies of cognition in infants. fNIRS signals are normally processed by applying ANOVA analysis on the grand average of the hemodynamic responses to investigate the cognitive-related differences between experimental groups. However, this averaging approach does not account for any differences in the temporal patterns of the responses. Therefore, we propose a new approach based on a combination of tensor decomposition and ANOVA. First, a four-way tensor of the hemodynamic responses is arranged as time × frequency × channel× subject and decomposed using Canonical Polyadic Decomposition (CPD). Next, ANOVA is applied to identify significant patterns between subjects. Instead of averaging, the CPD can capture the distinct patterns between groups in all the dimensions. We used fNIRS dataset of 70 infants who participated in an experiment to investigate cortical activation to an agent (i.e., mechanical claws vs. human hands) with different events (i.e., function and non-function). In the comparison with the traditional ANOVA, CPD+ANOVA identified the same significance factors. However, CPD+ANOVA discovered new information on the temporal and spatial patterns indicating a longer interval hemodynamic responses, which was missed using the traditional ANOVA. This new analysis of hemodynamic responses as captured using fNIRS will improve neuroscience and cognitive studies.

Dual-Task Gait Assessment and Machine Learning for Early-detection of Cognitive Decline.

Alzheimer's disease (AD) affects approximately 30 million people worldwide, and this number is predicted to triple by 2050 unless further discoveries facilitate the early detection and prevention of the disease. Computerized walkways for simultaneous assessment of motor-cognitive performance, known as a dual-task assessment, has been used to associate changes in gait characteristics to mild cognitive impairment (MCI) with early-stage disease. However, to our best knowledge, there is no validated method to detect MCI using the collective analysis of these gait characteristics. In this paper, we develop a machine learning approach to analyze the gait data from the dual-task assessment in order to detect subjects with cognitive impairment from healthy individuals. We collected dual-task gait data from a computerized walkway of a total of 92 subjects with 31 healthy control (HC) and 61 MCI. Using support vector machine (SVM) and gradient tree boosting, we developed a classifier to differentiate MCI from HC subjects and compared the results with a paper-based questionnaire assessment that has been commonly used in clinical practice. SVM provided the highest accuracy of 77.17% with 81.97% sensitivity and 67.74% specificity. Our results indicate the potential of machine learning + dual-task assessment to enable early diagnosis of cognitive decline before it advances to dementia and AD, so that early intervention or prevention strategies can be initiated.

Dyskinesia Severity Estimation in Patients with Parkinson's Disease Using Wearable Sensors and A Deep LSTM Network.

Dyskinesias are abnormal involuntary movements that patients with mid-stage and advanced Parkinson's disease (PD) may suffer from. These troublesome motor impairments are reduced by adjusting the dose or frequency of medication levodopa. However, to make a successful adjustment, the treating physician needs information about the severity rating of dyskinesia as patients experience in their natural living environment. In this work, we used movement data collected from the upper and lower extremities of PD patients along with a deep model based on Long Short-Term Memory to estimate the severity of dyskinesia. We trained and validated our model on a dataset of 14 PD subjects with dyskinesia. The subjects performed a variety of daily living activities while their dyskinesia severity was rated by a neurologist. The estimated dyskinesia severity ratings from our developed model highly correlated with the neurologist-rated dyskinesia scores (r=0.86 (p<0.001) and 1.77 MAE (6%)) indicating the potential of the developed the approach in providing the information required for effective medication adjustments for dyskinesia management.

Multilevel Features for Sensor-Based Assessment of Motor Fluctuation in Parkinson's Disease Subjects.

Motor fluctuations are a frequent complication in patients with Parkinson's disease (PD) where the response to medication fluctuates between ON states (medication working) and OFF states (medication has worn off). This paper describes a new data analysis approach that can be used along with two wearable IMU (inertial measurement units) sensors to continuously assess motor fluctuations in PD patients while moving in their natural environment. We hypothesized that joint analysis of the sensor data in its spectral, temporal and sensor domain could generate multilevel features that can be used to detect PD-related patterns successfully as the subject's motor state fluctuates between medication ON and OFF states. For this purpose, we utilized time-frequency (TF) representation and multiway data analysis tools (i.e., tensor decomposition) to decompose the TF representation of the two sensors' data into its multilevel structures, which were next used to extract multilevel features representing the PD symptoms in different medication states. The extracted multilevel features were used in a classification model based on support vector machine to detect medication ON and OFF states. For comparison purposes, we implemented a traditional feature extraction method. We also developed a hierarchical feature extraction method based on the combination of those two methods. The performances of the three methods were evaluated using a dataset of 19 PD subjects with a total duration of 17.54 hours. The multilevel features achieved 8.25% improvement in the accuracy over the traditional features, and the hierarchical features resulted in 10.73% improvement indicating that our approach holds great promise to continuously detect medication states from continuous monitoring of the subjects' movement. Such information can be used by the treating physician to tailor the adjustments to each subject's unique impairment(s), thereby improving therapeutic decision-making and patient outcomes.

Wearable Sensors for Estimation of Parkinsonian Tremor Severity during Free Body Movements.

Tremor is one of the main symptoms of Parkinson's Disease (PD) that reduces the quality of life. Tremor is measured as part of the Unified Parkinson Disease Rating Scale (UPDRS) part III. However, the assessment is based on onsite physical examinations and does not fully represent the patients' tremor experience in their day-to-day life. Our objective in this paper was to develop algorithms that, combined with wearable sensors, can estimate total Parkinsonian tremor as the patients performed a variety of free body movements. We developed two methods: an ensemble model based on gradient tree boosting and a deep learning model based on long short-term memory (LSTM) networks. The developed methods were assessed on gyroscope sensor data from 24 PD subjects. Our analysis demonstrated that the method based on gradient tree boosting provided a high correlation (r = 0.96 using held-out testing and r = 0.93 using subject-based, leave-one-out cross-validation) between the estimated and clinically assessed tremor subscores in comparison to the LSTM-based method with a moderate correlation (r = 0.84 using held-out testing and r = 0.77 using subject-based, leave-one-out cross-validation). These results indicate that our approach holds great promise in providing a full spectrum of the patients' tremor from continuous monitoring of the subjects' movement in their natural environment.

Assessment of response to medication in individuals with Parkinson's disease.

BACKGROUND AND OBJECTIVE: Motor fluctuations between akinetic (medication OFF) and mobile phases (medication ON) states are one of the most prevalent complications of patients with Parkinson's disease (PD). There is a need for a technology-based system to provide reliable information about the duration in different medication phases that can be used by the treating physician to successfully adjust therapy. METHODS: Two KinetiSense motion sensors were mounted on the most affected wrist and ankle of 19 PD subjects (age: 42-77, 14 males) and collected movement signals as the participants performed seven daily living activities in their medication OFF and ON phases. A feature selection and a classification algorithm based on support vector machine with fuzzy labeling was developed to detect medication ON/OFF states using gyroscope signals. The algorithm was trained using approximately 15% of the data from four activities and tested on the remaining data. RESULTS: The algorithm was able to detect medication ON and OFF states with 90.5% accuracy, 94.2% sensitivity, and 85.4% specificity. It performed equally well for all the activities with an average accuracy of 91.3% for the activities that were used in the training phase and 88.4% for the new activities. CONCLUSIONS: The developed sensor-based algorithm could provide objective and accurate assessment of medication states that can lead to successful adjustment of the therapy resulting in considerably improved care delivery and quality of life of PD patients.

Hybrid Feature Extraction for Detection of Degree of Motor Fluctuation Severity in Parkinson's Disease Patients.

The success of medication adjustment in Parkinson's disease (PD) patients with motor fluctuation relies on the knowledge about their fluctuation severity. However, because of the temporal and spatial variability in motor fluctuations, a single clinical examination often fails to capture the spectrum of motor impairment experienced in routine daily life. In this study, we developed an algorithm to estimate the degree of motor fluctuation severity from two wearable sensors' data during subjects' free body movements. Specifically, we developed a new hybrid feature extraction method to represent the longitudinal changes of motor function from the sensor data. Next, we developed a classification model based on random forest to learn the changes in the patterns of the sensor data as the severity of the motor function changes. We evaluated our algorithm using data from 24 subjects with idiopathic PD as they performed a variety of daily routine activities. A leave-one-subject-out assessment of the algorithm resulted in 83.33% accuracy, indicating that our approach holds a great promise to passively detect degree of motor fluctuation severity from continuous monitoring of an individual's free body movements. Such a sensor-based assessment system and algorithm combination could provide the objective and comprehensive information about the fluctuation severity that can be used by the treating physician to effectively adjust therapy for PD patients with troublesome motor fluctuation.

Deep Learning for Medication Assessment of Individuals with Parkinson's Disease Using Wearable Sensors.

Motor fluctuations between "OFF" state (with no benefit from medication) and " ON" state (with optimum benefit from medication) are a major focus of clinical managements in individuals with mid-stage and advance Parkinson's disease (PD). In this work, an automated algorithm based on Long Short-Term Memory (LSTM) as a deep learning method is developed to identify motor fluctuations in individuals with PD using wearable sensors during a variety of daily living activities. This network was evaluated on two datasets i.e., Dataset 1 and Dataset 2) that included recordings of 19 individuals with PD using subject-based leave-one-out cross-validation. The designed LSTM network yielded promising results using only one ankle sensor with an average classification rate of 73% and 77% for Dataset 1 and Dataset 2, respectively.

Automatic assessment of medication states of patients with Parkinson's disease using wearable sensors.

Motor fluctuations are a major focus of clinical managements in patients with mid-stage and advance Parkinson's disease (PD). In this paper, we develop a new patient-specific algorithm that can classify those fluctuations during a variety of activities. We extract a set of temporal and spectral features from the ambulatory signals and then introduce a semi-supervised classification algorithm based on K-means and self-organizing tree map clustering methods. Two different types of cluster labeling are introduced: hard and fuzzy labeling. The developed algorithm is evaluated on a dataset from triaxial gyroscope sensors for 12 PD patients. The average result of using K-means and fuzzy labeling on the trunk and the more affected leg sensors' readings was 75.96%, 70.57%, and 86.93% for accuracy, sensitivity, and specificity, respectively. The accuracy for individual patients varied from 99.95% to 42.53%, which was correlated with dyskinesia severity and the improvement of the PD symptoms with medication.