Project 1 - I am currently involved in a collaborative project with Children's Hospital of Michigan in Detroit, MI. I am working on application of deep learning and machine learning algorithms on neuro- imaging data. In my current project, I apply graph representation learning algorithms on images obtained from diffusion-tensor imaging (DTI) scans of the brain of pediatric epilepsy patients. The goal of this project is to predict preoperative language impairment scores for pre-surgical evaluation which is a regression task and postoperative seizure outcomes 1 year after surgery which is a binary classification task. These graph representation algorithms are also used to identify the Seizure Onset Zones (SOZ's) in the brain which would help the neurosurgeon perform accurate resective surgeries. The stored weights obtained from the trained deep learning algorithms are used in identifying the regions of the brain that cause the epileptic seizures, by leveraging gradient-based saliency and class activation maps.

I worked with images obtained from Single Photon Emission Computed Tomography (SPECT) systems and developed machine learning and image processing algorithms to improve image quality by efficient design of Multipinhole (MPH) Collimator in the gamma camera in a combined Fan Beam and MPH Collimator SPECT system. The process involves estimation of the spatially-varying blur functions or point spread functions (PSFs) at pixel level, for the SPECT MPH system which would be used during system calibration and image reconstruction. The goal in this process was to store the least number of coefficients for the PSF and calculate most of the coefficients on the fly to make the iterative image reconstruction process faster.