Team members

Faculty
James M. Rehg (Georgia Tech)
Benjamin Marlin (UMass Amherst)
Deepak Ganesan (UMass Amherst)
Emre Ertin (Ohio State)
Santosh Kumar (Memphis)
Susan Murphy (Michigan)

Students
Roy Adams (UMass Amherst)
Ju Gao (Ohio State)
Walter Dempsey (Harvard)
Nazir Saleheen (Memphis)

Big Data Analytics

MD2K’s goal is to create generalizable theory, methods, tools, and software that addresses the major barriers to processing complex mobile sensor data. This is to enable the use of this data by the broader community for biomedical knowledge discovery and just-in-time care delivery. MD2K’s Data Science Research Core seeks to demonstrate the feasibility, utility, and generalizability of the MD2K approach by implementing the entire MD2K data analytics system in the context of two biomedical applications - reducing relapse among abstinent daily smokers and reducing readmission among congestive heart failure (CHF) patients.

All of the tools and software developed by MD2K are freely available as open-source projects for mHealth and data science researchers. Biomedical researchers will be able to install the MD2K software on mobile devices to collect mobile sensor data and the MD2K analytics software on their servers to analyze these data for biomedical discovery.

Computational Model Development
comp model

Specific goals of the DSR Core include:
 • Develop general principles, computational methods, and a toolbox for inferring markers (i.e., measures) of patient health, as well as markers of the behavioral, physical, social, and environmental risk factors that are often found in a wide variability of behaviors. In addition, develop methods to address known and unknown confounders, errors in self-report data, and the variable quality and availability of sensor data.

• Develop time series pattern mining algorithms and interactive visualization tools to help biomedical researchers accurately discover vulnerable states from sensor-based markers. This is done with the goal of being able to predict adverse health events, ahead of the onset of adverse clinical events, and to develop online learning algorithms for delivering just-in-time adaptive intervention that can move an individual back to a healthy state.

• Develop and implement a standards-based, interoperable, extensible and open-source big data software platform for efficient implementation of MD2K data analytics developed in the previous two points. This supports both off-line analysis of mobile sensor data at population-scale and online data analysis at the individual scale, and provide a reliable and responsive user experience to biomedical researchers, patients, and care providers.

• Conduct user studies with smokers, CHF patients, researchers, and clinicians to inform the design of MD2K tools and to evaluate their utility, usability, and validity for biomedical research and care delivery.

Some of the major activities that MD2K has undertaken to reach the above-stated goals include:

• Detection of smoking from respiration and wrist-worn accelerometers
•Detecting craving from respiration and ECG data
• Detection of eating events using wrist-worn accelerometers
• Use of iShadow glasses for detecting drowsiness and fatigue
• Automated detection of visual smoking cues using first person vision
• Analysis of compressively sampled ECG data
• Structured prediction models for multivariate time series data with temporally imprecise labels
• Use of a Discovery Dashboard to discover and visualize predictors for adverse events.
• Predicting risk for adverse health events from marker data
• Learning decision rules for timing and content of just-in-time adaptive interventions.

Publications

 

 

 

Copyright © 2017 MD2K. MD2K is supported by the National Institutes of Health Big Data to Knowledge Initiative (Grant #1U54EB020404)

Team: Cornell Tech, GA Tech, Harvard, U. Memphis, Northwestern, Ohio State, Open mHealth, UCLA, UCSD, UCSF, UMass, U. Michigan, Utah, WVU