Can molecular information from primary tumor regions identify prostate cancer that is more likely to spread, sparing some patients from unnecessary lymph node testing? Gabriel Wasinger shares results from a study attempting to answer this question with machine learning, presented at ECP 2026.
Gabriel Wasinger is a pathologist at the Medical University of Vienna.
The following transcript has been edited for clarity.
Hello, my name is Dr. Gabriel Wasinger, and I am a pathologist at the Department of Pathology at the Medical University of Vienna. At this year's European Congress of Pathology, I am presenting our research on lymph node metastatic prostate cancer. This project was conducted at our Department of Pathology in close collaboration with other pathologists, as well as bioinformaticians, and under the supervision of Professor Gerda Egger.
To give you some background, in prostate cancer surgery pelvic lymph nodes are sometimes removed to check whether the tumor has already spread. This procedure can provide very important staging information, but not every patient benefits from it. Some patients that are selected for surgery based on clinical risk assessment methods or nomograms ultimately have no metastasis detected in the pelvic lymph nodes. This highlights the need for better prediction methods, and our goal was to investigate whether molecular information from primary tumor regions could help identify prostate cancer that is more likely to spread to the lymph nodes.
A major challenge in this is that prostate cancer is often highly heterogeneous. A single prostate can contain several tumor areas, and these may not perform the same way or behave the same way biologically. We therefore decided to analyze the tissue using spatial transcriptomics, a technology that allows us to measure gene activity directly in selected regions while keeping the tissue context and the location intact.
Our study cohort consisted of a total of fifty-one prostate cancer patients, with and without pelvic lymph node metastases. The analyzed tissue included primary tumor regions, matched lymph node metastases, benign prostate glands, and the surrounding stromal areas. To address the challenge of tumor heterogeneity in patients with multiple tumor foci, we compared the molecular profiles of the primary tumor regions with their matched lymph node metastases to identify the tumor focus most likely linked to the metastatic spread.
We found that primary tumor regions associated with lymph node metastases had a distinct molecular profile. Using this information, we trained machine learning models to recognize patterns linked to lymph node involvement directly in the primary tumor tissue. In internal testing, these models showed promising performance, suggesting that spatially resolved molecular information may help identify patients at higher risk of lymph node metastasis. However, this work is still exploratory and will need further validation before it can be used clinically.
As a next step, we're also interested in whether morphology-based machine learning approaches could complement the molecular data, making it possible to combine what pathologists see on the slide with the spatial transcriptomic information. In the long term, this may support precise patient stratification and help to reduce unnecessary removal of pelvic lymph nodes.
Thank you.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
