Research Interests

In my PhD, the focus of my research was on machine learning and medical imaging and recently I have become interested in bioinformatics and in particular its conjunction with medical imaging. The objective of my research is to simultaneously exploit wealth of information in medical image and human genomics to improve understanding of human disease via building statistical and computational methods. Here is a list of my recent manuscripts (see My CV for a complete list):

Joint Modeling of Imaging-Genetics

There are two main purposes for combining imaging and genetic information: 1) using imaging as intermediate information (so-called intermediate phenotype) to understand underlying biological processes of a disease, 2) using known genetic markers of the disease to interpret observed anatomy in the imaging data. Here are some of my published works toward those achieving goals:

  • K.N. Batmanghelich, A.V. Dalca, M.R. Sabuncu, P. Golland. Joint Modeling of Imaging and Genetics, In Proc. IPMI: International Conference on Information Processing and Medical Imaging, LNCS 7917, pp. 766-777, 2013. (Oral Presentation)

Point Process for Diverse Feature Selection

This work is related to the previous projects. The goal is to find a diverse subset of features that are optimal for a task (e.g. classification or regression) while being diverse with respect to some given side information. We formulate the problem as learning the quality scores of items in a Determinantal Point Process (DPP).

Dimensionality Reduction for Medical Imaging Applications

This project addresses the large dimension, low sample size problems (p>>n) in the medical imaging applications and more specifically for the diagnosis of the Alzheimer's disease. The goal is to transform very high-dimensional input to a low-dimensional representation that preserves discriminative signal and the resultant representation is clinically interpretable. We formulate the task as a constrained optimization problem that combines generative and discriminative objectives. We propose a novel large-scale algorithm to solve the resulting optimization problem.

(Software website)

Matrix Decomposition for Deformation Modeling and its applications in Computational Anatomy

Given a set of normal and abnormal brain images, in this project we try to solve an inverse problem: how to decompose a brain deformation to a normal and abnormal deformations from an average brain. The deformations are parameterized by stationary velocity fields and the velocity fields are decomposed to low-rank and sparse components. The algorithm iterates between image registration and the decomposition step and hence can be viewed as a group-wise registration algorithm.