Projects
1. Navchetan Awasthi, "QSM Reconstruction using lp (non-convex norm) regularization", Medical Imaging.
Susceptibility-weighted imaging (SWI) is a magnetic resonance imaging (MRI) technique that enhances image contrast by using the susceptibility differences between tissues. SWI does not provide quantitative measures of magnetic susceptibility. This limitation is currently being addressed with the development of quantitative susceptibility mapping (QSM). Magnetic susceptibility is a physical quantity that measures the extent to which a material is magnetized by an applied magnetic field. Magnetic materials are classified as diamagnetic, paramagnetic, or ferromagnetic. Biological tissues can be either diamagnetic or paramagnetic depending on their molecular contents and microstructure. The dominant magnetization that contributes to magnetic susceptibility originates from orbital electrons. Enhancing susceptibility induced contrast by combining both magnitude and phase. Magnitude or phase by itself only uses half the acquired information. Amplitude and phase could be combined in a number of ways so that they enhanced each other, rather than canceled out. SWI could be used for enhancing gray matter / white matter (GM/WM) contrast, water/fat contrast, and identifying brain iron, thus extending its application beyond visualizing veins in the brain. The most severe and obvious artifact in the phase is the discontinuity caused by phase wrapping. Phase contributed by sources outside a region of interest is commonly referred to as the background phase. A common approach in using phase is to perform a phase unwrapping procedure followed by a highpass filtering operation. The difficulty of phase unwrapping can be further compounded by the low signal-to-noise ratio (SNR). Limitation of signal phase is that phase is nonlocal and orientation-dependent, thus not easily reproducible. The magnetic field at any given voxel is a superposition of all dipole fields generated by surrounding voxels. These techniques are generally called quantitative susceptibility mapping (QSM). Methods based on lp norm were developed and compared against the state of the art techniques in this project.
2. Navchetan Awasthi, J. Abhijith, "GPU Implementation of Implicit Runge-Kutta Methods", Parallel Programming.
Runge-Kutta methods are an important family of implicit and explicit iterative methods used for the approximation of solutions of ordinary differential equations. Explicit Runge-Kutta methods are unsuitable for the solution of stiff equations as their region of stability is small. Stiff equation is a differential equation for which certain numerical methods for solving the equation are numerically unstable, unless the step size is very small. This project aims in parallelizing implicit Runge-Kutta methods which are stable even for the stiff problems. The major disadvantage is that Implicit Runge-Kutta methods are more computationally expensive. We have shown how to accelerate the execution of implicit Runge-Kutta methods using GPUs.
3. Navchetan Awasthi, Venkata Suryanarayana K., "Texture based glioblastoma segmentation using MR images", Machine Learning for Signal Processing.
Glioblastoma multiforme (GBM), also known as glioblastoma and grade IV astrocytoma, is the most common and most aggressive form of cancer that begins in the brain. Signs and symptoms are initially non-specific. They may include headaches, personality changes, nausea, and symptoms similar to those of a stroke. Worsening of symptoms is often rapid. This can progress to unconsciousness. Accurate segmentation of GBM is very important and it helps in diagnosis, selecting appropriate therapies, and determining tumor regions in surgical planning. The task of segmentation is difficult, challenging, as well as time-consuming as the characteristics of GBM, are different and often deforms the nearby tissues. In this project, we have used magnetic resonance (MR) imaging sequences for accurate segmentation of GBM lesions in the brain. In this work, we have implemented the reference paper and compared the results against the proposed texture feature based segmentation. We have proposed the use of textons as image features for more accurate segmentation of GBMs.
4. Navchetan Awasthi, J. Abhijith, "Molecular Visualization and Bond Characterization", Data Analysis and Visualization.
Interaction between atoms play a very important role in determining the properties of a molecule. The understanding of these interactions is necessary for understanding the various chemical and biological systems. While we can extract the electron density to understand the chemical behavior of a molecule, non-covalent interactions are characterized by low electron densities. Even if the electron densities change slightly, it can significantly change the actual non-covalent interactions. Recently the two scalar fields, signed electron density and reduced gradient are playing an important role in visualizing these interactions even for complex molecules. This project aims in characterizing covalent and non-covalent interactions using topological analysis of signed electron density and reduced gradient.