Projects

Non-Rigid Wound Image Registration System

Capturing standard images of wounds is challenging since wounds can appear in many places on the body and are not necessarily well visualized in standard views. Clinicians usually center a digital image of the at-risk area or wound in the field of view but variations in distance, orientation, rotation make comparison of wound imagery over time extremely difficult. We develop a system for non-rigid image registration. Achieving this aim will require developing machine learning models that extract image features from visible, infrared, and near infrared images to learn the spatial transformations across longitudinal wound images. Incorporating this system within the multispectral imaging device will greatly benefit medical practitioners that provide care and patients that need wound care.

Non-Invasive Methods for Detecting Human Physiological and Psychological Conditions

We develop a machine learning model that uses video images of the human face to pick-up signals and analyze these signals to determine physiological conditions. There are a multitude of applications for this research, from non-invasive measurement of clinical data (such temperature, blood pressure) to lie detection that could be used by law enforcement.

Reduced Order Modelling for Friction Stir Welding

Friction stir welding is a new method to fuse metals that are otherwise difficult to join. This project explores reduced order modelling for friction stir welding. The parameterization of the process requires complex computer simulations and human intuition. We propose to simplify the process by developing a mathematical model to optimize the friction stir welding process. Furthermore, we are interested in exploring how these methods may be applied to problems in other unrelated domains such as health, energy materials, and/or nanophotonics.

Pulmonary Ventilator Perfusion Mismatch

Develop a mathematical model and methodologies for evaluating pulmonary ventilation perfusion mismatch. Ventilation/perfusion mismatch occurs in pulmonary diseases such as ARDS and COPD. While low tidal volume ventilator strategy has shown a reduction in mortality rate of ARDS, there is no consensus among clinicians on other ventilator settings for ARDS patients. Mathematical models or ARDS provide useful insights into the condition but they are not readily applicable to investigate ventilation/perfusion mismatch. We are interested in deriving a measurement of the intrapulmonary shunt of a lung by using inhale and exhale gas fractions. We would like to be able to identify differences in ventilation/perfusion mismatch induced by dead space and/or intrapulmonary shunt in the lung.