Researchers develop AI-based projects for tumor scoring, vessel annotation
Click Here to Manage Email Alerts
Reidunn Edelmann, PhD, has always been interested in blood vessels.
“They orchestrate so many of the functions of the body and also the call for immune cells,” Edelmann, postdoctoral researcher in the Center for Cancer Biomarkers at University of Bergen in Norway, told Healio. “If the immune cells need to know where to go and when to go there, they need to be told by the blood cell.”
Edelmann’s interest in blood vessels led her to pursue an artificial intelligence-based method for tumor vessel annotation. Her project was among the first to be chosen for development by the aiForward program, an initiative to promote and facilitate use of AI in scientific projects. AiForward is organized and run by Aiforia Technologies, a company founded in 2013 as a spinoff of the Finnish Institute for Molecular Medicine at University of Helsinki.
Liesbeth Hondelink, MSc, medical student at Leiden University Medical Center in the Netherlands, also has a project supported by aiForward. Hondelink used the Aiforia software to develop an algorithm to score PD-L1 in tumor cells.
“I think [AI] is kind of a black box for some physicians,” Hondelink told Healio. “Some of them might be hesitant about it, especially more conservative doctors, because it might seem to be a very big change.”
As members of their field, Edelmann and Hondelink are exceptional. According to data from the Unesco Institute for Statistics, less than 30% of the world’s researchers are women.
In recognition of International Day of Women and Girls in Science, which was observed Feb. 11, Edelmann and Hondelink discussed their AI projects with Healio.
Question: What was the goal of your aiForward research project?
Edelmann: I had two goals. One was to make a proof-of-concept paper using AI in analyzing multiplexed ‘high-content’ tumor vessels. This hasn’t been done before. My other goal was to uncover biologically relevant tumor vessel biomarkers.
Q: What needs in oncology will your AI projects meet?
Hondelink: The pathology department gets a lot of lung cancer biopsies, which can be very small and sometimes contain only a small amount of tumor cells. Pulmonologists ask their pathologist about the PD-L1 score on every stage IV lung cancer biopsy, so they can decide whether to give the patient immunotherapy or chemotherapy. Studies have shown that pathologists are quite bad at estimating that score — they have to identify the cancer cells and estimate which percentage of them stains positive for the PD-L1 protein. Treatment choices depend on this score, and if you ask the same pathologist to score the case again the next day, 10% will have a different score. If you ask two pathologists to score the same case, they will be in agreement only 80% of the time. With our algorithm, a computer will count all the cells one by one, and it will give a result.
Edelmann: The arrangement of blood vessels in a tumor is an important part of assessing the tumor microenvironment, which is an important prognostic factor in cancer. In our project, we seek to more completely characterize these vessels and correlate them to outcome-related endpoints, as well as other biological traits of tumors in different types of cancers. For this project, we are focusing on breast and colon cancer.
QWhat is next for the development of your projects?
Hondelink: We’ve finished training and validating this algorithm. Now, we’re in the process of writing everything down, and we want to see if we can have our results published. The next steps are to run the scoring system alongside regular diagnostics for further validation. In the future, we might use the algorithm as an assistant to the pathologists for difficult cases; the algorithm could serve as a ‘second opinion’.
Edelmann: We are building the algorithm and it looks very good. I’m confident we will be able to fit this algorithm with the different vessel types, but we haven’t run it on a big cohort. Our preliminary results have not been published, so I can’t discuss them yet.
Q: What obstacles have you encountered in your work because of your sex?
Hondelink: I’ve realized I’m very lucky to be a woman in science in the Netherlands. This was my graduation project. I haven’t really experienced many obstacles like that so far. Maybe that could happen in the future, though.
Edelman: This wasn’t an issue for me, but I know that it is for thousands of women, and it’s very important to me that it shouldn’t be that way.
Q: Do you think AI will become an important tool in clinical practice?
Hondelink: Yes. It’s comparable to when we first captured X-ray images, and health care professionals thought, “this isn’t something we’re really ever going to do.” Some were quite hesitant to use them at first. I think we have to educate each other as doctors on how AI works and how we can benefit from it. – by Jennifer Byrne
For more information:
Reidunn Edelman, PhD, can be reached at University of Bergen, P.O. Box 7800, 5020 Bergen, Norway; email: reidunn.edelmann@uib.no.
Liesbeth Hondelink, MSc, can be reached at Leiden University Medical Center, Albinusdreef 2, 2333 ZA, Leiden, Netherlands; email: l.m.hondelink@lumc.nl.
Disclosures: Edelmann and Hondelink report no relevant financial disclosures.
Collapse
Read more about
Click Here to Manage Email Alerts