Supporting the stratification of non-small cell lung carcinoma for Anti PD-L1 immunotherapy with digital image registration

摘要

The analysis of differently-stained pathology slides constitutes an important way of collecting both morphological and functional tissue-related information and is a widely used as a tool for stratification of the diagnosis of a wide range on oncological diseases. Due to deformations induced during the sectioning process and the need to view differently stained slides alternately, or using multiple microscopes, such combined analysis corresponds to an inefficient and prone to error process. In this work we propose a fast and semi-supervised algorithm which extends a multiresolution approach, adopted from methods in literature that firstly addressed the problem of combined analysis of tissue slides using intensity-based registration of whole-slide scanned images. Our method suggests that the inclusion of a supervised step substantially contributes to the complexity reduction of the registration problem, allowing to halve the number of resolutions used by the registration algorithm without compromising a rapid and accurate approximation of tissue slides alignment at a zoom level of 40x, thus representing a well-adapted solution for pathology labs.