Ongoing projects

In diagnostic radiology, the use of ionizing radiation is justified by benefits surpassing risks. From an epidemiological perspective, this balance is difficult to assess because accurate dose values for individuals are not available. This project consists in developing tools to automatically report dose-to-organs from Computed Tomography (CT) images. First, a machine-learning based, multiclass segmentation tool will be developed to automatically contour organs in CT imaging studies. Then, a fast GPU-based Monte Carlo code will be used to compute dose maps from technical scanning parameters stores in DICOM headers of medical images. A large database of dose-to-organ values will be constituted as well as interactive dashboards to explore dose usage as a function of site explored, device used, etc.

On the long term, this database will be linked with epidemiological cancer data to assess potential causal relations.

Recent breakthroughs in medicine have shown a link between cerebrovascular pathologies and the risk of developing mild or major cognitive disorders of the like of Alzheimer’s disease (AD). The project will thus consist in developing a tool for the characterisation of the cerebrovascular system.
Indeed, the measurement of arteries and veins (diameter, density, etc.) in every brain region will
provide a better understanding of the vascular health changes throughout the aging process and if these changes are linked to two markers heavily correlated with neurocognitive disorders: cerebral tissue atrophy and cerebrovascular lesions (mainly hyperintensities in white matter and cerebral microbleeds). Thus, the cerebrovascular system for cognitively healthy participants, patients with mild cognitive disorder and patients with AD will be characterized to better our understanding of cognitive health’s link with cerebrovascular health. 
To summarize, the project is an exploration of the relation between the cerebrovascular system and the AD in order to comprehend the interactions involved and help with the diagnosis of the disease.
 

It is often difficult to share denominated data between different organisations and researchers due to ethical constraints related to respondent's confidentiality. This is a frequent reality in healthcare, given the inherent sensitivity of the data involved. One option in this case is to not share the data directly, but rather to provide access to it via a tool that controls the risk of disclosure of the queries made and allows only those it considers safe. DataSHIELD is such a tool that has been proposed to protect the confidentiality of a dataset, and which is used via the statistical software R. It also allows statistical analyses to be carried out on several datasets hosted in different locations, always ensuring the confidentiality of the respondents. In this project, we are interested in the confidentiality guarantees provided by the software, and in its limitations. In particular, we study the potential uses of the software for advanced statistical analyses, such meta-analyses and the use of neural networks.

It is often difficult to share denominalized data between different organizations and researchers due to ethical constraints related to respondent confidentiality. This is a common reality in the healthcare field, given the inherent sensitivity of this type of data. One option in this case is not to share the data directly, but rather to provide access to it via a tool that controls the risk of disclosure of the queries made and allows only those it considers safe. DataSHIELD is such a tool which has been proposed to protect the confidentiality of a dataset, and which can be used via the statistical software R. It also allows statistical analysis on several datasets hosted in different locations, always ensuring the confidentiality of the respondents. 

In this project, we are interested in the confidentiality guarantees provided by the software, and in its limitations. In particular, we wish to establish principles to guide the choice of disclosure control parameters offered with the tool, and to understand more precisely the impact of these controls on the quality of the descriptive statistics, linear models and graphs produced.
 

Radiotherapy treatments currently used in the clinical field are rarely modified. They generally consist of a global therapy of 50 grays, fractionated in five treatments of two grays every week for five weeks.
Thus, it could be worthwhile to develop a numeric tool, based on mathematical models found in the literature, in order to compare different types of treatment without having to test them on real tissues. Several parameters are known to alter the tissue response after irradiation including oxygen
partial pressure in irradiated regions, particle type hitting the tissue, and treatment duration.

The Python code created as the main part of the project is intended to facilitate the optimization of radiotherapy treatment by generating graphs showing cell survival after a certain number of fractions, taking many parameters into account. When completed and integrated to a graphical interface, the code will be easy to use and helpful for ongoing research projects.

In medical imaging, radiomic features make it possible to characterize heterogeneity of a region of interest at the anatomical level. This way of quantifying the heterogeneity of a region of interest can be useful, for example, in order to identify the more aggressive tumors in oncology. To do this, we hypothesize here that variation in magnetic resonance imaging (MRI) acquisition sequences and its resulting different levels of contrast would make it possible to optimize the subsequent radiomic analysis.
In this project, a pipeline for the analysis of real medical images will first be set up in order to quantify the robustness of radiomic characteristics according to variations in acquisition protocols. Then, an MRI acquisition simulation pipeline will be developed in order to evaluate the potential for optimizing radiomic features in medicine.
 

Interpretability of Artificial Intelligence, that is the capacity of an expert to understand why a prediction is made, is of great importance in health analysis. Firstly, because it matters to understand why a decision is made by an algorithm when it has such impact on a person’s life. Moreover, in research, interpretable algorithms are useful because they often unveil new investigation path. 

This study aims to combine two supervised machine learning algorithms to optimize both interpretability and performance, for instance, with mathematical logic tools. This new algorithm intends to help better predictions by lightly increasing model complexity while preserving high interpretability. 

This algorithm is developed to analyze fat data, which are data with a lot of characteristics (features) but with few samples (observations). This type of data is recurrent in health data, mainly in genomics, metagenomics and metabolomics data, which are all state of the art in medical analysis. More precisely, we are interested in problems such as antibiotic resistance or long corona virus disease (COVID-19). 
 

The PARTAGE project (Patients and Researchers Team-up and Generate Evidence) is a research project that is examining mechanisms to allow patients to securely share their clinical data with researchers using the Opal patient portal. As part of the overall PARTAGE project, this specific sub-project is about obtaining stakeholder feedback on the concept of data-sharing. To do so, we are using a process of stakeholder co-design in which patients, clinicians and researchers are embedded in the research team and we are obtaining additional feedback from each stakeholder group through focus groups and surveys.

Treatment of cancer with radiation is a proven technique used worldwide. One of the ways to treat prostate cancer is by using brachytherapy either alone or as a boost. At the moment, the techniques used depend on the experience of the treatment team and researchers are trying to overcome this problem.
In our case, the technology considered to address this problem is deep learning. Therefore, the aim of this project is to use deep learning to develop tools for planning in high dose rate brachytherapy for the treatment of prostate cancer.
Four different phases are initially targeted. The first consists of a classification of treatment plans. The second is a “reinforce learning” approach to help optimize treatment plans, by modifying the optimization objectives in order to consider each patient in a unique way. The third is dose map prediction based on patient anatomy. The fourth is the generation of treatment plans; from the patient's anatomy or from a dose map to find an adequate treatment plan.
The proposed work is a new approach that will ultimately help with the planning of high dose rate brachytherapy treatments for prostate cancer.
 

Metabolomics is one way of studying metabolism. The presence of certain metabolites, or the breakdown of metabolic pathways can serve as indicators of a patient's health. They can serve as markers for certain diseases such as cancers, or provide information on the quality of an individual's diet. Untargeted metabolomics acquisition methods produce large data matrices. The aim is to develop machine learning methods specifically suited to handle high dimensional data sets. For example models based on decision rules.
The purpose of these models being the search for biomarkers, they must be sparse in order to be able to be interpreted by a human expert. We also try to develop new approaches to better interpret some existing and efficient models. Interpretability is essential in the application of machine learning to health. Models cannot be diagnostic black boxes but rather analytical tools available to experts to better understand human metabolism.