Grant: Discovery Grants Program - Individual
University of Manitoba
$36,000.00
Physic Based Deep Learning for Microwave Imaging
FY 2024-25
Summary
Quantitative Microwave Imaging (MWI) employs electromagnetic waves to reconstruct the physical properties and internal structures of an object or region of interest. It has been used for a broad range of imaging applications, such as breast cancer detection and treatment monitoring, subsurface prospecting, and stored-grain monitoring. For biomedical applications MWI presents unique advantages, but there still remain the challenges of it producing lower resolution images compared to other imaging modalities, as well as the production of reconstruction artifacts, related to the nonlinearity and ill-posedness of the associated inverse problem. Recently, Deep Learning (DL) techniques have been used to address these challenges and enhance the resulting images. Current research indicates that DL approaches can be employed to successfully overcome these difficulties. However, the use of DL techniques relies heavily on the availability of large volumes of labelled data, a scarcity in MWI. This limitation often results in the creation of DL techniques that lack robustness and generalizability, and this limits the applicability of MWI for practical use.
My proposed research program will investigate developing semi-supervised techniques that are guided by physical and mathematical information about the MWI process. Unlike supervised DL approaches that require large amounts of labeled data, semi-supervised DL techniques have been designed to discover unknown patterns in data with a limited labelled dataset. This research program will pursue four objectives to address these challenges and create robust learning-based MWI. First, a large labelled dataset of 3D numerical breast permittivity phantoms will be created using generative models to address the current lack of realistic phantoms. Second, MWI reconstruction results will be improved using state of the art DL models to reduce low-resolution and remove imaging artifacts. Third, a robust DL-based MWI inverse solver will be developed to capture direct raw MW scattered-field data and optimization-based solvers will be eliminated. Finally, specialized physics-based features and DL-based calibration models will be used to make the DL inverse solver more robust when inverting experimental data.
The proposed research will impact the applicability of MWI by developing robust deep learning techniques to overcome the lack of generalizability, which is the primary obstacle to using learning-based techniques for MWI. The proposed research will facilitate the transfer of learning-based MWI from theoretical research concepts to practical applications. Learning-based MWI will not only enhance the MWI resolution and reduce the reconstruction artifacts, but it will reduce the computational cost and potentially provide real-time MWI, leading to the development of new MWI applications. The proposed research will support the training of 3 MSc, 2 PhD, and 5 undergraduate students.
Program
Discovery Grants Program - Individual
Recipient
Location
Manitoba, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-04894
Name nom
Khoshdel, Vahab
Department département
Electrical and Computer Engineering
Organization id
19
Institution établissement
University of Manitoba
Province en
Manitoba
Province fr
Manitoba
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
36000
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
1510
Committee name en
Electrical and Computer Engineering
Committee name fr
Génie électrique et informatique
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
2520
Research subject group en
Electrical and electronic engineering
Research subject group fr
Génie électrique et électronique
Research subject en
Microwave and millimeterwave devices, circuits and technologies
Research subject fr
Dispositifs, circuits et technologie hyperfréquences et d'ondes millimétriques
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Physic Based Deep Learning for Microwave Imaging
Keywords
microwave imaging; inverse problem; physic-based deep learning; machine learning; conditional diffusion model; biomedical imaging; breast cancer imaging
Application summary
Quantitative Microwave Imaging (MWI) employs electromagnetic waves to reconstruct the physical properties and internal structures of an object or region of interest. It has been used for a broad range of imaging applications, such as breast cancer detection and treatment monitoring, subsurface prospecting, and stored-grain monitoring. For biomedical applications MWI presents unique advantages, but there still remain the challenges of it producing lower resolution images compared to other imaging modalities, as well as the production of reconstruction artifacts, related to the nonlinearity and ill-posedness of the associated inverse problem. Recently, Deep Learning (DL) techniques have been used to address these challenges and enhance the resulting images. Current research indicates that DL approaches can be employed to successfully overcome these difficulties. However, the use of DL techniques relies heavily on the availability of large volumes of labelled data, a scarcity in MWI. This limitation often results in the creation of DL techniques that lack robustness and generalizability, and this limits the applicability of MWI for practical use.
My proposed research program will investigate developing semi-supervised techniques that are guided by physical and mathematical information about the MWI process. Unlike supervised DL approaches that require large amounts of labeled data, semi-supervised DL techniques have been designed to discover unknown patterns in data with a limited labelled dataset. This research program will pursue four objectives to address these challenges and create robust learning-based MWI. First, a large labelled dataset of 3D numerical breast permittivity phantoms will be created using generative models to address the current lack of realistic phantoms. Second, MWI reconstruction results will be improved using state of the art DL models to reduce low-resolution and remove imaging artifacts. Third, a robust DL-based MWI inverse solver will be developed to capture direct raw MW scattered-field data and optimization-based solvers will be eliminated. Finally, specialized physics-based features and DL-based calibration models will be used to make the DL inverse solver more robust when inverting experimental data.
The proposed research will impact the applicability of MWI by developing robust deep learning techniques to overcome the lack of generalizability, which is the primary obstacle to using learning-based techniques for MWI. The proposed research will facilitate the transfer of learning-based MWI from theoretical research concepts to practical applications. Learning-based MWI will not only enhance the MWI resolution and reduce the reconstruction artifacts, but it will reduce the computational cost and potentially provide real-time MWI, leading to the development of new MWI applications. The proposed research will support the training of 3 MSc, 2 PhD, and 5 undergraduate students.