Innovative AI Algorithm Enhances Detection of Pancreatic Cancer Metastasis to Prevent Unnecessary Surgeries

Pancreatic cancer remains a significant challenge in oncology, with rising case numbers and limited success from current personalized and immunotherapy treatments. Early detection of the tumor is crucial, as most cases are diagnosed at advanced stages, making treatment options more limited. Beyond early diagnosis, accurately assessing whether the cancer has spread to other organs—metastasized—is vital for determining the appropriate clinical approach. Surgery is beneficial only if the primary tumor has not metastasized; if metastasis is present, surgery is unlikely to improve prognosis and may cause unnecessary harm.

Researchers led by Núria Malats at the Spanish National Cancer Research Centre (CNIO) have developed a cutting-edge deep learning algorithm capable of predicting the presence of metastasis in pancreatic cancer using routine medical imaging. This high-performance artificial intelligence tool analyzes images of the primary tumor to assess the likelihood of metastasis, offering a promising advance in clinical decision-making. The algorithm's potential to identify metastases accurately could help clinicians fine-tune treatment plans, avoiding non-beneficial surgeries and reducing patient risk.

The new deep learning model, called the Pancreatic cancer Metastasis Prediction Deep-learning (PMPD) algorithm, was validated using data from approximately 250 patients involved in the Dutch PREOPANC1 trial. The results showed that the algorithm successfully predicted metastasis in 56% of cases within a complex dataset, demonstrating significant promise for real-world application. Notably, the model’s predictions were unaffected by variables such as the primary tumor’s size and location, patient age, or gender.

A particularly important finding was the algorithm’s ability to foresee metastasis that was only discovered during surgery. In such cases, the PMPD predicted 65.8% of metastases, suggesting that its routine use could potentially spare patients from undergoing invasive surgeries unnecessarily. As Malats explains, "If a patient with pancreatic cancer already has metastasis, surgery will not cure the disease and might worsen their condition. Knowing this beforehand using our algorithm helps in making more informed treatment choices."

In addition to detecting existing metastases, the algorithm also predicts the likelihood of future spread. This capability allows medical teams to tailor treatment strategies, manage patient expectations, and avoid unnecessary procedures. Developed through collaboration between experts in medicine, computer science, and statistics from Spain and the Netherlands, the algorithm leverages extensive real patient data and artificial intelligence techniques to detect subtle patterns imperceptible to the human eye.

While promising, the researchers acknowledge the need for further validation across different hospitals and patient populations. Potential challenges include false positives, where the algorithm might incorrectly identify metastasis, and false negatives, where it might miss existing metastasis. The team aims to test the algorithm in real-time clinical settings at hospitals such as Vall d'Hebron, Ramón y Cajal, Gregorio Marañón, and centers in China and Uruguay to ensure broader applicability and reliability.

Overall, this innovation represents a significant step towards personalized, less invasive treatment plans for pancreatic cancer patients. By providing a data-driven second opinion, the PMPD algorithm has the potential to improve diagnostic accuracy, optimize treatment decisions, and reduce unnecessary surgical interventions.

Source: https://medicalxpress.com/news/2025-09-algorithm-pancreatic-cancer-metastasis-unnecessary.html