Speaker: Oren Dovrat

M.Sc. student under the supervision of Prof. Shai Avidan

Sunday, December .30^th, 2018 at 15:30

Room 011, Kitot Bldg., Faculty of Engineering

Learning to Sample

Abstract

Processing large point clouds is a challenging task. Therefore, the data is often sampled to a size that can be processed more easily. The question is how to sample the data? A popular sampling technique is Farthest Point Sampling (FPS). However, FPS is agnostic to a downstream application (classification, retrieval, etc.). The underlying assumption seems to be that minimizing the farthest point distance, as done by FPS, is a good proxy to other objective functions. We show that it is better to learn how to sample. To do that, we propose a deep network to simplify 3D point clouds. The network, termed S-NET, takes a point cloud and produces a smaller point cloud that is optimized for a particular task. The simplified point cloud is not guaranteed to be a subset of the original point cloud. Therefore, we match it to a subset of the original points in a post-processing step. We contrast our approach with FPS by experimenting on two standard data sets and show significantly better results for a variety of applications.

Speaker:  Yuval Tamir

M.Sc. student under the supervision of Hedva Spitzer

Wednesday, December 26^th, 2018 at 15:30

Room 011, Kitot Bldg., Faculty of Engineering

Companding algorithm for enabling detection of malignant lesions in HDR CT mandibular images
 

Abstract

            Compressing and expanding (companding) HDR computerized Tomography (CT) images of X-ray to a single window is still a relevant challenge for several body organs, even though different approaches have been examined. Besides the general need for simplification of the window setting method for diagnosis requirements, there are clinical needs for resection of malignant lesions in the mandible, for example. In this kind of organ (mandible) the lesions often exist in both the soft tissue and the bone. A suggested algorithm has to take into account the requirement to expose differently the same range of specific grey levels, which is presented at different body tissues.

We propose an adaptive multi-scale contrast companding (AMCC) algorithm. A contrast appears stronger or weaker according to its own value and according to its context contrast, at the different spatial resolutions. The AMCC algorithm successfully compands, with adaptive set of parameters, a large variety of mandibular CT HDR images and natural images as well. Two collaborating physicians evaluated the tumor boundaries, by the single presentation method in 108 mandibular CT window setting and algorithm output images. They found that 92% of all algorithm output images showed at least the same evaluations as the window method did, while 50% of the algorithm output images even yielded improved evaluations. When the evaluation of each slice was done simultaneously, by using both window setting (bone and soft tissue) and algorithm output images, 93% of the evaluations showed preference for the algorithm’s output images. We suggest here, for the first time, a low-cost method for companding the HDR images and ability to perform resection with optimal boundaries definition of mandibular lesions.