Brain Health Model and Simulator
Description: Background The long-term goals of my research program are to understand, assess and predict brain health. In the last four years, I developed methodologies and metrological approaches to process neuroimages and extract biomarkers of interest. This work was developed by 13 HQP, disseminated in >29 peer-reviewed articles, and released as datasets or protocols, protected by six patents, and spun-off to a start-up. Objectives In this second phase, I propose to develop a brain health model and virtual simulator for personalized medicine. This requires solutions to multiple science and engineering problems, namely missing modalities; interpretability of simulator output; and a significant increase in data processing: -Objective 1: Palliate missing modalities via image synthesis. PET is a radiation-driven modality, minimally invasive, not ubiquitous, and costly, and is therefore not routinely performed. Synthetic PET images, carrying similar informational content, would reduce reliance on this modality. I propose to advance the field of image synthesis by using MRI and non-imaging data, together with deep learning, to generate synthetic FDG PET images; -Objective 2: Increase interpretability via image generation for visualization. The brain health model will not provide an image representation of predicted change on an individual basis. I propose a deep learning framework to generate future maps of brain health proxies (neuronal, metabolic, and cerebrovascular integrities; dendritic density) related to a change in the baseline condition of interest due to either cognitively healthy aging; neurodegeneration due to Alzheimer's; or non-pharmacological interventions; and -Objective 3: Manage a significant increase in data via a flexible data science infrastructure for image processing. Finally, our current database and image processing system, while significant (9,150 participants, >47,000 MRIs and >4,000 PET) needs significant overhaul. I propose a flexible data science infrastructure for large-scale image processing based on contemporary languages, a container architecture, and a structured image data format interoperable with Brain Canada's Canadian Open Neuroscience Platform. HQP training My research program will support one Research assistant and 11 students - five undergraduates; three M.Sc./M.Eng. candidates; and three Ph.D. candidate - in computer engineering (high performance computing), computer science (artificial intelligence, machine learning, simulations), computer vision (image processing, multiresolution images), and mathematics/statistics (modelling). This expertise is in high demand in several sectors: health, civil security, technological arts, finance, engineering, and entertainment. Impact Besides reducing our reliance on radiation-driven PET imaging, our model and simulator will directly support collaborators that devise interventions to change the decline trajectory in AD, and therefore achieve substantial societal impact.
Amount: CA$23,000
Funding type: research
Funder scheme: Discovery Grants Program - Individual
Start date: 2023-04-01
End date: 2024-03-31
Investigator: Simon Duchesne
Institution: Université Laval
Funded outputs count: 0
Funder award ID: rgpin-2019-06492
OpenAlex ID: G6938069486