ISP Seminars

Previous Seminars

Implementation Aspects of the Random Demodulator for Compressive Sensing
Pawel Jerzy Pankiewicz (AAU, Aalborg, Denmark)
26 April 2013
The compressive sensing paradigm has gained significant attention from the applied sciences and engineering in recent years. The random demodulator is one of the compressive sensing architectures providing efficient sub-Nyquist sampling of sparse band-limited signals using simple off-the-shelf components.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Friday, 26 April 2013 at 11:15 (45 min.)

Statistical models of the spine with applications in medical image processing
Dr Fabian Lecron (UMons, Belgium)
17 April 2013
Statistical shape models are useful tools for the extraction of an object in a given image. In this respect, this presentation introduces two novel ways of modeling shapes with a particular interest for the representation of the spine in biomedical applications. The characteristic of the first proposed model is to represent the link existing between several items composing a global shape. It is particularly the case for the spine where the vertebrae are linked together. The advantage of the second model is that no assumption on the statistical distribution of the sample is required. It is based on One-Class Support Vector Machine and on the definition of an hyperplane that holds the data of a sample in its positive side. Part of the presentation is dedicated to the use of these models to extract the shape of the spine in 2D or in 3D. The aim of the first application is to automatically segment cervical vertebrae on radiographs in order to compute orientation angles and to assess the mobility of the cervical spine. The second application concerns the 3D reconstruction of the spine from bi-planar radiographs to assess the scoliosis of a patient with 3D clinical indices.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 17 April 2013 at 14:00 (45 min.)

4D signal processing for lung cancer treatment by radiotherapy
Nicolas Gallego-Ortiz (ICTEAM/ELEN, UCL)
10 April 2013
Cancer treatment by radiotherapy relies on accurate patient positioning. The aim of radiotherapy is to target, with a high energy X-ray beam, cancerous cells while sparing surrounding healthy tissue. This requirement is especially hard to accomplish for lung tumors. The thorax is subject to respiratory motion, heart beating and deformations due to organ fillings. Tumor position uncertainty implies the addition of margins to the target volume. The largest the margins the more of healthy tissue is irradiated reaching even organs risk. Research to assure treatment quality and safety to moving tumors is currently very active. Efforts range from treatment planning and delivery protocols, to technological developments of dynamic imaging modalities. The development of 4D Computed Tomography during the last decade opened the door to image-guided therapy for lung tumors. At the same time, tumor tracking in real-time is still not feasible in most of the cases, thus there are still many open issues. In this talk 4D signal processing is going to be discussed in the context of radiotherapy. 4D signals (3D + time) come from a variety of imaging modalities implemented in the treatment room. The state of the art of imaging and tumor motion models is presented and a discussion on research questions in the field is proposed.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 10 April 2013 at 14:00 (45 min.)

A Deconvolution Problem in Astronomy
Adriana Gonzalez
27 March 2013
Optical sensors distort the observation of an object by the imperfection of their elements and by the natural physics of the observation. The acquired image is frequently corrupted by the noise coming from the sensor itself and by the Point Spread Function (PSF) of the instrument, which filters the hypothetically pure image. In most situations, we do not have any knowledge on this PSF, hence we face a blind deconvolution problem when estimating the pure image.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 27 March 2013 at 14:00 (45 min.)

A virtual tour of free viewpoint rendering
Cédric Verleysen (ICTEAM/ELEN, UCL)
18 December 2012
A principal limitation of the conventional video technology is that the viewpoint is fixed by the camera that captures a scene. Since the last decade, researchers have tried to overcome this limitation by allowing a remote viewer to interactively navigate across the scene. The potential applications of such free viewpoint rendering are tremendous, ranging from 3D TV to scene analysis, and from interactive media, augmented reality to video content/game production, etc.
Ressources :

Shannon Seminar Room (a105) -- Tuesday, 18 December 2012 at 15:00 (45 min.)

Compressed Sensing in RF Communication and Analog-to-Digital Conversion
Thomas Arildsen (TPS/DES, Aalborg U., Denmark)
30 November 2012
Compressed sensing has experienced explosive growth in interest from the research community in recent years. It is currently being investigated in numerous, typically signal processing-related areas.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Friday, 30 November 2012 at 11:15 (45 min.)

Dictionary learning methods for single-channel source separation
Augustin Lefèvre (INMA/ICTEAM, UCL)
29 November 2012
We present three main contributions to single-channel source separation. Our first contribution is a group-sparsity inducing penalty specifically tailored for nonnegative matrix factorization (NMF) with the Itakura-Saito divergence: in many music tracks, there are whole intervals where at least one source is inactive. The group-sparsity penalty we propose allows identifying these intervals blindly and learn source specific dictionaries. As a consequence, those learned dictionaries can be used to do source separation in other parts of the track were several sources are active. These two tasks of identification and separation are performed simultaneously in one run of group-sparsity Itakura-Saito NMF.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Thursday, 29 November 2012 at 10:45 (45 min.)

Measuring the brain connectivity
Maxime Taquet (ICTEAM/ELEN, UCL)
16 November 2012
For decades, brain diseases have been studied in terms of abnormalities in particular regions. However, neurologists and psychiatrists faced challenges when attempting to elucidate the causes and mechanisms of some mental illnesses that do not seem to be bound to a particular locus. Among them, autism and schizophrenia are now identified as diseases which affect the connections -rather than the regions- of the brain. During this seminar, you will learn how brain connections can be detected, characterized and compared, based on medical images. Recent advances in brain imaging and connectivity modeling, as well as their applications to population studies will be presented.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Friday, 16 November 2012 at 14:00 (45 min.)

A 3D MRI segmentation method and a 2D/3D X-ray image registration method for orthopaedic applications
Taha Jerbi (ELI, UCL)
16 October 2012
First, I present a new segmentation method of 3D MRI images that evaluates the necrotic area in the femoral head. This method is based on PDE equations and a dynamic robust estimator of the noise and the irrelevant contours. I show the advantages of our results by using entropy based evaluation technique. Second, I present a registration method in frequency domain between 2D and 3D data. This method is used to determine the positions of the bones during flexion (pseudo static). It is well suited to the acquisition system which produces 3D reconstructions of the bones in the initial standing position and only 2D frontal and sagittal radiographic acquisitions for other positions.
Ressources :

-- Tuesday, 16 October 2012 at 10:45 (45 min.)

Proximal methods for Poisson Intensity CBCT and PET
Prof. Yannick Boursier (Université Aix-Marseille 2 / ESIL) (invited talk)
2 October 2012
Cone-Beam Computerized Tomography (CBCT) and Positron Emission Tomography (PET) are two complementary medical imaging modalities providing respectively anatomic and metabolic information of a patient. In the context of public health, one must address the problem of dose reduction of the potentially harmful quantities related to each exam protocol : X-rays for CBCT and radiotracer for PET. Two demonstrators based on a technological breakthrough (acquisition devices work in photon-counting mode) have been developed and we investigate in this paper the two related tomographic reconstruction problems. We formulate separately the CBCT and the PET problems in two general frameworks that encompass the physics of the acquisition devices and the specific discretization of the object to reconstruct. These objects may be observed from a limited number of angles of views and we take into account the specificity of the Poisson noise. We propose various fast numerical schemes based on proximal methods to compute the solution of each problem. In particular, we show that primal-dual approaches are well suited in the PET case when considering non differentiable regularizations such as Total Variation. Experiments on numerical simulations and real data are in favor of the proposed algorithms when compared with the well-established methods and has contributed to establish the first results for spectral CT.

-- Tuesday, 2 October 2012 at 10:45 (55 min.)