Ph. D. Project
Model inversion and separation of optical spectroscopic signals for in vivo characterization of skin tissues
2016/10/01 - 2019/09/30
Several methods of tissue optical spectroscopy (also called "optical biopsy") have been proposed to diagnose non-invasively pre-cancerous lesions that are hardly detectable in clinical settings. This diagnostic is possible because pathological and early metabolic and morphological modifications induce changes in the optical properties of the biological tissues at subcellular, cellular and tissue scales. The interest of these approaches has been demonstrated for identifying the optical properties of ex vivo tissues and for improving the diagnosis efficiency in several preclinical studies: skin hypertrophic lesions, bladder cancer, precancerous skin lesions.

The collected data are sets of spectroscopic signals (intensity spectra) acquired under two modalities (autofluorescence and diffuse reflectance). Until now, our contributions were dedicated to the conception of algorithms combining various methods of discriminant feature extraction and selection, and supervised classification. A new challenge consists in extracting from spectroscopic signals, discriminant information related to the characteristics and concentrations of the constitutive elements (absorbers, diffusers and fluorophores) in order to determine the nature of the tissue in situ. This analysis must be based on physical models describing the propagation of light in tissues.

The proposed PhD thesis aims at developing an approach based on the inversion of these physical/mathematical models. This approach should lead to solving an unmixing problem through a source separation procedure. The sought source signals are spectroscopic signals which characterize the absorption and emission of chromophores and fluorophores laying inside the tissues.
Source separation is a classical inverse problem. The originality of the problem we will tackle comes from the non-linearity of the mixing model. This non-linear behavior is due to the complexity of the models of absorption and diffusion of light in human tissues. Contrary to blind source separation approaches, some constraints will be imposed on the shape of sources, which are related to the physical models of light-tissue interactions. Some of these models have been already investigated at CRAN (direct and inverse Monte Carlo). The estimation of source signals will then allow us to both reconstruct physically plausible source signals, and infer on the parameters or tissues.

The PhD project will take place within the SpectroLive project at CRAN (instrumentation and clinical prototype) and in collaboration with CHR hospital at Metz (concerning the clinical study and the recording of in vivo data).
skin, optical spectroscopy, fluorescence, diffusion, physical models, source separation.
Required skills :

The candidate needs a Master or Engineering degree. He/she should have a strong background in mathematics and signal and image processing. He/she will have excellent programming skills (Matlab, C/C++, etc.) and good skills in English, both written and oral.


The PhD thesis will be co-supervised by Charles Soussen (CRAN, PhD supervisor) and Walter Blondel (CRAN).

Application :

Please contact Charles Soussen and Walter Blondel ( , and send your CV, covering letter, one (at least) reference letter and your last academic results with rankings, when applicable.
Biology, Signals and Systems in Cancer and Neuroscience
Academic grant application