French
English
Scientific Context
Compton cameras, due to the absence of a mechanical collimator, can provide better sensitivity than Anger cameras which
are the standard detector employed in nuclear medicine for SPECT scans. This results in a reduced patient dose or a faster
acquisition. Their application in nuclear medicine was first proposed by ([1]). Since then, the interest in Compton cameras
has steadily grown. Recently, the application of these systems for targeted radiotherapy ([2], [3]) or as multimodal PET /
Pinhole / Compton camera systems ([4], [5]) has been considered. However, the uncertainties in measurements caused by
physical effect, such as detector geometry or Doppler broadening, demand strong requirements on both instrumentation and
reconstruction, especially at low photon energies of clinical interest.
PhD fellowship project
In this project, we will focus on reaching the technological requirements for nuclear medicine applications by implementing
optimized data processing and enhanced tomographic reconstruction algorithms for systems based on different detector
technologies considering recent advances on instrumentation. In this context, we have developed a ML-EM reconstruction
algorithm based on precise physics modeling which includes Total variation (TV) prior and Point Spread Function (PSF)
corrections for image quality enhancement ([6], [7]). PSF correction leads to noisy images with artifacts, especially for low
doses. One of the aims of this project is to develop blind deconvolution techniques using neural networks, specific to detector
technologies and robust for low doses. In order to train the network and evaluate the performance of the reconstruction
algorithm, accurate modeling of the response of Compton camera systems is essential ([7].) To this end, the Compton camera
module (CCMod, [8]) developed within GATE/Geant4 ([9]) Monte Carlo simulation toolkit will be employed. The developed
model of a scintillation detector based Compton camera prototype for the validation of the module against experimental
data (MACACO, [10]) in the framework of a collaboration between CREATIS and IRIS groups, will be a starting point for
modeling the response of different technologies. Finally, in this project we will study the application of these imaging systems
for the monitoring of thyroid cancer, which is one of the most common cancers in young women. The proof of concept will
be made on simulated data. Preliminary tests could be made on real data obtained within the framework of collaborations.
Objective and tasks
The goal of this PhD fellowship is to study and evaluate the ability of deep learning to improve the quality of Compton
camera imaging based on different detector technologies with respect to standard techniques, in thyroid cancer monitoring.
• Model Compton camera systems based on scintillators
• Simulate Compton camera acquisition data using CCMod
• Prepare and optimize the training data set
• Develop and evaluate blind deconvolution techniques using neural networks robust for low dosis
• Evaluate the performance of these cameras equipped with an advanced reconstruction algorithm for thyroid cancer
monitoring in realistic conditions
• Repeat each step for Compton cameras based on semiconductor detectors
• Compare performance of different detector technologies in the studied clinical case
Thesis supervision
The PhD candidate will be recruited at CREATIS and INSA Lyon and will be co-supervised by Voichita Maxim and Ane
Etxebeste. He/she will work in the Tomoradio team in a stimulating environment composed of researchers in inverse problems,
tomography, imaging for radio-therapy and Monte Carlo simulation. This work will be done in a strong collaboration with
physicists from IP2I Lyon and LPSC Grenoble.
Profile Required
We are looking for enthusiastic and autonomous students with strong motivation and interest in multidisciplinary research.
• Education: Master in Applied or Pure Mathematics, Computer Science, Signal and Image processing, Biomedical
Physics or engineering degree in related fields.
• Scientific interests: computer sciences, deep-learning, medical applications, applied mathematics, Monte Carlo
simulations.
• Programming skills: Python, Tensorflow.
• Languages: English required, French optional.