You probably know how reassuring it feels when a CT scan comes back clean.
For most people that's the end of the story. But in pancreatic cancer research, that reassurance is now being questioned.
A new AI model called REDMOD reread CT scans from more than 200 patients who were later diagnosed with pancreatic cancer. The scans had originally been read as normal.
It flagged the disease a median of 475 days before diagnosis, in a study published this month in the journal Gut.
Also read: You don't have to lose weight to improve your heart health, research suggests
On an independent test set, REDMOD caught 73 percent of the hidden cancers. Experienced radiologists reviewing the same scans caught 39 percent.
For scans taken more than two years before a clinical diagnosis, the gap widened. The AI caught 68 percent. The radiologists caught 23 percent.
Why early detection matters so much for pancreatic cancer
Pancreatic cancer is almost always found late. That is the main reason it is one of the deadliest cancers.
According to the Pancreatic Cancer Action Network, the five-year survival rate stands at 13 percent.
Also read: These 8 symptoms could mean your thyroid is overactive
For the most common form, pancreatic ductal adenocarcinoma, it is only 8 percent. The disease has no routine screening test, unlike breast, colon or prostate cancer.
By the time symptoms appear, the cancer has usually spread. Only a minority of patients are diagnosed while the tumor is still localized.
Where early detection would change the picture is dramatic. Modelling studies cited by the authors suggest that lifting the share of localized diagnoses from 10 percent to 50 percent would more than double survival rates.
How REDMOD finds what the eye cannot
REDMOD stands for Radiomics-based Early Detection MODel. Radiomics means pulling thousands of quantitative texture features out of a medical image. These are patterns of pixel intensity and shape that a human reader cannot see.
Also read: Sleeping more than 9 hours? Studies link it to a higher risk of early death
The Mayo Clinic team, led by Sovanlal Mukherjee, trained the model on routine abdominal CT scans. They tested it against 219 patients who later developed pancreatic ductal adenocarcinoma.
The comparison group was 1,243 cancer-free controls matched by age, sex and scan date. All scans had been taken from three months to three years before diagnosis.
The crucial point is that no new equipment is involved. REDMOD works on CT scans people are already having for other reasons. A years-early warning can come out of imaging that would otherwise sit in a file.
What is still missing before any patient benefits
The study is a proof of concept. The researchers acknowledged that their test population lacked ethnic diversity. REDMOD cannot be used in routine care yet.
Also read: Daily fiber supplement eased knee arthritis pain in six-week trial
The next step is prospective validation in high-risk patients, such as those with unexplained weight loss or newly diagnosed diabetes.
If you have symptoms that worry you, including persistent abdominal or back pain, unexplained weight loss, new jaundice or new-onset diabetes, talk to your doctor.
Also read: Your tongue may reveal health problems before you notice other symptoms
This article is made and published by Ida-Marie Palm Varbæk, who may have used AI in the preparation.