Research Interests

My research interests primarily include areas at the intersection of inverse problems, statistical signal processing, machine learning, and optimization. More recently, I have been focusing on the applications of deep learning for solving inverse problems that frequently arise in medical imaging applications such as computed tomography, magnetic resonance imaging, etc. I take interest in designing novel deep neural network architectures and learning strategies that are particularly suited for image reconstruction problems. Although I primarily focus on medical imaging applications, the learning paradigms I develop often transcend the application at hand and apply to more general computer vision tasks. I am also keenly interested in developing robust and scalable optimization algorithms with theoretical guarantees for high-dimensional signal estimation problems with structural constraints such as sparsity and low-rank.

 

I am currently working on developing a deep learning-based framework for CT-to-MRI synthesis without exploiting any pixel-wise correspondence between the scans in two modalities during training. The approach assumes that each pair of scans in the training database originates from the same brain region of the subject being imaged. Given this minimal pairing information, I seek to find latent structural similarities between two modalities to stabilize the learning framework based on generative adversarial networks (GANs). The primary motivation behind this line of work is to computationally extract the soft-tissue contrast from the CT scans for diagnosis without requiring the patients to undergo expensive MRI scanning simultaneously. Inter alia, I am working on combining sparse dictionary learning with generative modeling for medical imaging applications.