Vegh Group

Research and development of technologies to improve the diagnosis, intervention, and prevention of health conditions.

The Vegh Group creates technologies to make measurements, or use already collected measurements to make inferences, in relation to health conditions. .

The Group aims to produce quantitative information using technologies founded on mathematical methods or data driven approaches, including machine learning and deep learning. The primary technologies of interest for the Group have been those linked with human medical imaging (primarily, MRI, PET, SPECT, CT). More recently, the Group has extended their effort into other types of data, including medical records, genetics, amongst others.  

For more than a decade, the Vegh Group have been developing low field MRI instrumentation with the aim of making MRI technology more accessible, portable, all at a lower cost. In parallel they have designed and implemented medical imaging protocols to improve clinical imaging workflow and the information which can be extracted from images. Topics of interest include tissue microstructure imaging using MRI, non-traditional relaxation and diffusion processes in MRI, and data driven methods for information mapping. Techniques employed within the Group have now been extended from mathematical foundations to machine learning and deep learning. The Group is increasingly focusing on how our methods can be applied for prevention of disease, not just at the time of intervention.  

Research Areas

  • Functional MRI and the use of naturalistic stimuli to create brain activation maps.
  • Diffusion MRI for inferring tissue microstructure properties. 
  • Machine learning applied to medical images and other types of medical data.
  • Deep learning and its application in medical imaging.
  • Classification of disease states including seizure recurrence in epilepsy.
  • PET image reconstruction.
  • Radiomics feature based classification.
  • Hand crafting of features for machine learning. 
  • Explainable AI. 
  • Low field MRI instrumentation.
  • Brain asymmetry and its relation to disease. 
  • Electromagnetic noise suppression.
  • Cross-modality medical image synthesis using machine learning and depp learning. 
  • Healthspan prediction from large multifaceted datasets. 
  • Radiotherapy treatment planning using machine learning.
  • Identification of dysplastic tissue using machine learning.
  • Automated segmentation in medical images and histology using machine learning. 
     

Research Approach

The Group's goal is to create quantitative information. Thus, information extracted from data (medial images, clinical information, genetics, etc.) can be compared across sites, and can be used to classify disease state. For example, we parameterise images with the aid of mathematical models (i.e., tissue microstructure imaging) and by using data driven approaches (i.e., kinetic model parameter estimation in PET). The Group is unique because their approaches rely heavily on analytical / data driven frameworks, and the data they use is generally multifaceted and often not interpretable in its acquired form. The Group not only develops the frameworks, but they also collect their own data. This allows the Group to modify the framework and data collection process to achieve an optimal outcome. The use of machine learning allows us to scale to large data sets, where intricate patterns most likely elude the observer.

Funding

Since 2019

  • ARC Training Centre for Innovation in Biomedical Imaging Technology - MRI technology development in collaboration with industry partners Siemens Healthcare.
  • Cancer Council Queensland Project Grant - esophageal cancer, radiomics, machine learning and deep learning.
  • BiomedTech Horizons 2.0 - portable low field MRI hardware, instrumentation and protocol development.
     

Group members