French
English
(see PDF file for full topic)
Medical Background
Surgery for high-grade gliomas represents a major challenge in neurosurgery.
Accurate identification of tumor margins is crucial for complete resection, but remains difficult because healthy and diseased tissue may appear visually identical. Intraoperative fluorescence spectroscopy offers a promising solution, but requires sophisticated mathematical methods to extract relevant spectral information.
Proposed Approach
This internship proposes an innovative two-level optimization framework to automate the selection of fluorescence wavelengths and maximize healthy-tumor discrimination. The project combines three key elements: (1) realistic physical modeling based on the scattering equation to describe photon transport in heterogeneous tissue, (2) the formulation of a sophisticated nonlinear inverse problem to reconstruct the optical properties of the tissue, and (3) a two-level optimization algorithm that automatically optimizes acquisition parameters.
Algorithms and Tools
You will implement advanced mathematical algorithms in Python/PyTorch, including implicit differentiation for gradient calculation, proximal methods for convex optimization, and nonlinear solvers (Gauss-Newton, Levenberg-Marquardt) for physical model inversion.
Desired Profile and Practical Information
Master’s degree in Applied Mathematics, Signal Processing, or Computer Science with a solid command of Python and convex optimization. Familiarity with PyTorch, proximal algorithms, or inverse problems is highly desirable.
Duration: 6 months | Compensation: ~500–700 €/month | Supervisors: J. Cohen (CNRS), A. Gautheron (UCBL)