Phantom Project (Courtois-Neuromod Project)

General objectives

The first long-term goal is to train an AI capable of general intelligence (a computational model of brain function called "Phantom") by training a handful of individuals in a key cognitive domain for learning, i.e. memory. Thus, it is possible to force an artificial network (deep learning) to adopt an architecture close to the biological neural network during learning to generate appropriate responses to a series of stimuli, such as associating semantic categories with images, and to predict brain activity.

The second objective is to demonstrate the utility of artificial networks driven with biological networks to understand the functioning of the human brain. As part of the memory component of the project, the focus of interest is Alzheimer's disease, a neurological disorder associated with marked cognitive (mostly memory) alterations.

The hypothesis is that neuroimaging data can constrain the network architecture, with great potential for flexibility, generalization, and quick adaptation to new tasks, even though training is a simple task.

An example of future application is the realization of in-silico (virtual) cortical ablations to perform lesion studies. In addition, longitudinal data collection of this magnitude is of incomparable value, even using a more traditional data analysis approach. The number of measurement points per subject, as well as the duration of the participants' follow-up, make it a valuable resource for studying the variability of cerebral anatomical functional architecture at different time scales.

Memory project component

This component of the project is co-directed by Dre Sylvie Belleville and Dr Pierre Bellec. 15 images (14 unique and 1 repeated target 1 time) are presented to the subjects who must recognize if the target image was presented to them during the test and recognize the spatial position of the image (remember which quadrant is it? appeared the first time). The training (occurring weekly during 5 years) of the Phantoms (N = 4) for this task is done on a large base of complex images compiled by Samie-Jade Allard and I in order to reproduce the cognitive processes of encoding, memorization (short-term: after a block of 15 images, long-term: after 1 hour of training) as well as visual representation in healthy subjects. From Phantoms, the association between artificial activity profiles and biological responses is tested to see if it differs between control subjects and patients with Alzheimer's disease, and which systems are involved. To do this, a large cohort of elderly patients with subjective (N = 150) or mild (N = 100) cognitive impairment, as well as control subjects (N = 50) are monitored every two years by the Consortium for Early Identification of Alzheimer's Disease-Quebec (CIMA-Q), co-directed by Dr. Belleville. This cohort will also be followed for the 5 years of the Courtois-Neuromod project.