June 2022
Ana Leni Frei is a PhD student in Computational Pathology at the University of Bern. With her research on cell interactions in rectal cancer, Ana Leni hopes to help build Artificial Intelligence (AI) tools that can assist pathologists to understand changes that occur in the tumor microenvironments of patients receiving neoadjuvant chemoradiotherapy and hopefully understand why 20% of those patients do not respond to the therapy.
Ana, what is your research about?
My project employs a specific method (cell-based graphs) to study tumor environment in patients with rectal cancer with and without neoadjuvant treatment. This type of cancer is usually diagnosed in late stages and these patients receive chemo-radiotherapies in order to shrink the tumor before the resection and prevent relapses. But up to 20% of the patients do not respond at all to those preoperative chemo-radiotherapies. Together with pathologists, we try to find information in the tumor tissue of why some patients respond better than others to this treatment. I look at the tumor microenvironment as a network of interacting cells and based on this try to understand how the tumor is behaving. With this we hope to identify features that are significant for the prognosis of a patient. These new insights can then serve to develop more targeted therapies.
What motivates you about your work?
During my master thesis at EPFL I had the opportunity to work in the field of computational pathology. My project was the application of machine learning and cell-based graphs to automatically classify different tissue types present in colorectal cancer. I really enjoyed working on this project and I particularly appreciated to use cell-based graphs to model the tissues. I strongly believe that looking at the tissues as a network of interacting cells is highly informative about the biology of that tissue and could help better understand disease development, especially in cancer. For these reasons, I wanted to pursue with a PhD in this domain. By chance, at the end of my master thesis, Prof. Inti Zlobec was looking for a PhD student working on exactly these topics so I applied and I had the chance to join her team!
No one will trust an algorithm that recommends a treatment without understanding why the decision is being made.
How can your research be transferred into clinical care?
We should strive to make the AI as explainable as possible and establish a good interaction between computer scientists and clinicians if we want to use these new AI tools in clinical setup. AI is not going to replace clinicians. It’s a tool for more information. One critical point therefore is to develop more techniques to explain the output of our models - to show which part of the input data was important for the AI to come to a certain conclusion. No one will trust an algorithm that recommends a treatment without understanding why the decision is being made.
As cell-based graphs represent the structure of the tissue they allow to keep the biological information present in the initial images. For that reason, predictions arising from graph-based analyses are easier to understand for humans and relate back to biology.
