Deep Learning for spectral Photon Counting-CT in vivo imaging and treatment design in liver cancer mouse models

Stage numéro : Doctorat-2023-IM-01
Laboratoire :Centre de Physique des Particules de Marseille Case 902
 163 avenue de Luminy - 13288 Marseille Cedex 9
Directeur :Cristinel Diaconu - -
Correspondant :William Gillard - -
Groupe d'accueil :imXgam
Chef de groupe :Christian Morel - -
Directeur de thèse :Yannick Boursier / Maître de Conférences - 04 91 82 76 41 -

Thématique : Imagerie médicale

Location: imXgam group, Centre de Physique des Particules de Marseille (CPPM), France.

Collaboration: Maina Team, Developmental Biology Institute of Marseille (IBDM), France.

Advisors: Flavio Maina, Yannick Boursier.

Keywords: Spectral CT, deep learning, in vivo cancer imaging, immuno therapy.

Context and objective: The PIXSCAN-FLI prototype is a spectral micro PC-CT scanner developed at CPPM. Dedicated to in vivo imaging, it is based on the technological breakthrough of hybrid pixels, which makes possible in particular to identify and quantify contrast agents in the imaged object. Recently, a collaborative research between IBDM and CPPM illustrated its capability to perform unprecedented longitudinal studies to track spontaneous liver tumors in mouse models. This study highlighted the tumor dynamics features throughout the tumorigenic program, and provided promising results for a proposed immuno therapy approach designed to remodel the tumor microenvironment while targeting cancer cells with drugs.

The aims of this thesis are twofold : i) further explore the effectiveness of this immuno-therapy approach and characterize the biological mechanisms involved ii) develop and validate a specific deep learning methodology for an optimal processing of in vivo spectral data. For that purpose, the student will interact with both biologists from IBDM and physicists and mathematicians from CPPM, and will be able to rely on a successful multidisciplinary collaboration of more than 4 years.

The student will be in charge of the longitudinal data acquisition campaigns (and will undergo training if necessary to be certified) and of the data processing with the PIXSCAN-FLI existing routines. The goal is to implement acquisition strategies using biological markers for qualitative/quantitative evaluation and 3D reconstruction of the tumors, their microenvironement and their vascularization. The student will also participate to the development of a deep learning methodology to deal with spectral data. The goal is to improve the accuracy of the existing data processing pipeline, and to speed up the processing time. The student will benefit from the experience of the imXgam group in that field and from its active collaborations from more than 6 years (I2M Lab., CREATIS Lab.). The development will be done with Python.

References: [1] Kronland-Martinet, C. et al., Development of K-Edge Spectral Tomography Using XPAD3 Composite Pixels, IEEE International Conference in Medical Imaging 2014. [2] Cassol, F. et al. “Tracking Dynamics of Spontaneous Tumors in Mice Using Photon-Counting Computed Tomography.” iScience vol. 21 (2019): 68-83.

Candidate: Motivation, perseverance, rigor, creativity and curiosity, strongly interested on an interdisciplinary project combining biology, physic, and mathematic. Students with interdisciplinary profile are strongly encouraged to apply.

How to apply: Please send an email containing “[PC-PT PhD]” in its subject, with motivation letter, CV, Grades and ranking during Master 1 and Master 2, name and complete coordinates with email address of two reference persons to and

Funding: start date: 01/10/2020.