News release

From: BMJ Group

AI model detects very early normally ‘invisible’ tissue changes of pancreatic cancer

Offers potential to shift diagnosis to early stage, treatable disease, say researchers

More accurate than radiologists, but it requires testing in high risk patients before clinical use

An AI model (REDMOD) can pick up the very early subtle tissue changes of pancreatic ductal adenocarcinoma, the most common form of pancreatic cancer, which conventional imaging and the human eye find difficult to detect, finds research published online in the journal Gut.

As such, it offers the potential to shift an all too common late stage, terminal disease diagnosis to one that is at an early stage (stage 0) and treatable, say the researchers.

While REDMOD was more accurate than experienced radiologists, it requires testing in high risk patients, defined as those with unexpected weight loss and newly diagnosed diabetes, before it can be widely used in clinical practice, they add.

Pancreatic ductal adenocarcinoma has a poor rate of survival. It’s usually diagnosed late, in the absence of symptoms and visible tissue changes in the early stages, and it rapidly progresses, explain the researchers.

To overcome these challenges, the researchers developed an AI framework, called Radiomics-based Early Detection MODel (REDMOD), designed specifically to pick up the subtle tissue texture patterns (radiomics) of very early pancreatic cancer, which standard computed tomography (CT) scans can’t see.

The framework includes automated pancreatic segmentation—clear delineation of the borders of the pancreas from surrounding tissue/organs, obviating the need for this to be done manually with the attendant risk of variable accuracy.

To test its reliability and effectiveness, the researchers applied REDMOD on abdominal CT scans from 219 patients from several different hospitals, who were deemed to show no evidence of disease after radiologist review, but who were subsequently diagnosed with pancreatic cancer.

In 87 (40%), this was 3-12 months; in 76 (35%) this was 12–24 months; and in 56 (25%) more than 24 months (up to around 3 years) before diagnosis. Disease was located in the head of the pancreas in nearly two thirds (64%) of patients.

Their scans were compared with those of a total of 1243 patients who hadn’t developed the disease up to 3 years later, matched by age, sex, and scan date.

The average age of those who were subsequently diagnosed with pancreatic cancer was 69, but ranged from 34 to 88; and the average age of the comparison group was 64, but ranged from 34 to 88.

REDMOD detected the ‘invisible’ signature of pre-clinical pancreatic ductal adenocarcinoma an average of 475 days before clinical diagnosis.

“This temporal window holds profound significance, as attaining such early detection would substantially augment the probability of cure and improved survival,” highlight the researchers.

“In fact, modelling studies indicate that increasing the proportion of localised [pancreatic ductal carcinomas] from 10% to 50% would more than double survival rates, thereby underscoring that the timing of diagnosis is the single most critical determinant of survival outcomes,” they add.

And REDMOD performed better than radiologists: it was nearly twice as sensitive–the ability to pick up true, rather than false, positive results—at accurately picking up ‘invisible’ early malignant cellular changes: 73% compared with 39%. And it was nearly 3 times as accurate as radiologists for cases detected more than 2 years before clinical diagnosis: 68% vs 23%.

It also correctly identified just over 81% of scans in an independent group (539 patients) drawn from several hospitals and 87.5% in the public US National Institutes of Health NIH-PCT dataset (80 patients) as free of pancreatic cancer.

The pre-clinical changes detected were a reliable indicator of subsequent clinical disease because REDMOD gave the same answer for 90–92% of scans when the same patient was scanned again some months earlier.

The researchers acknowledge various limitations to their findings, including that they weren’t based on an ethnically diverse group of patients.

Nevertheless, they conclude: “This study validates REDMOD as a fully automated AI framework capable of identifying the imaging signatures of stage 0 [pancreatic ductal adenocarcinoma] in normal pancreas, achieving this with substantial lead times and performance superior to expert radiologists.”

They add: “While prospective validation is paramount to confirm clinical utility, the REDMOD framework represents a significant advance towards shifting the paradigm for sporadic [pancreatic ductal adenocarcinoma] from a late-stage symptomatic diagnosis to proactive pre-clinical interception, offering tangible hope for improving outcomes in this challenging disease.”