Embedded Files
About SINEW Project
About SINEW Project
Sensors In-Home for Elder Wellbeing (SINEW) Project
SINEW is a longitudinal study in assessing senior adults' cognitive health, identifying mild cognitive impairment (MCI) from normal cognition (NC), and indicating aspects of daily lives causing the condition for timely intervention. As the first large scale study in Asia of its kind involving over 200 senior subjects (since year 2020) who live alone in their homes, SINEW leverages non-intrusive ambient sensor devices and wearables installed in the every participant's home to monitor and collect data about the individuals' daily activities and patterns of behavior.
Based on the sensor data collected from numerous households, digital biomarkers comprising aspects of daily life are extracted, wherein artificial intelligence (AI) and machine learning techniques can be applied to predict the participants' cognitive health.
Participants and Clinical Protocol
Participants in SINEW project were recruited from the community.
Community-dwelling seniors above 65 years, living alone, functioning independently, able to communicate in English and/or Mandarin.
Baseline screening and assessment applied at the start to determine initial cognitive health status, MCI or Normal Cognition (NC).
Yearly comprehensive assessments are conducted using clinically validated instruments measuring various aspects of health including functional, psychosocial, mental, and cognitive wellbeing. The results are used to label the data about the participants with their cognitive status so that a classification method or machine learning may perform.
System Configuration
System Configuration
Sensor Network and Data Collection
To collect the participants' daily activity data, a sensor network system, is installed in every participant's home.
Based on Internet of Things (IoT) architecture consisting of wireless sensor devices and an intelligent gateway,
Raw sensor readings data are transmitted to a cloud-based database via the in-house gateway device.
The kinds of sensor and wearable are chosen based on preference and privacy of the subject
some view that recording images, audio, or video data in their activities are still not preferable and can be seen as invasion of privacy
Motion Sensors
M-01, M-02, M-03
Contact Sensors
D-01, D-02
Dry Bed Sensor
+
Contact Sensor
D-03
Beacons
B-01 (Key)
B-02 (Wallet)
Wearable
SmartWatch
Gateway Set
Raspberry Pi, Dongles
Based on the data sent to the cloud-based server and database, the data can be further processed.
Monitoring the condition/status of sensor readings for system maintenance and participants' conditions
Applying Dashboard and data visualization supporting decision making to handle technical issues and deployment schedule
Extracting data features and digital biomarkers for analysis, prediction and machine learning
SINEW Sensor Network Architecture
Digital Biomarkers
To capture the high level information relating to the daily routines and activities of each participant, digital biomarkers are defined indicating different aspects of daily living including:
Physical. Tracking bodily movement using wearable sensors such as smartwatch.
Heartrate indicates daily average of heartbeat per minute in beat per minute (bpm)
Step count indicates the number of step taken for walking in the day
Activity. Tracking activity in relation to movement to spatial zones/locations at home and outside.
Transitions to bedroom indicates the number of movement from another room or zone to bedroom in the day
Transitions to kitchen indicates the number of movement from another room or zone to kitchen in the day
Transitions to living room indicates the number of movement from another room or zone to living room in the day
Outing indicates how many times the participant is going outside
Outing duration indicates the total duration of the participant spent outside in the day
Cognitive. Monitoring the participant's memory performance related to memory forgetfulness
Forgetting wallet indicates how many times the participant forget to bring the wallet (the wallet is left in the house) when going outside
Forgetting keys indicates how many times the participant forget to bring keys (the keys are left in the house) when going outside
Forgetting medication indicates how many times the participant does not take the medicine on time as prescribed
Sleep. Capture the statistics of sleep patterns of the participants
Sleep time indicates the time the participant go to sleep at nighttime
Wake time indicates the time the participant wake up in the morning next day
Sleep duration indicates how long, in minutes, totally the participant is sleeping at nighttime
Sleep interruption indicates how many times, during the night sleeping time, the participant awakes
Sleep interruption duration indicates the total duration when the participant is awaking several times during sleep time
Machine Learning for MCI Prediction and Discovery
The extracted biomarkers and the labels based on clinical assessments, a machine learning model is trained to detect MCI conditions, to extract classification rules, and to predict likelihood of MCI over time
The biomarker-based predictive model is implemented using Fusion ART employing a multi-channel neural network architecture. Prediction can be conducted through the bottom-up activation and competitive selection to select the most activated node before reading out the output layer.
The elements of the input vector in the input channel corresponds to the features in the daily biomarkers. The label or output channel corresponds to the cognitive status to predict (MCI or NC).
Page updated
Report abuse