Bio Image Analysis.ORG

Description :

Histopathology is the gold standard for the diagnosis of many diseases such as cancers. It is based on the visual assessment of tissue samples with multiple stains highlighting the multiple tissue structures. Nowadays, with the advent of Digital Pathology, the slide scanners generate gigapixel images, called whole slide images (WSI). For diagnostic purposes, superimposing WSIs of differently stained tissues enables us to study patterns of correspondence and divergence between stains, and to assess intra-tumoral heterogeneity, for example.
Image registration or Image alignment enables the pathologist to assess the histology and expression of multiple markers for a patient in a single area. In addition, due to tissue processing and pre-analytical steps, sections may undergo non-linear deformations. In other words, they stretch and change shape from one section to the next. At present, there are only a few automatic alignment tools capable of processing large images with sufficient accuracy and reasonable processing time.

Pré-requis : Pyrhon, java

Travail demandé : The aim of this project is to align multi-colored whole slide (WSI) images to characterize and quantify immune cells within tumor tissue. We have already developed a framework based on rigid registration followed by B-unwarp registration that handles gigapixel image registration [1]. However, there is still room for improvement, in particular to detect the set of robust key points per pair images in order to estimate the registration transform. After reviewing state-of-the-art methods in the field of WSI registration, the work consists in developing a model capable of estimating transform to be applied to regions of interest (ROI), such as tumor regions or any other structure in the tissue, by investigating optimal transport. A comparison with existing methods will be carried out, and particular attention will be paid to the metrics used. This study is a part of the European BigPicture project.