Development of automatic gamma spectrum analysis using a hybrid machine learning algorithm for the radiological characterization of nuclear facilities decommissioning.
The application of gamma spectrometry to radiological characterization in nuclear facility decommissioning, requires the development of specific algorithms for automatic gamma spectrum analysis. In particular, the classification of concrete waste according to its level of contamination, is a crucial issue for controlling decommissioning costs.
Within CEA/List, LNHB, in collaboration with CEA/DEDIP, has been involved for several years in the development of tools for the automatic analysis of low-statistics gamma spectra, which can be applied to scintillator detectors (NaI(Tl), plastics). In this context, an original approach based on a hybrid machine learning/statistics spectral unmixing algorithm has been developed for the identification and quantification of radionuclides in the presence of significant deformations in the measured spectrum, due in particular to interactions between the gamma emission from the radioactive source and its environment.
The proposed subject follows on from thesis work that led to the development of the hybrid algorithm with the aim of extending this approach to the radiological characterization of concrete surfaces. The candidate will be involved in the evolution of the hybrid machine learning/statistical algorithm for the characterization of concrete for classification as conventional waste. The work will include a feasibility study of modeling the deviations of the learned model to optimize the robustness of decision-making.
Development of a 3D gel dosimetry method for quality control of radiotherapy treatment plans using ultra-high dose rate charged particle beams (FLASH)
Ultra-high-dose-rate FLASH radiotherapy is one of the most promising innovations of the last decade in radiation oncology. It has the potential to eradicate radioresistant tumours and reduce unwanted side effects, that in turn increases cure rates and improves patient quality of life. However, dosimetry infrastructure is lagging behind this clinical and technological advance, with current dosimeters no longer suitable and none of those under development achieving consensus.
The optically read dosimetric gel developed at LNHB-MD (CEA Paris-Saclay) is a promising candidate, as photon beam measurements have shown a linear response over a wide dose range (0.25 - 10 Gy) as well as independence in energy (6 - 20 MV) and dose rate (1 - 6 Gy/min). In addition, this water-equivalent dosimeter has the unique ability to provide three-dimensional measurements with high spatial resolution (< 1 mm) with an associated combined uncertainty of approximately 2% (k = 1). This dosimetry method has been validated for quality control of conventional radiotherapy treatment plans but has never been tested with FLASH beams.
This doctoral project aims to develop a 3D gel dosimetry method suitable for FLASH radiotherapy delivered by charged particle beams: (1) conventional energy electrons (= 10 MeV), (2) very high energy electrons (VHEE = 50 MeV), and (3) protons (= 100 MeV). For each of these types of beams, available at the Institut Curie in Orsay and also at Gustave Roussy in Villejuif, the validation of the dose distribution measured by gel will be carried out by comparison with measurements using other dosimeters (e.g. diamond, alanine) and Monte Carlo simulations.
This study will make a significant contribution to improving patient safety, optimising treatment efficacy and the future integration of FLASH radiotherapy into clinical practice.
Advancing Health Data Exploitation through Secure Collaborative Learning
Recently, deep learning has been successfully applied in numerous domains and is increasingly being integrated into healthcare and clinical research. The ability to combine diverse data sources such as genomics and imaging enhances medical decision-making. Access to large and heterogeneous datasets is essential for improving model quality and predictive accuracy. Federated learning is currently developed to support this requirement offering an alternative by enabling decentralized model training while ensuring that raw data remains stored locally at the client side. Several open-source frameworks integrate secure computation protocols for federated learning but remains limited in its applicability to healthcare and raises issues related to data sovereignty. In this context, a French framework is currently developed by the CEA-LIST, introduces an edge-to-cloud federated learning architecture that incorporates end-to-end encryption, including fully homomorphic encryption (FHE) and resilience against adversarial threats. Through this framework, this project aims to deliver modular and secure federated learning components that foster further innovation in healthcare AI.
This project will focus on three core themes:
1) Deployment, monitoring and optimization of deep learning models within federated and decentralized learning solutions.
2) Integrating large models in collaborative learning.
3) Developing aggregation methods for non-IID situation.