Automated quantification in preclinical nuclear imaging

Context 

Targeted radionuclide therapy is a cancer treatment modality that is currently attracting growing interest in oncology. It relies on the administration of a radiopharmaceutical compound composed of a vector molecule specifically targeting tumor tissues, linked to a radioisotope emitting ionizing radiation. During their preclinical development phase, these therapeutic agents are administered to tumor-bearing rodents undergoing longitudinal small-animal nuclear imaging examinations such as positron emission tomography (µPET) or single-photon emission computed tomography (µSPECT). The registration of these functional images with morphological acquisitions obtained through micro-computed tomography (µCT) or magnetic resonance imaging (µMRI) enables the delineation of organs and tumors, as well as in vivo quantification of radioactivity distribution throughout the animals’ bodies. µPET/MRI offers several advantages that make it a promising alternative to µPET/CT, which has long been considered the reference imaging modality for biodistribution studies. However, both techniques have inherent limitations that may affect the accuracy of the experimental results they provide.

Scientific Background and Objectives

At present, one of the main limitations to the development of preclinical PET/MR imaging lies in the difficulty of correcting PET data for photon attenuation based on MR information, as the MR signal does not reflect the electron density of tissues. In contrast, attenuation maps can be more easily derived from CT imaging. However, CT suffers from poor soft-tissue contrast, which complicates organ delineation. The overall objective of this PhD project will be to contribute to improving the quantitative accuracy of multimodal data. 

PhD Work Plan 

In the initial phase, the candidate will conduct a literature review on MR-based attenuation correction methods for PET, focusing on techniques applicable to small-animal imaging. An initial assessment will then be performed to quantify the bias introduced by the absence of attenuation correction in mice and rats, and to evaluate the performance of a simplified uniform correction model assuming water-equivalent tissues. This preliminary work could combine Monte Carlo simulations with experimental validation using the preclinical imaging systems and refillable murine phantoms available at the ImaThera laboratory. The candidate will also take part in studies on rodents in collaboration with the biology team. Subsequent research will be structured around three main objectives.

Axis 1 – Implementation of Animal-Specific Attenuation Maps 

The candidate will contribute to the optimization of MRI sequences designed to differentiate the tissues within the field of view. Based on these data, the candidate will develop personalized attenuation maps by assigning appropriate linear attenuation coefficients (LACs) to each tissue, in order to generate multi-class models tailored to individual animals. To facilitate and improve the reliability of this process, an automated organ segmentation approach will be investigated. Initially, this segmentation may rely on artificial intelligence (AI) tools trained to automatically identify and delineate the main organs from MR images. This step will make the generation of attenuation maps faster, more reproducible, and less dependent on manual operations. In the longer term, the segmentation tool could be extended to other imaging modalities (such as PET, µCT, or multicontrast MRI), paving the way for a multimodal organ segmentation solution. The candidate will collaborate with the British company MR Solutions Group for this work. The performance of the newly developed attenuation maps will be evaluated on cohorts of animals with varying body masses by comparing quantitative image-based results with ex vivo gamma counting of organs after necropsy.

Axis 2 – Implementation of RF Hardware-Specific Attenuation Maps 

In PET/MRI, several hardware components, particularly radiofrequency (RF) coils, are positioned between the animal and the PET detector. These components cause non-uniform attenuation along the detector’s lines of response. The candidate will develop a hardware attenuation correction by generating attenuation maps from 3D models of RF coils created using computer-aided design (CAD), or from clinical CT scans of the coils. In the latter case, the energy difference between CT Xrays and PET annihilation photons will need to be compensated to ensure the accuracy of the resulting attenuation map. This part of the project will be conducted in collaboration with MR Solutions Group.

Axis 3 – Automated Generation of Attenuation Maps Using Deep Learning

The use of AI for PET data correction has been the subject of several key studies. In clinical PET/MRI, researchers have demonstrated the feasibility of generating multi-class attenuation maps from trained neural networks without requiring additional acquisitions or radiation exposure, notably by synthesizing realistic CT-like images from MR data. A proof-ofconcept study will be conducted to assess the feasibility of implementing such an approach in preclinical imaging. Quantification results obtained from PET reconstructions corrected with these AI-derived attenuation maps will be compared with ex vivo gamma counting data used as a reference.

The selected candidate will be recruited within the framework of the FEDER i-NanoT project (€17M, 4 years) and will join two dynamic hospital and university teams collaborating within a highly multidisciplinary environment. The candidate will be hosted at the ImaThera research unit of the Centre Georges-François Leclerc (CGFL), located on the university hospital campus of the University of Burgundy in Dijon, France, which features a 575 m² laboratory dedicated to medical research. Additionally, the IFTIM team (Functional and Molecular Imaging and Medical Image Processing) of the Institute of Molecular Chemistry at the University of Burgundy aims to cover two essential areas in clinical imaging: the development and deployment of new imaging tracers in both preclinical and clinical settings, and the processing and analysis of medical images. The candidate will also collaborate with MR Solutions Group (Guildford, UK).

Master’s degree (M2) in Medical Physics, Medical Imaging, Health AI, or equivalent field.

Basic familiarity with Monte Carlo simulation methods and/or artificial intelligence would be an advantage.

Strong interest in experimental work (physics, instrumentation, metrology) and multidisciplinary approach (medicine, biology, pharmacy). This PhD involves significant laboratory-based measurements in molecular imaging, including work with laboratory animals.

Professional-level English; proficiency in French is an advantage.

Curiosity, autonomy, initiative, and scientific rigor.

Professionalism and discretion, as the work will take place in a hospital environment.