AI-enhanced uterine cancer diagnosis

DNA changes in a tumor determine the behavior of the tumor and the course of the disease

Using microscopic images of a tumor to see what DNA changes are present in the tumor and thus determine what type of uterine cancer the patient has. This is impossible for a human eye. That’s why pathologists at Leiden University Medical Center (LUMC) turned to artificial intelligence (AI) for help. And successfully. Read here how the findings, published in The Lancet Digital Health, can improve uterine cancer diagnosis and treatment.

DNA changes in a tumor determine tumor behavior and disease course, as has become clear in recent years, thanks in part to work at the LUMC. There are four types of uterine cancer, each with a different disease course. It is important for the patient and the doctor to know what type of cancer it is, but this currently requires expensive additional DNA testing. Pathologists wondered whether these “molecular” types of uterine cancer could also be distinguished under a microscope.

Artificial intelligence predicts DNA changes
To do this, they used microscopic images of uterine cancer from more than 2,000 women who participated in the PORTEC clinical trials, coordinated by the LUMC by Professor Carien Creutzberg. All of these patients underwent surgery and were given permission to use the residual material for scientific research. With this unique collection of images, researchers in the pathology department have created an artificial intelligence model that predicts DNA abnormalities and therefore various types of uterine cancer. It’s important here that the model shows the researchers where the visual information for predictions is hidden in the fabric. So it’s not a black box, like other AI models.

Improving the diagnosis of uterine cancer
“The application of artificial intelligence to imaging microscopy is still in its infancy. Through this study, we wanted to learn more about the relationship between tumor appearance and underlying DNA changes. With this work, we learned which areas of tumors contain the most important visual information for diagnosis, and therefore what pathologists should focus on,” says Sarah Fremond, PhD candidate in Pathology.

Uterine cancer is the most common cancer of the female genital tract. A lot of research is being done on this type of tumor at the LUMC. According to the researchers, this study contributes to the further improvement of the diagnosis and treatment of uterine cancer. “As a next step, our team will now develop an AI model that can predict the risk of metastasis,” adds pathologist Tjalling Bosse.

Read the entire article The Lancet’s digital health.

This work was funded by the Hanarth Fund and was carried out in close collaboration with the University of Zurich.