Mergelas Award Recipients Announced 

Ph.D. Student at the Institute of Medical Science (IMS), University of Toronto
Supervisor: Dr. Farzad Khalvati

My Ph.D. thesis is focused on identifying molecular biomarkers of pediatric low-grade glioma (pLGG) using MRI and Artificial Intelligence (AI), as an alternative to the invasive procedure of biopsy. Additionally, I emphasize reproducible research, translational medicine, and Human-AI interaction. To that end, I have introduced OpenRadiomics, which provides the research community with the largest and most comprehensive open-source AI-ready radiomics data. OpenRadiomics also proposes a reproducible research protocol, stressing generalizable training and evaluation pipelines instead of individual trained models. The vision of my research is to make the pipelines end-to-end and extend them beyond pLGG.

MSc Candidate at the Department of Computer Science.  Supervisor – Dr. Farzad Khalvati

I started my academic research when I was in my third year of undergraduate career. My first research project with my supervisor Dr. Yanglei Song is on the Contextual Multi-Armed Bandit Problem, which is a very classical problem in traditional reinforcement learning. We introduce a novel algorithm “Truncated Linear UCB” and provided an optimized regret bound. I have also worked on the Medical AI research project on the Prostate Cancer (PCa) Detection using MRI Data at Queen’s University. We develop a novel pipeline for PCa classification by using multi-modality, multi-domain MRI data and incorporating uncertainty information in the classification model. When the model deploys to real clinical settings, radiologists will know the predicted label from the model as well as how confident of the model in its prediction. At the IMICS Lab, I am currently working on to improve the classification performance on the Pediatric Low-Grade Gliomas (pLGG) subtypes with limited data. I hope the model can produce more accurate labels to radiologists during the diagnostic process. 

MSc Candidate 
Supervisor - Dr. Farzad Khalvati

Research Overview: Segmentation of the boundaries of regions of interest (e.g., tumour) is a leading problem in medical imaging. Using Machine Learning (ML) for this problem generally requires manually annotated ground-truth segmentations, demanding extensive time and resources from radiologists. Weakly supervised segmentation simplifies the manual effort to assigning a binary image-level label to each image, indicating whether the image has the anomaly of interest. My research investigates a novel weakly supervised method for segmenting brain tumours in 2D Magnetic Resonance (MR) images that leverages the image-level labels to acquire two different sources of information that lead to the segmentations. The first source is from explanation maps, which can be generated from classifier models trained using binary image-level labels. The second source of information is acquired using non-cancerous variants of cancerous MR images generated using a deep convolutional generative adversarial network (DCGAN). The combination of these two priors enables weakly supervised segmentations that significantly outperform the state-of-the-art results for 2D MR images.