Numerical Modeling and Simulation of Inverse Problems in Medical  Imaging

Part of DROITE ANR project

and ARC6 "Modélisation et Simulation Numérique de Problèmes Inverses en Imagerie Médicale"

supported by the labex CAMI, PRIMES, PERSYVAL and the ECCAMI project.

10/09/2013 and 11/09/2013

in UJF Grenoble,

at LJK accès / directions

(cf. IRMA CAMPUS  SITE, direction: practical information)

Participant list

Registration is free but mandatory / Inscription gratuite mais obligatoire

Program

Tuesday 10 September

8h45 : Welcome, (LJK  TOUR IRMA, salle 1) in Grenoble
9h00 Introduction to pinhole system calibration. Application to the calibration of a C arm
. . . . . Benjamin Spencer, TIMC-IMAG, Grenoble.

9h45 2D fan-beam CT with independent source and detector rotation
. . . . . Simon Rit, CREATIS-CLB-ESRF, Lyon and Grenoble

10h30 Break
11h00 About efficient sampling in tomography

Alvin Ihsani, Mc Master University and TIMC-IMAG

11h30 Introduction to ROI tomography (summer school type presentation)

Laurent Desbat, TIMC-IMAG, Grenoble

12h30 to 14h00 Lunch
14h00 Compressed sensing methods for cardiac C-arm computed tomography

Cyril MORY, PHILIPS and CREATIS, 20-30mn

14h30 In-room breathing motion estimation from limited projection views using a sliding deformation model

Vivien Delmon, CREATIS, Lyon

15h00 About hybrid imaging

Faouzi Triki, LJK, Grenoble.

16h00 Break
16h30 Data Consistency Conditions for Linograms and Planograms

Rolf  Clackdoyle,  LHC, St Etienne.

20h00 Possible dinner at a nice place.

Wednesday 11 September

9h30 Reconstruction tomographique pour la caméra Compton,

Voichita Maxim, CREATIS, Lyon

10h30 Break
10h45 Bone microstructure reconstruction from few projections with binary tomography

L.Wang, B.Sixou, F.Peyrin, CREATIS, Lyon.

11h30-13h30 Lunch
13h30 3D PET list-mode reconstruction including all information provided by the detector

Julien Bert, PhD, CHRU Brest, LaTIM - INSERM

14h15 Simultaneous Reconstruction of the Activity Image and Registration of the CT image in TOF-PET

Ahmadreza Rezaei,  Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

15h00 Spatially Variant Resolution Modelling for Iterative List-Mode PET Reconstruction

Matthew Bickell,  Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

15h45 Break

16h15 A modified likelihood distribution for bias reduction in low-statistics PET

Katrien Van Slambrouck, Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

17h00 Parallel data processing on GPU and CPU using OpenCL

Koen Michielsen, Johan Nuyts, Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

This DROITE workshops are supported by a grant from << Région Rhône Alpes>> (ARC 6, T.I.C. et Usages Informatiques Innovants)

This workshop is supported by the labex CAMI, PRIMES, PERSYVAL, and the ECCAMI project. DROITE is supported by ANR (ANR project ANR-12-BS01-0018).

Tuesday 10 September

8h45 : Welcome, (LJK  TOUR IRMA, salle 1) in Grenoble

9h00 Introduction to pinhole system calibration. Application to the calibration of a C arm

Benjamin Spencer, TIMC-IMAG, Grenoble.

9h45 2D fan-beam CT with independent source and detector rotation

Simon RIT, CREATIS-CLB-ESRF, Lyon and Grenoble

We investigate the fan-beam situation where the source and the detector can rotate independently around the object. In this case, the detector can be tilted with respect to the typical situation where the detector perpendicular to the line defined by the source and the center of rotation. We re-visit the work of Gullberg and Crawford with displaced center of rotation and add an additional possibility to let the distance between the source center of rotation vary.

10h30 Break

11h00 About efficient sampling in tomography

Alvin Ihsani, Mc Master University and TIMC-IMAG

11h30 Introduction to ROI tomography (summer school type presentation)

Laurent Desbat, TIMC-IMAG, Grenoble

We first recall basic results of CT: the Radon transform, the central slice theorem, the FBP formula for parallel geometry, the fan-beam geometry and the associated FBP formula. We then present a possible road through 2D ROI CT: reconstruction from limited angle and truncated projections (ex. the interior problem) ; the Hilbert Projection Equality, Short Scan, Less than a Short Scan and Virtual Fan Beam reconstruction ; Differentiated BackProjection and Hilbert transform, DBP method.

12h30 to 14h00 Lunch

14h00 Compressed sensing methods for cardiac C-arm computed tomography

Cyril MORY, PHILIPS and CREATIS, 20-30mn

In cardiac C-arm CT, electrocardiogram gating leads to a limited view reconstruction problem. Few projections are available to reconstruct each phase, and their angular distribution is not optimal. In such conditions, traditional reconstruction algorithms like SART or FDK prove insufficient. Compressed sensing allows using a priori information in the reconstruction, potentially compensating for the loss of information caused by ECG-gating. This talk will be a quick tour of the 3D compressed sensing methods available for cardiac C-arm CT and of the 3D+time methods developed during my PhD thesis.

14h30 In-room breathing motion estimation from limited projection views using a sliding deformation model

Vivien Delmon, CREATIS, Lyon

15h00 About hybrid imaging

Faouzi Triki, LJK, Grenoble.

My talk  will focus on multi-waves or hybrid imaging modalities that combine the best imaging properties of different types of waves (optical waves, pressure wave, electric wave, elastic wave, etc). These imaging modalities overcome the limitations (resolution, contrast, etc) of classical imaging techniques using one particular type of waves like EIT, MRI,etc. I will give a review of recent mathematical results in the field.

16h00 Break

16h30 Data Consistency Conditions for Linograms and Planograms

Rolf  Clackdoyle,  LHC, St Etienne.

1/ Helgason Ludwig range conditions in 2D (well known DCC for the Radon transform)

2/ Linograms (link with the parallel geometry, DCC for Linogram geometry and link to the HL DCC for the Radon Transform)

3/Linogram Fan Beam: complete DCC.

4/ Planograms and complete DCC.

20h00 Dinner together  at a nice place...


Wednesday 11 September


9h30 Reconstruction tomographique pour la caméra Compton,

Voichita Maxim, CREATIS, Lyon

The Compton camera is an imaging device dedicated for SPECT, having superior sensitivity compared to the Anger camera.
Pre-clinical imaging in nuclear medicine and on-line therapy control in hadron-therapy are some of the foreseen applications.
The projections measured by the camera are integrals of the intensity of the source on conical surfaces.
The first purpose of this talk is to present a review of the models that were proposed in the literature to describe the direct problem.
The second is to quickly introduce some of the analytical reconstruction techniques. Examples of reconstructed images will also  be shown.

10h30 Break

10h45 Bone microstructure reconstruction from few projections with binary tomography

L.Wang, B.Sixou, F.Peyrin, CREATIS, Lyon.

11h30-13h30 Lunch

13h30 3D PET list-mode reconstruction including all information provided by the detector

Julien Bert, PhD, CHRU Brest, LaTIM - INSERM

List-mode based image reconstruction is continuously gaining ground in Positron Emission Tomography (PET) imaging. Such reconstruction has many benefits, as the preser- vation of the temporal sampling of the acquired data that facilitates the study and correction of dynamic processes, as well as maintaining the highest spatial sampling available for a given detector geometry. The centerpiece of the reconstruction process is the projector. This projector computes on-the-fly the contribution of a line-of-response for each voxel of the field-of- view considering detector effects. In this study, we propose a new projector that incorporates all available information provided by the detector to improve the reconstructed image accuracy. This projector includes geometric and detector scatter effects, time-of-flight and depth-of-interaction. As the computation burden is a main obstacle to obtain such reconstruction, we implemented the reconstruction including detector corrections on graphic processing units (GPU). Results showed that our projector provides reconstructed images with a high accuracy (low noise, high contrast and resolution).

14h15 Simultaneous Reconstruction of the Activity Image and Registration of the CT image in TOF-PET

Ahmadreza Rezaei,  Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

Abstract: Previously, maximum-likelihood methods have been proposed to jointly estimate the activity image and the attenuation image (or the attenuation sinogram) from TOF-PET data. In this contribution, we propose a method that addresses the same problem for TOF-PET/CT by combining reconstruction and registration. The method, called MLRR, iteratively reconstructs the activity image while registering the available CT-based attenuation image, so that the pair of activity and attenuation maximizes the likelihood of the TOF emission sinogram. The algorithm is evaluated on 2D and 3D simulations and on a clinical data set.

 
15h30 Spatially Variant Resolution Modelling for Iterative List-Mode PET Reconstruction

Matthew Bickell,  Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

Abstract: In PET imaging accurately modelling the physical measurement process to account for the point spread function (PSF) can improve the subsequent resolution of the reconstruction. During MLEM iterative reconstruction this may be accomplished by convolving the image with a suitable Gaussian distribution before forward projection and after backprojection. This Gaussian is, however, spatially invariant and is only an accurate model for the resolution near the centre of the field-of-view (FOV) of the PET scanner, and becomes less accurate with increasing radial distance. In this work a spatially variant model has been developed which models the two most dominant factors affecting the point spread function, namely the crystal response function and the acollinearity of the photon pair. This model is incorporated directly into the list-mode data by randomly shifting the lines-of-response (LORs) according to the derived distributions, after which a list-mode MLEM reconstruction is performed. I will demonstrate that this model agrees with measured PSFs, and I will present results from phantom studies when using this new model compared to those when using the Gaussian convolution model. An improved resolution is observed, although along with an increase in noise. This work will be applied to motion correction such that the resolution for each LOR can be accurately modelled before they are corrected for motion.

16h15 A modified likelihood distribution for bias reduction in low-statistics PET

Katrien Van Slambrouck, Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

Abstract: PET data are typically reconstructed with maximum likelihood reconstruction methods. However, these methods suffer from positive bias due to the nonnegativity constraint in the reconstruction and the strong asymmetry of the Poisson distribution near zero. We developed a modified maximum likelihood method with a combined Poisson and Gaussian distribution. The Poisson distribution is replaced with a Gaussian distribution for very low counts, for higher counts the Poisson distribution is still used. The resulting reconstruction algorithm is an extension of the NEGML algorithm. This new NEGML method reduces bias significantly with comparable results to other bias reduction methods. Moreover, the parameter tuning for the algorithm is rather easy.

17h00 Break

17h30Parallel data processing on GPU and CPU using OpenCL

Koen Michielsen, Johan Nuyts, Nuclear Medicine and Medical Imaging Research Center, K.U.Leuven.

Abstract: Image reconstruction contains many operations where the same instructions are applied to different parts of the data. This type of problems is ideally suited to be solved in parallel on off-the-shelf graphics processing units (GPU). The two prevalent programming interfaces for general purpose computing on GPU are CUDA (NVIDIA Corporation) and OpenCL (Khronos Group). The latter has a significant advantage in flexibility since it is not limited to one vendors hardware, but will run on any GPU, CPU or other processing unit and operating system with supported hardware drivers. In our case this allows to run the same implementation on both a GPU on a Windows 7 workstation and a multicore CPU Linux cluster.
This presentation will show the work in progress on parallelizing the different aspects of our iterative reconstruction methods, such as distance driven projection and backprojection, image (or sinogram) convolution and smoothing priors. We will concentrate on showing the implementation details of the device code and benchmarks.

This DROITE workshops are supported by a grant from << Région Rhône Alpes>> (ARC 6, T.I.C. et Usages Informatiques Innovants)

This workshop is supported by the labex CAMI, PRIMES, PERSYVAL, and the ECCAMI project. DROITE is supported by ANR (ANR project ANR-12-BS01-0018).