| KEYWORD |
Representation learning for Whole Histological Slide Imaging
Tesi all'estero
Parole chiave IMAGE ANALYSIS, MACHINE LEARNING, ARTIFICIAL NEURAL NETWORKS, MACHINE LEARNING, DEEP LEARNING, OPTIMIZATION, MEDICAL IMAGING
Riferimenti SANTA DI CATALDO
Riferimenti esterni francesco.ponzio@polito.it
Gruppi di ricerca DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Descrizione Whole Histological Slide Images (WSI) are very large (>= GB) multi-resolution files obtained from the digitalization of human tissue samples. The WSIs’ multi-resolution nature (similar to Google maps) allows pathologists to navigate among the depicted tissue, looking for something that can be related to a disease (typically cancer), and that can be quantified into a score, eventually exploited to a survival rate prediction. This visual analysis is time-consuming and affected by a high inter-observer variability. Thus, an important research effort is devoted to providing tools (often Deep learning-based) capable of reducing the pathologists’ effort required for the histological analysis. One of the main challenges in this context is the paucity of available annotated dataset, absolutely needed for supervised learning approaches. Moreover, in a single WSI many different kinds of tissue coexist together and typically are associated with a shared common label.
The aim of the thesis is developing a feasible deep learning-based representation (namely featurization or embedding) of the tissue in the context of histological slides analysis with the final aim of predicting the survival rate for patients.
The thesis period is abroad at the research center Inria Sophia Antipolis - Méditerranée.
Conoscenze richieste Good programming skills (Python). Prior knowledge of Machine Learning and Deep Learning design frameworks is a plus.
Scadenza validita proposta 28/07/2024
PROPONI LA TUA CANDIDATURA