Novel data augmentation with image mixing for machine learning-based classification of preclinical positron emission tomography images

We have developed a novel data augmentation procedure that can significantly enhance machine learning-based classification of different brain imaging scans. We will further expand this program to be used to create a novel animal model of brain diseases which will allow researchers to investigate earliest signs of diseases, the progression rates of which may vary across individuals. As a starter, we will look into the animal model of Parkinson’s disease. Parkinson’s disease, affecting approximately 100,000 Canadians, is the second most prevalent neurodegenerative disorder. Over half of patients develop levodopa-induced dyskinesia, a challenging motor side effect. We have recently demonstrated the rats that will develop this side effect shows different pattern of brain activities compared those that do not (i.e., resistant) even on the first day of levodopa treatment. With the proposed machine learning technology, we believe that we can accurately identify these “at-risk” animals, which will give researchers a new window of opportunities to investigate their brains under the microscope and discover novel therapeutic targets. This project will serve as a proof-of-concept for innovative imaging-based classification approaches in animal models, applicable to investigating early disease stages with varying progression rates.

Faculty Supervisor:

Ji Hyun Ko

Student:

Partner:

Cubresa Inc

Discipline:

Engineering

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Manitoba

Program: