Machine learning tools for mHealth data analysis focus of new project

A University of Massachusetts Amherst computer scientist is embarking on a project to build machine learning-based tools for analyzing large-scale mobile health and clinical data.

The effort is funded by a five-year, $536,527 National Science Foundation Faculty Early Career Development award and will be a collaborative effort with scientists, clinicians and researchers at the University of Memphis, Yale University School of Medicine and Children's Hospital Los Angeles. Each institution is providing access to mHealth and clinical data, according to an announcement.

"Electronic health records are seeing wide adoption across the United States and we're starting to see the emergence of large stores of complex clinical data as a result," Benjamin Marlin, of UMass Amherst, states in the announcement. "There's significant interest in leveraging these data to enhance all kinds of clinical decision support tools with the hope that they can ultimately improve quality of care."

The project will include analyzing data, such as physiological measurements, collected via mHealth wearable sensor systems.

Wearable mHealth devices are being used in a variety of healthcare scenarios. One of the most recent is a project launched by chip maker Intel and the Michael J. Fox Foundation, which will use wearable devices and data analytics to monitor treatment of patients suffering from Parkinson's disease. Another new wearable device and app may soon provide greater independence to ALS patients given a proof of concept that taps EEG brainwaves to command electronic devices via a wearable display, a tablet and software. 

In many situations, the data being collected is what Marlin describes as "noisy" data due to factors such as high velocity, heterogeneity and incompleteness.

"We're not dealing with nice, clean data in these areas," Marlin says in the announcement. "The data are noisy, parts are missing due to sensors disconnecting or clinicians not recording measurements. A number of these issues can break current data analysis methods. The goal of this work is to design new machine learning-based data analysis tools that are significantly more robust and accurate."

To learn more:
- read the announcement

Related Articles:
Wearables to play starring role in Parkinson's disease research effort
Docs: Wearable devices a possible game changer in medical care
For wearables to take deeper root, user incentives a must
3 areas where the biosensing wearables market needs to expand