Multidisciplinary Institute in Artificial intelligence
Thomas Christen,
CRCN INSERM, MR physicist; GIN, Grenoble
thomas.christen@univ-grenoble-alpes.fr
Benjamin Lemasson
CRCN INSERM, biologist and data analysts; GIN, Grenoble
benjamin.lemasson @univ-grenoble-alpes.fr
We propose to develop the next-generation of MR exams by jointly designing new acquisitions, physics-informed AI reconstruction and image analysis. By employing AI tools at all stages of our new MRI pipeline, we will (1) enable new ways to pilot the scanners, thus producing large number of different parametric images tuned to pathophysiological characteristics (2) allow under-sampling strategies that dramatically shorten the acquisition times (<5min full exam) (3) take into account scanners variability and, (4) provide analysis tools, jointly optimized with the acquisition strategy, that integrate all the information and provide guidance for patient diagnostic and prognostic. We will thus simplify the current clinical imaging workflow by creating fast, robust, one click exams that provide all necessary information within one acquisition.
To start, we propose to tune our new MRI exam for the management of acute stroke patients that will not only be faster that actual exams but also provides access to new information that can be used to detect lesions and predict patient evolution. Because enhancing the acute stroke imaging workflow is inherently difficult, primarily due to time sensitivity which could delay treatment, we will minimize potential adverse effects on patient’s care, by conducting our studies first in silico followed by preclinical MRI using rodent models of stroke and then process to clinical settings with healthy volunteers. Our project will create a unique dataset, which will include the acquisition of the same optimized MRF sequences for stroke lesions in both healthy and pathological animals, as well as in healthy human subjects. Such a dataset will allow a more fundamental and exploratory investigation of the transferability of observations made from animal data to human data using AI-based causality models.
Training activities: As part of the MIAI chair, we will develop theoretical and practical training modules designed to teach AI-based tools for medical imaging. We are considering two components: one for initial education (all medical students in the first years of their training as well as medical residents specializing in medical imaging) and another aimed at professionals (AI for medical image as a culture and AI for medical imaging as a job).
License:
