Talks

PhD Proposal: Breath Analytics: A sensor-driven study on day-to-day human respiration

Faizan Wajid

5105 Brendan Iribe Center for Computer Science and Engineering (IRB)

Wednesday, March 30, 2022, 1:00-3:00 pm

You are subscribed to this talk through .
You are watching this talk through .
You are subscribed to this talk. (unsubscribe, watch)
You are watching this talk. (unwatch, subscribe)
You are not subscribed to this talk. (watch, subscribe)

Abstract

We believe that breathing reflects a variety of conditions of a person and can be used to detect physical as well as mental conditions. It is well understood that respiratory illnesses such as influenza or COVID- 19 affect our breathing, as do our various mental states, such as anxiety, depression, etc. However, breath characteristics have not been used for any diagnosis purposes in the past as the measurements used have not been detailed enough.

While most research centered around breathing focuses more on the signal properties, our approach puts a magnifying lens on the signal to study the physiology. This is enabled by the recent availability of sensors, but specifically the Spire Tag, a wearable sensor that makes measurements 25 times a second, allowing us to take a close look at the breathing process on a breath-by-breath basis throughout the day in the wearer’s normal environment. Our collaboration with the School of Public Health has given us access to a large cohort study that tracks the onset of illness and recovery. Since each participant was required to wear the Spire tag, we aim to identify the same from the sensor alone. This presents some unique challenges in using the raw data with all its physical characteristics and noises.

In the first part, we describe the process of transforming the raw data into a more usable form. This consists primarily of testing the Spire tag’s various sensors, but specifically the accelerometer and force sensors, and the various characteristics that could manifest, such as noise, drift, hysteresis, and so on.

In the second part, we outline the assumptions and relaxations we made targeting a single participant’s respiratory data while they are asleep. We chose this participant’s data as the person had been infected with the flu virus and had ample data for days around this period. From this, we select some days before and during onset of illness, and after recovery. We use this data set to develop the analysis techniques.

Finally, we describe some analytical results from our study of the feature space. Operating on individual breaths, we investigate how changes in these features are reflected in the feature space before carrying on to the task of classifying sick and non-sick breaths.

We conclude with presenting the proposed work with practical applications. Since our study so far has been focused around one participant, we aim to extend our analysis to more participants. Part of this also entails grouping the various types of respiratory events, such as deep and shallow breaths, coughs, sneezes, moments of apnea, etc. in attempts to develop a respiratory health index.

Examining Committee:

Chair:
Department Representative:

Dr. Ashok Agrawala
Dr. Aravind Srinivasan
Dr. Nirupam Roy

Bio

Faizan Wajid is a fifth-year PhD student in the Department of Computer Science at the University of Maryland, College Park. He is advised by Dr. Ashok Agrawala and is a member of the MIND Lab. His research focuses on sensors and signals, and developing technologies that improve well-being.

This talk is organized by Tom Hurst