Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : ICube Laboratory – University of Strasbourg
Durée : 32 months
Contact : wemmert@unistra.fr
Date limite de publication : 2023-12-01

Contexte :
Primary liver cancers define a wide spectrum of tumors including hepatocellular carcinomas (HCC), cholangiocarcinomas (CCA) and combined hepatocellular-cholangiocarcinomas (cHCC-CCA) sharing both components. Due to high intratumor heterogeneity, accurate diagnosis of cHCC-CCA is still challenging. In addition, studies aiming to evaluate its prognosis provided discordant outcomes, with tumor behavior closer either to HCC or CCA. Considering the different management and prognosis of the types of primary liver cancers, improving their morphological characterization and recognition is needed and helpful to accurately identify cHCC-CCA.
In order to provide a comprehensive morphological signature of cHCC-CCA, we aim to develop a multiscale morphological approach (from molecular to microscopic) integrating molecular pathology using MALDI imaging (a global in situ proteomic approach), histology and immunohistochemistry (IHC). Firstly, phenotypical features of cHCC-CCA will be derived from direct comparison with HCC and CCA. Secondly, we will search for specific phenotypical features of cHCC-CCA in order to develop a diagnostic application and a prognostic correlation on the clinical outcomes. For this purpose, specific artificial intelligence algorithms based on deep learning will be developed to extract useful information and features from each image modality. The project will benefit from the collaboration and the expertise of computer scientists specialized in data and image analysis, pathologists, analytical chemists specialized in molecular imaging by mass spectrometry and clinicians. We aim to build a comprehensive exhaustive classification of cHCC-CCA based on their multilevel morphological features and identify prognostic subgroups allowing to propose a tailored management of patients

Sujet :
The candidate recruited will be in charge of developing and testing new models of deep neural architecture for multi-modal analysis of mass spectrometry and histopathology data.
We propose a sparing and original approach, relying on the use of a common backbone unsupervisely trained in an autoencoder. For that, we will rely on a pre-trained model that has proved its capacity to accurately identify and classify liver tumors between HCC and CCA. This model will be fine-tuned on our own dataset of pure HCC, pure CCA, and mixed tumors. To have enough data and a more robust model, patches from TMA and WSI will be used for that task. Once the autoencoder is trained, the first layers to the latent space will be kept and used to train simultaneously 3 fully connected classifiers: one to distinguish between HCC and CCA, one to evaluate the mVI and one to quantify the fibrosis of the tumor (the 3 features associated to clinical outcomes). This architecture will be trained on the annotated TMA: each TMA image will be divided into small tiles associated with the pathologist annotation. Finally, the trained models will be applied in a patch-based manner and evaluated on WSI. The evaluation will rely on the annotations provided by the pathologists. For each patch on the WSI our networks will provide a probability on the three aspects of the disease (tumor composition, mVI and fibrosis). Thus, for each WSI, we will obtain spatial information on the different features: which part of the tumor is considered as CCA or HCC, parts of the tissue that indicate that there is or not a vascular invasion and localization of the fibrosis.

Profil du candidat :
– Qualifications/knowledge:
PhD in Computer Science, specialized in machine learning.
Solid knowledge of Data Science and more particularly of supervised and unsupervised deep learning methods.
Experience in (medical) image analysis would also be valuable.

– Operational skills/expertise:
Good experience in Python programming and deep learning libraries (Keras, PyTorch…).
Good verbal (English or French) and written (English) communication skills.

– Personal qualities:
Interpersonal skills and the ability to work individually or as part of a project team.

Formation et compétences requises :
PhD in computer science, with knowledge and experience in data mining and image analysis.

Adresse d’emploi :
ICube – UMR 7357
300 boulevard Brant
67400 ILLKIRCH

Document attaché : 202310111158_Fiche_poste_anglais.pdf