Graduate Seminar Series: Clinical Stream – Michael Tisi

Graduate Seminar Series: Clinical Stream
Graduate Seminar Series for the Institute of Biomedical Engineering (BME). This day is for clinical stream presenters.
If you would like to invite your Principal Investigator, please add their email via the ‘Add Guest’ button and they will also be notified of your presentation.
Presentation Title: Machine Learning Assisted Pulmonary Function Assessment via Intrabreath Oscillometry
Abstract:
Intrabreath oscillometry is a pulmonary function test that enables the dynamic tracking of mechanical changes in the respiratory system by measuring airway impedance. Raw volume, pressure, and flow measurements from intrabreath oscillometry tests were input to a novel classification architecture to resolve one of the following physiologies: normal, restrictive, obstructive, or mixed obstructive-restrictive functionality. The MiniROCKET algorithm was first used to generate features from input time series
data and train a ridge regression classifier model. When used for inference, output classifier scores are utilized in a soft voting scheme to acquire a final, confident prediction of pulmonary function based on all valid intrabreath oscillometry trials for
a given subject. The proposed classification architecture was able to distinguish between normal, restrictive, and obstructive pulmonary function with state of the art accuracy; however, it struggled, at first, to differentiate between restrictive and mixed obstructive-restrictive physiologies. After identifying and removing anomalous training examples and defining a region of uncertainty instead of a rigid decision boundary for the classifier, the proposed architecture was also able to distinguish between restrictive and mixed obstructive-restrictive pulmonary function with state of the art accuracy.
Supervisor Name: Chung-Wai Chow; Shahrokh Valaee
Year of Study: 2
Program of Study: PhD
Zoom link: https://us02web.zoom.us/j/89610372821?pwd=azd4SCtYVWtreVovaGNPV1c2NGY2Zz09
Meeting ID: 896 1037 2821
Password: 483329
Powered by Calendly.com