Description
Date depot: 2 octobre 2019
Titre: Cancer detection in Full Field Optical Coherence Tomography images
Directeur de thèse:
Jean-Christophe OLIVO-MARIN (Analyse d'images biologiques)
Encadrante :
Vannary MEAS-YEDID (Institut Pasteur (UMR3691))
Domaine scientifique: Sciences pour l'ingénieur
Thématique CNRS : Sciences de l’information et sciences du vivant
Resumé:
Cancer has become the first cause of death in twelve European countries, including France. More than 140,000 people die of cancer, each year in France. Although there have been great improvements in the diagnosis and treatment of cancer, health outcomes after surgery and reoccurrence of the disease remain a major health problem. Hence, there is a crucial need to improve real-time intraoperative diagnosis to characterize as precisely as possible the tumor margins, in order to reduce the ablation of healthy tissue, to reduce the surgery time, and reduce the risk of additional surgery and cancer resurgence. Moreover, patient’s comfort can be increased and surgery time significantly reduced by introducing in the clinical practice the non-invasive tissue examination optical biopsy. This refers to non-invasive imaging methods that use the properties of light to visualize the tissue at different depths.
In the past years, two new microscopy modalities have emerged: namely Full Field Optical Coherence Tomography (FFOCT) and Dynamic Cell Imaging (DCI) that allow label-free high-resolution 3D imaging of biopsies. FFOCT is highly efficient to capture changes in the tissue microenvironment associated with tumors, but fails to detect single cells. The more recent DCI allows the visualization of cells within their microenvironment, and gives a signal related to single cell metabolism. Together the techniques can provide a comprehensive architecture of biopsies with minimal tissue preparation and a reasonable imaging time, and showed promising results, comparable to histology gold standard.
However, FFOCT and DCI images are still difficult to read and to be interpreted by surgeons, especially since DCI signal origin is not yet fully characterized, there is variability in the signal acquisition and display. Since the obtained contrasts are significantly different from standard histology, it can then be difficult for the medical personnel to adopt this technique.
The aim of our work is to develop a reproducible computer-aided diagnosis tool based on machine learning algorithms that can be used by pathologists at the time of surgery to accurately check for the presence of residual tumor in the margins of the resected specimen within a limited timeframe (5 minutes in comparison to current half-hour). The imaging tool will help and shorten the surgeon’s decision making to excise additional tissue.
Doctorant.e: Mandache Diana