~~(The talk will be given in English)

Speaker:   Dr. Ilya Soloveychik
                       School of Computer Science and Engineering, Hebrew University

Monday, February 29th, 2016
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering

Group Symmetric Covariance Estimation and Detection

Abstract
We focus on the problem of Gaussian and robust covariance estimation under group symmetry constraints. We assume the true estimated matrix to commute with a finite unitary matrix group, which is referred to as the group symmetry. Examples of group symmetric structures include circulant, persymmetric, proper quaternion and many others. We develop the group symmetric versions of the sample covariance and Tyler's estimators and determine their performance benefits. In particular, the classical results claim that at least n = p and n = p+1 sample points in general position are necessary to ensure the existence and uniqueness of the sample covariance and Tyler's estimator respectively, where p is the ambient dimension. We significantly improve these requirements for both estimates and show that in many cases even 1 or 2 samples are enough to guarantee the existence and uniqueness regardless of p. We build a unified framework for group symmetric estimation and explain the nature and consequences of the underlying block-diagonalization phenomenon. In addition, we develop a group symmetric detection framework and investigate the gains it yields.

~~Speaker: Ori Weber
M.Sc. student under the supervision of Prof. Nahum Kiryati

Wednesday, February 24th, 2016 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering

Robust registration between high-field preoperative and low-field intraoperative brain MRI

Abstract

Preoperative high-resolution brain MRI is commonly used to produce high quality anatomical MRI as well as diffusion tensor imaging trachtography and functional MRI. While preoperative high-resolution MRI is very useful for lesion detection and surgery planning, it has only a limited validity as an anatomical reference for navigation during open brain surgery. When the skull is opened (craniotomy), at the beginning of surgery, cerebrospinal fluid leaks out, causing the brain to drift progressively with regard to the skull. As surgery progresses, local deformations caused by ongoing tissue resection further increase the discrepancy between preoperative MRI and the actual brain anatomy.

During the last decade, low-field (.15T) intraoperative MRI (iMRI) has been developed to provide updated MRI images during brain surgery, thereby providing real-time images of the brain for improved navigation. While low-field intraoperative MRI provides some degree of anatomical guidance during surgery, it cannot replicate the functionality, in terms of resolution, signal-to-noise, field-of-view and the ability of diagnostic systems to provide fMRI and DTI information. Therefore, the ability to quickly map information obtained from the preoperative images directly to the intraoperative environment, while taking into account brain-shift and tissue resection, would provide significant improvement in image quality, leading to higher accuracy of navigation during surgery.

This thesis proposes a robust, automatic and real-time framework of coupling high-field preoperative and low-field intraoperative MRI, and projecting preoperatively acquired fMRI and tractography maps during the ongoing surgery. In the proposed method, open brain surgery is analyzed as a succession of three consecutive phases: pre-craniotomy, post-craniotomy and post-resection. Each phase has a tailor-made pipeline which explicitly takes into account its respective challenge: the limited intraoperative MRI field-of-view, the brain-shift effect, and the extracted brain tissues following resection. Quantitative validation results are presented on a set of 20 real interventional cases.