Mathews JACOB is currently an assistant professor at the Departments of Biomedical Engineering, ECE and Imaging Sciences (formerly Radiology) at the University of Rochester. His research interests include image reconstruction, image analysis and quantification in the context of a range of modalities including magnetic resonance imaging, near-infrared spectroscopic imaging and electron microscopy.
He was born on 16th of June, 1975 in Kerala, India. He obtained his B.Tech in Electronics and Communication Engineering and M.E in signal processing from the National Institute of Technology, Calicut Kerala in 1996 and the Indian Institute of Science, Bangalore in 1999 respectively. He was granted his Ph.D degree from the Biomedical Imaging Group at the Swiss Federal Institute of Technology in 2003. He was a Beckman postdoctoral fellow at the University of Illinois at Urbana Champaign between 2003 and 2006.
“Model Based Algorithms for Biomedical Imaging”
The focus of the lecture will be on constrained model-based algorithms for two important biomedical applications (1) the reconstruction of magnetic resonance spectroscopic imaging (MRSI) data and (2) the estimation of global shape from biomedical images. Although these problems appear unrelated, our solutions to them are very similar; the model-based framework is the underlying theme. This framework enables the combination of diverse set of priors with the available data, thus constraining the problems. The resulting algorithms are coherent and consistent, thus eliminating performance loss over standard multi-step schemes.
The ﬁrst part of the talk will start with the basics of MRSI, its clinical applications, the state of the art methods and their limitations. I will then introduce our model based algorithm that overcomes many of the problems associated with the current methods. Specifically, this algorithm combines the available anatomical and ﬁeld-inhomogeneity priors, derived using conventional MRI, to constrain the reconstructions. The performance improvement in using this scheme will be demonstrated using simulation and in-vivo experimental data. The computational challenges in automating the derivation of the MR priors and integrating the model-based algorithm with fast acquisition schemes such as echo-planar spectroscopic imaging (EPSI) and parallel imaging will then be discussed. To overcome these challenges, I will introduce (a) an algebraic method to derive the anatomical and ﬁeld-map priors automatically from the MRI data (b) a novel non uniform FFT interpolation kernel to account for the non-uniform EPSI trajectory and (c) an exact algorithm to derive the coil sensitivities in parallel MRI. The second part of the talk will start with a brief introduction of the basic steps involved in shape estimation: (1) local feature detection and (2) estimation of the global shape. As the first step, a method to design the optimal steerable template for the detection of a specified local image feature (eg. edge/ridge) will be introduced. I will then present the standard parametric active contour model to combine the local information to estimate the global shape. The snake model uses image energy to drive the curve to the object of interest, while it uses an internal energy term to enforce curve smoothness. In contrast to the standard image energy that rely on pre-detected local features (eg. gradient magnitude), a new energy that directly use the image information using steerable detectors will be introduced; in addition to providing more consistent results, this energy is also parameter-independent. Since the smoothness of the curve model is dependent on the parameterization, a new internal energy term that forces the snake to stay in the curvilinear abscissa during the evolution will be introduced. This algorithm will then be adapted for a specific application: 3-D reconstruction of DNA ﬁlaments from stereo cryo-electron micrographs. The rigorous validation of the resulting reconstruction algorithm using simulated and experimental data-sets will also be presented.