An algorithm that predicts whether pancreatic cancer has spread to other organs could help avoid unnecessary surgeries

• In pancreatic cancer, knowing if there is metastasis is key to deciding whether to operate or not. Nowadays, a significant number of patients undergo unnecessary invasive surgeries because their metastasis was not detected in time.
• "Our algorithm accurately predicts metastasis using images that are already routinely obtained," says Malats.
• The article is published in the journal 'GUT'.
• The algorithm will be tested in hospitals thanks to a project that has received nearly 800,000 euros in funding from the Spanish Department for Digital Transformation.

Pancreatic cancer remains one of the biggest challenges in the field of oncology. Numbers of cases are increasing, and the latest personalised therapies and immunotherapy are not yet yielding results. Much of the effort is focused on detecting the tumour as early as possible, because the majority of cases are diagnosed at a late stage. But the research also aims to help make the best clinical decision after diagnosis.

When deciding whether to operate or not, consultants must know the primary tumour has already spread to other organs. If it has spread – if there is metastasis – surgery is not recommended. The problem is that in pancreatic cancer, this is very difficult to determine. Nowadays, a significant number of patients are undergoing unnecessary surgery when their metastases has not been detected in time.

A team led by Núria Malats, from the National Cancer Research Centre (CNIO), has developed an algorithm that accurately predicts the presence of metastasis using medical images of the primary tumour.

It is a high-performance deep-learning model, which is looking promising in terms of helping surgeons and doctors to detect metastases, which could fine-tune planned surgery and improve outcomes for pancreatic cancer patients, as stated in the publication GUT.

It is essential to know if there is metastasis before deciding to operate.

“If a person with pancreatic cancer already has metastasis, operating will not cure this, and it can even make their situation worse,” explains Malats, head of the Genetic and Molecular Epidemiology group at CNIO. “Surgery is very invasive and can cause the patient to suffer more, without improving their prognosis. That’s why it is essential to know beforehand if there is metastasis when deciding about a surgery. Our algorithm accurately predicts the presence of metastasis using images that are already routinely obtained.”

The PMPD algorithm (Pancreatic cancer Metastasis Prediction Deep-learning algorithm), which uses artificial intelligence, was tested with data from around 250 patients from the Dutch clinical trial PREOPANC1 on first-line treatment for pancreatic cancer. The principal researcher on this trial, Casper Van Eijck, has also been involved in the paper that is now being published. The algorithm had a high success rate.

Avoiding unnecessary side effects

Specifically, the PMPD algorithm accurately classified 56% of the metastases in the PREOPANC-DPCG dataset, “a promising result in pancreatic cancer, especially in this type of complex diagnosis,” Malats points out.

The model performed consistently regardless of the location of the metastasis. The size and location of the primary tumour, as well as the patient’s gender and age, did not affect its predictive capacity.

The results are particularly positive when considering patients from the PREOPANC-DPCG study whose metastases were only detected in the operating room. The PMPD algorithm predicted 65.8% of these metastases, which means that, if it had been used at the time, “these patients could have been spared surgical intervention,” says Malats.

A data-driven second opinion

The algorithm also predicts the development of the disease. As Malats explains, “it not only tells us if there is metastasis now, but also tries to predict if it will appear in the coming months. This helps doctors make better decisions about whether to operate or not, plan treatments more tailored to the patient’s risk, and avoid unnecessary interventions.

It is a project led by the CNIO group with the collaboration of experts in medicine, computer science, and statistics from institutions in Spain and the Netherlands.

The success of the algorithm is due to being trained with a large amount of real medical data (CT scan images and clinical data). Malats also adds that it uses artificial intelligence techniques to detect patterns that are difficult for the human eye to see.

The algorithm is designed as a complementary tool, explains the CNIO researcher: “it helps doctors (radiologists, oncologists, and surgeons, especially) make decisions, but does not replace their professional judgement. It provides a data-driven second opinion, which can make diagnosis faster, more accurate, and less risky for the patient.”

Next step: validate with patients in hospitals

However, there are limitations. Malats asserts that “more validation in different hospitals and populations is needed.” And, like any AI development, it can produce false positives (incorrectly indicating the presence of metastases) or false negatives (failing to detect them when they are present).

Therefore, one of the group’s next objectives is to test the algorithm on real patients, in real time, in collaboration with hospitals such as Vall d’Hebron (Barcelona), Ramón y Cajal and Gregorio Marañón (Madrid), Centro Universitario de Navarra, and the Dutch Pancreatic Cancer Group (DPCG). They are also reaching out to hospitals in China and Uruguay, in order to obtain as diverse a range of images as possible.

The group has received nearly 800,000 euros in funding from the Spanish Department for Digital Transformation for the project Implementation in tertiary hospitals of the AI-PMPD algorithm for predicting pancreatic cancer metastasis and demonstrating its real-time performance. 

Funding:

Pancreatic cancer AI for genomics and personalized Medicine (PANCAIM Study). H2020 #101016851.

About the National Cancer Research Centre (CNIO)

The National Cancer Research Centre (CNIO) is a public research centre under the Department of Science, Innovation and Universities. It is the largest cancer research centre in Spain and one of the most important in Europe. It includes around five hundred scientists, along with support staff, who are working to improve the prevention, diagnosis and treatment of cancer.