The Computational Pathology Group try to understand the correlation between cell morphology, function and the underlaying molecular features. To link different features to morphology, deep learning and quantitative image analysis are used.
We want to closely connect our research to clinics and aim to improve decision making during diagnostics. For that we aim to develop tools which support medical professionals and provide fast information about additional features of a tumor based on the routinely generated H&E slices.
In Gastric cancer four molecular subclasses are defined (CIN, EBV, GS and MSI) which should be predicted based on morphological features extracted from H&E stained tissue slices using convolutional neuronal networks (cNN’s). Here we also focus an tumor heterogeneity, its influence onto the training process and we will use multiplex imaging for further validation.
We investigate different types of lymphomas (cHL and ALCL), which share some molecular properties (e.g. positive for CD30) but have a different clinical course. We quantify cell morphological structures used for cell migration, quantify cell movement and use cNN’s to characterize the migration type to collect information which could help to understand the early spreading of ALCL. Publication: Goncharova*, Flinner* et al. 2019; PMID: 31581676