Our group focuses on the following areas:

  • Multiparametric MRI image analysis: Our group has developed pipelines for quantitative and advanced MRI analysis and postprocessing using various diffusion MRI models (kurtosis, IVIM, stretched diffusion etc.), perfusion (DSC) and pharmacokinetic analysis. These features can be used for further analysis using machine learning and deep learning techniques.
  • Image registration: One of the major focus of the group has been in developing inline deformable image registration. Current state-of-the-art 3D deformable registration is iterative and takes a long time for multiphase or multi-timepoint datasets.  Current projects in the lab involve evaluating the use of deep learning to perform medical image registration using supervised as well as unsupervised methods.
  • Image segmentation: The group has expertise in developing segmentation models for various clinical translation medical imaging projects: prostate segmentation using domain transfer methods, aneurysm detection in MR angiography images, brain active tumor region, edema region and necrotic tumor segmentation. Lung nodule segmentation and classification using UNet and Attention Gate network.
  • Image super-resolution:  Our group has developed techniques for developing high isotropic 3D volumes for stack of thick 2D scans using cycleGAN deep learning technique which allows for data reformatting in any plane. This has greatly helped in prostate MRI visualization using these super resolution scans.  Another project currently going on in the lab entails using deep learning techniques for generating 7T-like images for epileptic foci detection from routine clinical 3T scans.
  • Image distortion correction and deblurring: The group has developed deep learning model for diffusion MRI distortion correction and deblurring using enhanced deep super resolution network. This would allow for retrospective image quality improvement of DWI scans from standard clinical scans.