By continuing to browse the site, you are agreeing to our use of cookies. (More details)
Portal > Offres > Offre UMR5220-DAVSAR-003 - CBCT image simulation for deep learning in prostate radiation therapy (H/F)

Post-doc: CBCT image simulation for deep learning in prostate radiation therapy (H/F)

This offer is available in the following languages:
Français - Anglais

Ensure that your candidate profile is correct before applying. Your profile information will be added to the details for each application. In order to increase your visibility on our Careers Portal and allow employers to see your candidate profile, you can upload your CV to our CV library in one click!

Faites connaître cette offre !

General information

Reference : UMR5220-DAVSAR-003
Date of publication : Tuesday, June 23, 2020
Type of Contract : FTC Scientist
Contract Period : 12 months
Expected date of employment : 1 September 2020
Proportion of work : Full time
Remuneration : about 2109 à 2228€ gross per month
Desired level of education : PhD
Experience required : Indifferent



External beam radiation therapy (EBRT) is a key step of the reference treatment of pelvic cancers, the most common cancers in the male population, especially with prostate cancer but also rectum and bladder cancers. However, its efficiency is hampered by the large deformations occurring between the treatment fractions, while the treatment plan is precisely optimized according to a fixed anatomy. The goal of the DELPEL project is to develop a full workflow to monitor the dose during EBRT.

This project will exploit the Cone-Beam CT (CBCT) images acquired by Image-Guided RT (IGRT) and take advantage of the recently developed deep learning (DL) approaches. The main challenge will be to identify an optimal workflow combining the data available in clinical routine and the involved processing steps, especially image simulation, segmentation and registration. The first task (WP1) of this project will be to compose a large database of images, including planning CT images with the associated delineations and dose distributions, and CBCT images. Since the delineation of the CBCT images is not realized in clinical routine, is very time consuming and difficult, our idea is to simulate realistic CBCT images from CT images that will thus be associated with reference delineations. Using this database, the second task will be dedicated to the segmentation of the images. Deep learning approaches will be investigated and validated using real and simulated delineated images. The third task will be to develop the dose monitoring process, thanks to two main steps: (i) deformable image registration (DIR) between the CBCT images and the planning image; (ii) daily dose calculation from IGRT images. For both these steps, deep learning approaches should enable to improve the state-of-the-art methods.


The recruited people will participate to WP1.

In order to create a meaningful training database, contours on both CT and CBCT are needed. However, in clinical routine CBCT contours are usually not available. Here, we will use some CBCT images that were delineated by an expert, but only on 23 patients, with more than 150 delineated CBCT. Moreover, the uncertainty associated with the CBCT contours is relatively high as the image quality in the pelvic region is usually relatively bad. The management of the complete database will be performed at CLB, based on previous expertise in image DB management.
To increase the number of cases of the database as well as to dispose of reliable ground truth, CBCT images will be simulated from CT images. Indeed, our group is experienced in the CBCT simulation of the Elekta Synergy device and we developed hybrid techniques allowing to efficiently simulate X-ray projections corresponding to the geometry and physical characteristic of the beam-detector CBCT device. The simulation method will be composed of a first analytical part simulating the primary portion of the radiography projections, like a conventional DRR (Digitally Reconstructed Radiograph); this is usually relatively fast, thanks to a GPU implementation in the RTK toolkit [Rit2014]. However, to allow more realistic simulations that include similar image noise, all scatter photon should be taken into account. For that part, we plan to use advanced Monte- Carlo and Fixed Forced Detection (FFD) method that we recently developed for SPECT images [Cajgfinger2018, Sarrut2014] and has been shown to be efficient for X-ray also [Arbor2015]. One difficulty will be to properly take into account the influence of bow-tie filter that adds a certain quantity of scatter, not really known for the moment. Methods developed in [Zöllner2017] may help. Hence, we hope to be able to generate highly realistic CBCT image from the planning CT. Because images will be simulated, the CT contours could be projected onto the CBCT images and be used as reference. We plan to generate simulated CBCT images for more than 100 patients cases treated at CLB.


Medical image processing and simulation, C++, python

Work Context

Lab CREATIS. The team is located at Léon Bérard cancer center

Constraints and risks


Additional Information


We talk about it on Twitter!