New AI tool can help doctors better detect early pancreatic cancer
REDMOD finds tumor signs as early as 3 years before standard clinical diagnosis
Written by |
An artificial intelligence (AI) tool can detect early-stage pancreatic cancer on standard medical scans up to three years before a clinical diagnosis, according to a Mayo Clinic study. By identifying subtle patterns invisible to the human eye, the tool nearly doubled the accuracy of specialists in spotting the disease while it is still potentially curable.
Researchers found that the AI model, known as REDMOD, could identify the “signature” of cancer even when a pancreas appeared entirely normal to expert radiologists.
“The greatest barrier to saving lives from pancreatic cancer has been our inability to see the disease when it is still curable,” Ajit Goenka, MD, senior author of the study at Mayo Clinic, said in a news story. “This AI can now identify the signature of cancer from a normal-appearing pancreas, and it can do so reliably over time and across diverse clinical settings.”
The study, “Next-generation AI for visually occult pancreatic cancer detection in a low-prevalence setting with longitudinal stability and multi-institutional generalisability,” was published in the journal Gut and funded by the National Institutes of Health (NIH), among others.
The challenge of early detection
Pancreatic cancer is one of the deadliest types of cancer. A large part of the reason that the disease has such a high mortality rate is that most patients aren’t diagnosed until the cancer has already advanced and spread beyond the pancreas, at which point it’s much harder to treat. Finding ways to improve early diagnosis of pancreatic cancer is, therefore, a key goal for scientists working to improve outcomes in this disease.
To visualize the pancreas, clinicians often rely on CT scans of the abdomen. This is pretty reliable for detecting large, advanced tumors, but when a tumor is in early stages, it’s usually too small to be clearly visible to experts interpreting the scan.
Now, researchers are exploring whether AI could help experts make more accurate calls at early stages. AI is a broad category of computational tools that work by feeding a computer a large dataset, which it uses to identify patterns that can then be applied to future datasets. The basic idea is that AI-based computer systems might be able to detect minute changes indicative of cancer that aren’t readily seen by even the most experienced human expert.
The AI tool used in this study is called the Radiomics-based Early Detection MODel, or REDMOD for short. To train the program, researchers used standard CT scans from nearly 1,000 people — 156 who went on to be diagnosed with pancreatic cancer, and 813 who did not have cancer.
“By leveraging a large clinically representative dataset and advanced AI methodologies, this work establishes the clinical validity of the REDMOD framework for pre-clinical [pancreatic cancer] detection,” the scientists wrote.
The researchers then tested their algorithm using CT scans from a separate group of nearly 500 patients, among whom 63 were later diagnosed with pancreatic cancer.
When experts reviewed these scans without AI assistance, they correctly identified just over a third (38.9%) of the cancer cases. However, with an assist from REDMOD, nearly three-quarters (73%) of the cancer cases were accurately identified, with a median time to diagnosis of more than a year.
When looking only at scans taken at least two years before diagnosis, experts using the AI program accurately identified about two-thirds (68%) of cancer cases, whereas experts without AI were able to identify less than a quarter (23%) of cases. The researchers also noted that the AI’s accuracy was consistent across scans from different centers and when analyzing scans from the same patient taken at different points in time.
Moving toward clinical use
“This study validates REDMOD as a fully automated AI framework that can identify the imaging signature of [early-stage pancreatic cancer] from routine CT scans with substantial lead times and performance superior to expert radiologists,” the scientists concluded. “The demonstrated ability of the framework to consistently detect these occult signals on a large clinically-oriented dataset, combined with its high longitudinal stability and validated specificity, establishes a robust foundation for AI-augmented early detection.”
Researchers at Mayo are currently running a study called Artificial Intelligence for Pancreatic Cancer Early Detection (AI-PACED) to prospectively test how this AI tool could be used to improve patient care.
“While prospective validation is paramount to confirm clinical utility, the REDMOD framework represents a significant advance towards shifting the paradigm for [pancreatic cancer] from a late-stage symptomatic diagnosis to proactive pre-clinical interception, offering tangible hope for improving outcomes in this challenging disease,” the researchers wrote.
