Artificial intelligence enables earlier assessment of endometrial cancer relapse risk

Researchers from the Institute of Clinical Medicine at the University of Tartu, in collaboration with Helsinki University Hospital, have developed an artificial intelligence-based machine learning model that enabled the relapse risk of endometrial cancer to be assessed before surgical intervention, thereby helping to shift important treatment decisions to an earlier stage.

Endometrial cancer, a malignant tumor of the uterine lining, is one of the most commonly diagnosed gynecological cancers. Despite primary treatment, nearly 20% of patients experience relapse, which is often difficult to treat.

Currently, the risk of endometrial cancer relapse is usually assessed only after surgery, based on post-operative histological findings. This limits physicians’ ability to plan follow-up treatment early and in a patient-centered manner, explained Vijayachitra Modhukur, a bioinformatics specialist at the Women's Clinic of the Institute of Clinical Medicine.

“The results of the study show that preoperative data, such as blood tests and other initial exams, can accurately predict the risk of cancer relapse already before surgical intervention,“ explained Modhukur. “This means that important treatment decisions no longer need to rely solely on post-operative information.”

In this study, researchers used the machine learning model on clinical and laboratory-derived data from 784 endometrial cancer patients enrolled in Helsinki University Hospital. As a key breakthrough, the results showed that endometrial cancer relapse is not a uniform phenomenon.

Early relapse (within 6 months after surgery) is associated with aggressive tumor biology, whereas late relapse (more than 6 months after surgery) is driven by factors such as tumor size and the extent it has spread within the body.

According to Modhukur, distinguishing between these two types of relapse allows treatment to be tailored more precisely to each patient’s individual risk profile. Although the model’s initial results are promising, it is not yet ready for routine clinical use. Further testing is required to identify and eliminate potential limitations and bias.

Once validated, the model could serve as a smart assistant for doctors, making treatment decisions more precise and personalised, noted Modhukur.

“In clinical practice, a doctor would input standard pre-surgery data. Artificial intelligence would then flag high-risk patients before the operation, allowing the medical team to prepare more aggressive treatments or closer monitoring immediately,” added Modhukur.

The collaborative study was published in Computational and Structural Biotechnology Journal.