AI model bests radiologists at detecting early signs of pancreatic cancer
A newly developed artificial intelligence model has the potential to significantly improve the chances of detecting pancreatic cancer in its earliest stages.
The Radiomics-based Early Detection MODel, or REDMOD for short, is an AI framework that was designed to identify radiomic signatures of pre-diagnostic pancreatic ductal adenocarcinoma (PDAC) on CT scans; it is capable of detecting patterns indicative of the disease that are not typically visible to the human eye. This makes it a promising development for improving pancreatic cancer outcomes, as it is often identified in later stages, leading to poor survival rates.
REDMOD was trained on imaging data, which included ample exams that were acquired prior to PDAC diagnoses, from multiple institutions. This helped experts develop a 40-feature radiomic signature. It was tested on over 200 abdominal CT scans of patients who were initially told their imaging showed no signs of PDAC, only to be diagnosed in the following years.
Testing revealed the model capable of detecting pancreatic cancer an average of 475 days prior to diagnosis; it did so with a sensitivity of nearly 75%, which was twofold higher than radiologists. At a 24-month lead time, REDMOD’s sensitivity was three times higher than that of radiologists.
“This temporal window holds profound significance, as attaining such early detection would substantially augment the probability of cure and improved survival,” Harishkumar Goenka, MD, with the department of radiology at Mayo Clinic, Rochester, Minnesota, and colleagues noted. “In fact, modelling studies indicate that increasing the proportion of localized [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.”
The model’s performance also was maintained across exams conducted at multiple institutions using CT scanners from different manufacturers. This highlights its generalizability. Its accuracy was consistent as well, as it correctly identified the same patients with pre-diagnostic cancer 92% of the time on subsequent scans completed months later.
Although further validation is needed, the team is optimistic about their model’s future clinical potential.
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.