Aim

Patients diagnosed with advanced rectal cancer get treated with neoadjuvant chemoradiotherapies to reduce the tumor size before the resection and prevent local recurrences. However, an important proportion of the patients do not respond to the neoadjuvant therapy and the tumor continues to grow during the course of the treatment.

In this project, we aim at exploring the tumor microenvironment of rectal biopsies. We capture the tissue composition and structure by building cell-based graphs on rectal biopsies. Individual cells are detected and classified and edges are built to connect close by cells. This cell graph is a representation of the tumor microenvironment and is used to train Graph Neural Networks to predict treatment response based on the biopsies with the aim to find non responders and redirect them towards other treatment options.