Functional pathology: computational tool bridging spatial omics and pathology in oncology

Visual assessment of clinical samples stained with Hematoxylin and Eosin has been the state-of-the-art for the past 100-years to discover patterns indicating cancer and decide on patient treatment.

In recent years, AI in the form of convolutional neural networks have shown great potential in mimicking visual assessment, e.g. automating tumor grading.

In parallell, we have seen the emergence of ‘spatial omics’, where many different biomolecules are detected in parallel directly in their tissue location, revealing local tissue function.

These techniques can be used to identify specific cell types and their interactions within the tissue architecture, and provide clinically relevant information.

In this project, we aim to develop computational methods to merge the two types of information to bridge the gap between function (from spatial omics) and large-scale learning-based pathology.

The purpose is to improve our understanding of two specific processes in cancer development: (i) early diagnostic signs in lymph node morphology (ii) the tumor micro-environment patterns in liver metastasis. We build on ongoing collaborations with clinicians, tumor biologists and developers of molecular tools.

The computational methods we develop have the potential to lead to new diagnostic and prognostic markers and the functional omics may provide clues for future therapies.