Grant: Discovery Grants Program - Individual
University of Toronto
$37,000.00
Computational Methods for Rapid and Robust Magnetic Resonance Imaging
EF 2023-24
Résumé
The development of new computational tools and techniques has ushered in a new generation of imaging technologies. Alongside progress due to the development of new imaging hardware, or the exploitation of new physical interactions, significant advances in biomedical imaging are being made due to methods for extracting information and reconstructing images with missing or corrupted samples. This enables faster rates of data acquisition, and increased robustness to noise, image artifacts, or other unwanted image features. This proposal outlines a research program with the objective of developing new computational methods to enable more rapid and robust reconstruction of magnetic resonance imaging (MRI) data. Specifically, this program of research focuses on two primary objectives: ?Development of an accelerated multi-contrast imaging framework for diagnostic brain exams using MRI, and ?Development of a computational imaging platform for image-guided therapy using focused ultrasound. To achieve this, several computational technologies will be explored. The emerging field of deep learning for reconstruction of MR images aims to leverage flexible, highly parameterized neural networks for recovery of images from under-sampled, noisy, or corrupted data. Conventionally, this is performed in a fully- or semi-supervised manner, necessitating fully-sampled population datasets for model training. In contrast, self-supervised learning approaches significantly relax the training data requirements, allowing under-sampled raw MR data to be used for training, instead of high fidelity, fully-sampled datasets, which may not exist or would be impractical to acquire. Self-supervised learning strategies broaden the scope and applicability of deep learning in contexts where there is scarcity of ground truth data. Unlike neural network models, which capture complex non-linear properties of imaging data, structured low-rank methods exploit linear dependencies between samples within a given dataset. These approaches leverage image properties such as limited image support or smooth image phase variations to enable accelerated imaging (recovery of missing samples), as well as strategies for dealing with data inconsistencies (e.g. unwanted phase modulations between samples). Finally, multi-dimensional tensor models make use of data that have multiple dimensions (e.g. space, time, contrast, receive channel, etc.). These data typically contain signals or information that have redundancy, or correlations across the respective tensor dimensions and have been increasingly explored in MRI, particularly for highly resolved and under-sampled dynamic imaging. The proposed program of research focuses on the development of computational methods for biomedical imaging, working towards improvements to the speed and robustness of data acquisition in MRI enabled by new computational approaches for reconstructing images from raw measurements.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2023-03410
Name nom
Chiew, Mark
Department département
Physical Sciences Platform
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
37000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1505
Committee name en
Physics
Committee name fr
Physique
Area of application code
1102
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Biomedical engineering
Area of application fr
Génie biomédical
Research subject code
2514
Research subject group en
Electrical and electronic engineering
Research subject group fr
Digital signal processing
Research subject en
Génie électrique et électronique
Research subject fr
Traitement des signaux numériques
Application title
Computational Methods for Rapid and Robust Magnetic Resonance Imaging
Keyword
Biomedical Imaging; Computational Imaging; Deep Learning; Image Reconstruction; Low-Rank Tensor Models; Magnetic Resonance Imaging; Neuroimaging; Self-Supervised Learning
Application summary
The development of new computational tools and techniques has ushered in a new generation of imaging technologies. Alongside progress due to the development of new imaging hardware, or the exploitation of new physical interactions, significant advances in biomedical imaging are being made due to methods for extracting information and reconstructing images with missing or corrupted samples. This enables faster rates of data acquisition, and increased robustness to noise, image artifacts, or other unwanted image features. This proposal outlines a research program with the objective of developing new computational methods to enable more rapid and robust reconstruction of magnetic resonance imaging (MRI) data. Specifically, this program of research focuses on two primary objectives: ?Development of an accelerated multi-contrast imaging framework for diagnostic brain exams using MRI, and ?Development of a computational imaging platform for image-guided therapy using focused ultrasound. To achieve this, several computational technologies will be explored. The emerging field of deep learning for reconstruction of MR images aims to leverage flexible, highly parameterized neural networks for recovery of images from under-sampled, noisy, or corrupted data. Conventionally, this is performed in a fully- or semi-supervised manner, necessitating fully-sampled population datasets for model training. In contrast, self-supervised learning approaches significantly relax the training data requirements, allowing under-sampled raw MR data to be used for training, instead of high fidelity, fully-sampled datasets, which may not exist or would be impractical to acquire. Self-supervised learning strategies broaden the scope and applicability of deep learning in contexts where there is scarcity of ground truth data. Unlike neural network models, which capture complex non-linear properties of imaging data, structured low-rank methods exploit linear dependencies between samples within a given dataset. These approaches leverage image properties such as limited image support or smooth image phase variations to enable accelerated imaging (recovery of missing samples), as well as strategies for dealing with data inconsistencies (e.g. unwanted phase modulations between samples). Finally, multi-dimensional tensor models make use of data that have multiple dimensions (e.g. space, time, contrast, receive channel, etc.). These data typically contain signals or information that have redundancy, or correlations across the respective tensor dimensions and have been increasingly explored in MRI, particularly for highly resolved and under-sampled dynamic imaging. The proposed program of research focuses on the development of computational methods for biomedical imaging, working towards improvements to the speed and robustness of data acquisition in MRI enabled by new computational approaches for reconstructing images from raw measurements.