University of Memphis
AutoSense: Quantifying Exposures to Addictive Substances and Psychosocial Stress
AutoSense is an unobtrusively wearable wireless sensor system for continuous assessment of personal exposures to psychosocial stress and addictive behaviors as experienced by individuals in their natural environments. AutoSense consists of a chest band with six wireless sensors and an arm band with four wireless sensors. The Autosense chest band includes:
- A 2-lead ECG,
- A respiratory inductive plethysmograph (RIP) band for measurement of relative thorax volume and respiratory frequency,
- A 3-axis accelerometer to assess motion and physical activities,
- A galvanic skin response (GSR) assessment at ECG electrodes,
- Skin temperature sensors, and
- Ambient temperature sensors.
All six sensors are hosted on a small (1’’x 2.5’’) circuit board and are packaged in a custom plastic enclosure with a 700 mAh Li-Ion battery. The arm band consists of a transdermal WrisTAS alcohol sensor, a 3-axis accelerometer, a skin temperature, and a GSR sensor. The armband package measures 1” square and uses a small 210 mAh Li-Ion battery.
Smart Phone Data Collection
Both sensing units wirelessly transmit measurements from all 10 sensors to an Android smartphone using a low-power ANT radio and will last a week on a single battery charge. Sensors on the body are complemented by additional sensors on the phone (e.g., GPS, accelerometer). The phone also acts as a local server for heavier computation & storage and for collecting self-report.
Smart Phone Software
Measurements from the wireless sensors are processed on the smart phone to compute over 100 features from various sensors (e.g., heart rate, heart rate variability, etc.) in real-time. These features are used in machine learning-based algorithms to infer various psychosocial states and behaviors. These include measures of stress (from ECG and RIP), conversation episodes (from RIP), location such as home, work, etc. (from GPS), and posture and physical activity (from accelerometers).
Micropayments can be associated with each self-report prompt to encourage compliance. In addition, self-report can be solicited based on automatically detected behavioral events (e.g., stress, conversation), based on self-reported events, or time.
The mobile software can be used as effectively in a lab session. It provides an elaborate study-coordinator user interface that can be used to annotate various lab phases, which will be time-synchronized with the continuous measurements received from the sensors placed on the subjects. Finally, the entire software suite on the mobile phone can be customized for a scientific study by defining the measures to be collected in a configuration file.
The chest band and the mobile phone software have been used on 50+ participants for 2,000+ hours in the field in two separate studies. The automated assessment of stress has been evaluated in a 20+ subject under both lab and field conditions. In the lab, it predicted physiological stress with 90% accuracy and in the field had a median correlation of 0.71 with self-reported stress level.
To further evaluate the assessment of stress, and to evaluate the automated assessment of conversation, alcohol usage, drug usage, smoking, and craving for illicit drugs, AutoSense is being used in 4 ongoing user studies. In one study, 30 smokers and social drinkers in Memphis wear the entire suite of sensors for one week in the field. They mark smoking and drinking events and provide self-reports 20 times daily. In the second study, 20 drug users from an ongoing study at NIDA wear the chest band for 4 weeks in the field and in stress and craving sessions in the lab. In the field, participants mark craving, drug usage, and smoking events and provide self-reports 10 times daily. In the third study, 10 drug users in a residential facility at Johns Hopkins wear the chest band during self-administration of drugs. In the fourth study, healthy adults wear the chest band and a sensor suite from BioPac during a lab stress session to assess the validity of AutoSense sensors.
Upon completion of the ongoing user studies, AutoSense will be ready for use in a variety of scientific studies that involve automated and continuous assessment of stress, drinking, smoking, craving, drug usage, social interactions, and physical activity in both lab and field settings. We plan to make the wearable sensors commercially available and release the mobile phone software as open source under the name of “FieldStream”. See the project webpages for additional information ( http://sites.google.com/site/autosenseproject/ and http://www.fieldstream.org/ ).