A new artificial intelligence system could change the way pancreatic cancer, one of the most aggressive forms of the disease, is detected. The model can identify signs of cancer months or even years before it is diagnosed, at a time when the chances of survival can increase significantly.
The system, called REDMOD, was presented in a study published in the journal Gutappropriate Science Tech Daily. He analyzes medical scans of the abdomen (CT) and detects changes in the pancreas associated with the most common and aggressive form of pancreatic cancer. According to the researchers, these signs are so subtle that they cannot be noticed even by experienced radiologists.
The stakes are huge. Currently, pancreatic cancer is often discovered late, when the disease is already advanced and treatment options are limited. Researchers say such technology could move diagnosis to a much earlier stage, before the disease becomes clearly visible.
Why is difficult to discover
Pancreatic cancer has one of the lowest survival rates. The main reason: in the early stages it does not give symptoms and is not clearly visible on the images. Plus, it evolves quickly, and wasted time makes a difference.
To overcome this problem, the researchers developed REDMOD, a model that analyzes very fine details in the structure of tissues, which the human eye cannot distinguish. The system also includes an automatic function that separates the pancreas from the surrounding organs, so that the analysis is more accurate and does not depend on the intervention of the doctor.
The model was tested on abdominal CT scans from 219 patients from several hospitals. At the time of the scan, all were considered normal, but were later diagnosed with pancreatic cancer.
In 40% of cases, the images had been taken 3 to 12 months before diagnosis. In 35% of cases, they were between one and two years old. Otherwise, they exceeded two years and reached almost three years before the disease was confirmed.
The results were compared with scans from 1,243 people who did not develop pancreatic cancer over the next three years.
What the system found
REDMOD was able to identify the early signs of the disease, on average almost 475 days before the diagnosis was made. Researchers say this time frame is crucial, because detecting this type of cancer earlier significantly increases the chances of treatment and survival.
The data show that if the number of cases detected in the early stages increased from 10% to 50%, the survival rate could double.
Moreover, the system performed better than radiologists in identifying real cases. The model correctly detected 73% of cases, compared to 39% for doctors. The difference was even greater when the scans were done more than two years before diagnosis. In these situations, REDMOD had an accuracy of 68%, compared to 23% for radiologists.
In additional tests, the model analyzed scans from an independent group of patients and correctly indicated, more than 81 percent of the time, that they did not have cancer. In another data set, sourced from National Institutes of Healththe accuracy reached 87.5%.
Another important aspect: the results were consistent. When the same patients were analyzed on older scans, the system gave the same result more than 90% of the time.
What’s next
The researchers say the model needs to be tested further, especially in high-risk people – such as those who have unexplained weight loss or have recently been diagnosed with diabetes – before it can be widely used. There are also limitations to the study, including the lack of greater diversity among participants.
Even so, the bottom line is clear: AI could change the way pancreatic cancer is detected, shifting the focus from late diagnosis to early intervention. If the results are confirmed in other studies, this technology could bring an essential advantage in the fight against the disease: time.
