Multiple-Instance Learning for Sparse Behavior Modeling from Wearables: Toward Dementia-Related Agitation Prediction

Alam, Ridwan; Bankole, Azziza; Anderson, Martha; Lach, John

doi:10.1109/embc.2019.8856502

Cited by 16 publications

(33 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The data from the subjects was analyzed using an ANN with two classes (control subjects and AD), the approach is low cost and it does not have any side effects. Related to prediction of the agitative behavior for the patients with dementia an important challenge is overcoming weakly labeled and sparse data [82]. The goal is to infer proficiently the agitation episodes from data collected by wearable sensors using multiple-instance learning (MIL) models.…”

Section: For Behavior Assessmentmentioning

confidence: 99%

Smart Environments and Social Robots for Age-Friendly Integrated Care Services

Anghel

Cioara

Moldovan

et al. 2020

IJERPH

View full text Add to dashboard Cite

The world is facing major societal challenges because of an aging population that is putting increasing pressure on the sustainability of care. While demand for care and social services is steadily increasing, the supply is constrained by the decreasing workforce. The development of smart, physical, social and age-friendly environments is identified by World Health Organization (WHO) as a key intervention point for enabling older adults, enabling them to remain as much possible in their residences, delay institutionalization, and ultimately, improve quality of life. In this study, we survey smart environments, machine learning and robot assistive technologies that can offer support for the independent living of older adults and provide age-friendly care services. We describe two examples of integrated care services that are using assistive technologies in innovative ways to assess and deliver of timely interventions for polypharmacy management and for social and cognitive activity support in older adults. We describe the architectural views of these services, focusing on details about technology usage, end-user interaction flows and data models that are developed or enhanced to achieve the envisioned objective of healthier, safer, more independent and socially connected older people.

show abstract

Section: For Behavior Assessmentmentioning

confidence: 99%

Smart Environments and Social Robots for Age-Friendly Integrated Care Services

Anghel

Cioara

Moldovan

et al. 2020

IJERPH

View full text Add to dashboard Cite

show abstract

“…We design the IMU feature space to contain standard statistical, frequency, and power domain features [41][42][43]. The statistical features are extracted as mean, max, median, standard deviation, rms, variance, and interquartile range of the raw IMU signal.…”

Section: ) Feature Spacementioning

confidence: 99%

Wearable Respiration Monitoring: Interpretable Inference with Context and Sensor Biomarkers

Alam

Peden²,

Lach

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Breathing rate (BR), minute ventilation (VE), and other respiratory parameters are essential for real-time patient monitoring in many acute health conditions, such as asthma. The clinical standard for measuring respiration, namely Spirometry, is hardly suitable for continuous use. Wearables can track many physiological signals, like ECG and motion, yet not respiration. Deriving respiration from other modalities has become an area of active research. In this work, we infer respiratory parameters from wearable ECG and wrist motion signals. We propose a modular and generalizable classification-regression pipeline to utilize available context information, such as physical activity, in learning context-conditioned inference models. Morphological and power domain novel features from the wearable ECG are extracted to use with these models. Exploratory feature selection methods are incorporated in this pipeline to discover application-specific interpretable biomarkers. Using data from 15 subjects, we evaluate two implementations of the proposed pipeline: for inferring BR and VE. Each implementation compares generalized linear model, random forest, support vector machine, Gaussian process regression, and neighborhood component analysis as contextual regression models. Permutation, regularization, and relevance determination methods are used to rank the ECG features to identify robust ECG biomarkers across models and activities. This work demonstrates the potential of wearable sensors not only in continuous monitoring, but also in designing biomarker-driven preventive measures.

show abstract

“…With the global number of people with dementia currently at 50 million and predicted to reach 130 million by 2050 [3], it is vital that effective methods to manage the disease and the resulting difficulties are found, so that more people with the disease can have greater independence and maintained quality of life for as long as possible [4]. Technology has been shown to be helpful in this regard, and over recent years much work has been done to develop support systems which can aid in the manage-ment of dementia and the related difficulties [5][6][7][8][9][10][11]. In previous work [12], we reviewed the use of wearable computing-based systems for identifying the occurrence of dementia-related difficulties from physiological data.…”

Section: Introductionmentioning

confidence: 99%

“…Systems based on wearable computing can be used to collect physiological data to this end, in a passive, non-obstructive manner that is comfortable and convenient for the PwD. One such system for identifying dementia-related difficulties is the BESI system, in which a wrist-worn accelerometer, the Pebble smartwatch, is utilized to track movements of the subjects to detect agitated behaviors [5][6][7]. In the BESI study, the participant was asked to wear the smartwatch for 30 days, with subject and caregiver dyad numbers ranging from 3 to 10 in each paper and study iteration.…”

Section: Introductionmentioning

confidence: 99%

“…Machine learning was then used on the data collected by the wearable to predict the occurrences of agitation. The researchers trained a number of models, such as support vector machines (SVM), adaptive boosting (AdaBoost), and an ensemble of decision trees by bagging (Tree-Bagging), with the latter providing the most robust prediction of agitated behaviors from unknown data [7]. Though the available literature from the study does not specify the stage of dementia experienced by the participants, it is likely from the community-based setting in which they were tracked that they still lived somewhat independently, or at least in their own home, implying mild to moderate dementia.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Challenges in Data Capturing and Collection for Physiological Detection of Dementia-Related Difficulties and Proposed Solutions

Harper

Ghali

Hussain

et al. 2021

Intelligent Computing Theories and Application

View full text Add to dashboard Cite

Dementia is a neurodegenerative disease which leads to the individual experiencing difficulties in their daily lives. Often these difficulties cause a large amount of stress, frustration and upset in the individual, however identifying when the difficulties are occurring or beginning can be difficult for caregivers, until the difficulty has caused problematic behavior or undeniable difficulty to the person with dementia. Therefore, a system for identifying the onset of dementia-related difficulties would be helpful in the management of dementia. Previous work highlighted wearable computing-based systems for analyzing physiological data as particularly promising. In this paper, we outline the methodology used to perform a systematic search for a relevant dataset. However, no such dataset was found. As such, a methodology for collecting such a dataset and making it publicly available is proposed, as well as for using it to train classification models that can predict difficulties from the physiological data. Several solutions to overcome the lack of available data are identified and discussed: data collection experiments to collect novel datasets; anonymization and pseudonymization to remove all identifiable data from the dataset; and synthetic data generation to produce a larger, anonymous training dataset. In conclusion, a combination of all the identified methods should ideally be employed in future solutions. Future work should focus on the conductance of the proposed experiment and the sharing of the collected data in the manner proposed, with data ideally being collected from as many people as possible with as many different types of dementia as possible.

show abstract

Multiple-Instance Learning for Sparse Behavior Modeling from Wearables: Toward Dementia-Related Agitation Prediction

Cited by 16 publications

References 12 publications

Smart Environments and Social Robots for Age-Friendly Integrated Care Services

Smart Environments and Social Robots for Age-Friendly Integrated Care Services

Wearable Respiration Monitoring: Interpretable Inference with Context and Sensor Biomarkers

Challenges in Data Capturing and Collection for Physiological Detection of Dementia-Related Difficulties and Proposed Solutions

Contact Info

Product

Resources

About