The Laboratory of Medical Information Processing (LaTIM) is offering a PhD position on Radiation Imaging (PET, CT) Tomographic Reconstruction using Neural Network Approaches.
Thesis Location: Laboratory of Medical Information Processing (LaTIM), French Institute of Health and Medical Research (INSERM UMR 1101), Brest, France
Thesis Supervisors: Didier BENOIT, Dimitris VISVIKIS
Period: 3 years, starting on October 2018
During the last decade, dose reduction during a PET or CT examination has become a crucial issue. This reduction strongly impacts the results from the tomographic reconstruction. Indeed, iterative reconstruction algorithms are based on a noise model, with the statistical quality of the acquired data (number of detected photons) having a significant impact in the reconstructed image quality which can in turn strongly influence patient management. In the past, many different denoising techniques have been proposed but over the last couple of years new algorithms from the visual computing field (segmentation, object detection and super-resolution) using neural networks have been introduced in nuclear [1-2] and CT imaging [3-4]. These techniques have gained ground within the context of tomographic reconstruction given the computational power offered by dedicated hardware such as GPUs (Graphics Processing Unit) and multi-CPU processors.
The aim of this thesis is to propose new iterative reconstruction methods using neural networks for radiation based imaging.
Within this context different training databases (clinical and/or simulated raw datasets and images) will be constructed and evaluated for the training of neural networks using existing libraries such as Tensorflow or Caffe. Different implementations can be envisaged based on the combination of raw datasets (sinograms/projections) and corresponding reconstructed images both in the field of PET and CT imaging for training convolutional neural networks (CNN, ). Simulated datasets using GATE  or GGEMS  will be used. The performance of the proposed approaches will be finally compared with current state of the art in iterative image reconstruction in PET and CT imaging. The open source platform CASToR  dedicated to the generic tomographic reconstruction will be used within this context.
The candidate must hold a Master degree in Physics, Computer Science or Applied mathematics. Experience on neural networks and/or tomographic reconstruction programming would be appreciated but not required
Scientific Interests: Tomographic reconstruction, numerical simulation, neural networks
Programming Skills: C/C++, OpenCL (or CUDA)
Languages: English, French optional
Send CV, cover letter, grades/marks (Master, License/Bachelor) by e-mail to: firstname.lastname@example.org and email@example.com
 Litjens G., Kooi T., Bejnordi B. E., Setio A. A. A., Ciompi F., Ghafoorian M., van der Laak J. A. W. M., van Ginneken B. and Sánchez C. I. “A survey on deep learning in medical image analysis.” arXiv 2017
 Gong K., Guan J., Kim K., Zhang X., Fakhri G. E., Qi J. and Li Q. “Iterative PET image reconstruction using convolutional neural network representation.” arXiv 2017 preprint arXiv:1710.0334.
 Xu J., Gong E., Pauly J. and Zaharchuk, G. “200x Low-dose PET Reconstruction using Deep Learning.” arXiv 2017 preprint arXiv:1712.04119.
 Kang E., Min J. and Ye J. C. “A deep convolutional neural network using directional wavelets for low‐dose X‐ray CT reconstruction.” Medical Physics 2017, 44(10)
 Jain V., Seung H., “Natural image denoising with convolutional networks,” in Proc. Adv. Neural Inf. Process. Syst. (NIPS, 2008), pp.769–776.
 Jan S, et al “GATE: a simulation toolkit for PET and SPECT” Phys Med Biol. 2004; 49(19):4543-61
 Bert J, Perez-Ponce H, El Bitar Z, Jan S, Boursier Y, Vintache D, Bonissent A, Morel C, Brasse D, Visvikis D. “Geant4-based Monte Carlo simulations on GPU for medical applications.” Phys Med Biol. 2013, 58:5593-5611
(c) GdR 720 ISIS - CNRS - 2011-2018.