in News
A new artificial intelligence tool could help detect pancreatic cancer earlier, a study suggests.(Image credit: SEBASTIAN KAULITZKI/SCIENCE PHOTO LIBRARY via Getty Images)Share this article 0Join the conversationFollow usAdd us as a preferred source on GoogleSubscribe to our newsletter
A novel artificial intelligence (AI) model may assist medical professionals in identifying pancreatic cancer up to three years before physicians typically notice tumors on CT scans, according to recent research.
The program, detailed on April 28 in the journal Gut, was utilized to examine nearly 2,000 CT scans that had previously been classified as “normal,” showing no indications of illness. The instrument detected minute anomalies in the pancreas’s structure that subsequently evolved into tumorous tissue.
A chance to detect cancer early
Pancreatic cancer stands as one of the most lethal forms of cancer.
“The five-year survival rate [in the U.S.] is approximately 12% to 13% due to our inability to diagnose it at a stage when therapeutic interventions could prove effective,” study co-author Dr. Ajit Goenka, a radiologist and nuclear medicine specialist at the Mayo Clinic in Rochester, Minnesota, informed Live Science. The initial phases of pancreatic cancer frequently do not manifest any symptoms, leading to the disease often being advanced by the time of diagnosis.
While medical professionals’ capacity to detect and treat numerous other cancers has advanced in recent decades, a comparable breakthrough has not been observed in pancreatic cancer. Diagnosis typically involves a combination of tissue sampling and imaging examinations, including CT scans. However, by the time tumors become visible through these methods, the cancer is frequently in its terminal stages.
Nevertheless, earlier indicators of the disease might exist.
“Basic science research indicates that the process of cancer development does not commence six months prior,” Goenka stated. “It begins 10 to 15 years earlier, implying that a signal was present in the pancreas, and that signal was beyond the scope of human detectability.”
Ultimately, it is a matter of mathematics. It translates that image into a mathematical depiction and extracts those mathematical characteristics.
Dr. Ajit Goenka, radiologist and nuclear medicine specialist at the Mayo Clinic in Rochester, Minnesota
By harnessing AI to recognize patterns imperceptible to humans, Goenka and his associates devised a tool to amplify that existing signal and pinpoint early disease indicators in CT scans.
The model, designated the Radiomics-based Early Detection Model (REDMOD), essentially transforms the CT scan image into a mathematical challenge. It initially segments the organ, constructing a three-dimensional model of the pancreas from the two-dimensional images acquired by the CT apparatus. Subsequently, it scrutinizes the resultant structure on a pixel-by-pixel basis.
“It examines every single pixel within that image and quantifies the extent to which it deviates from the remainder of the organ, and then it contrasts this with controls where such a deviation is not anticipated,” Goenka elaborated. “In essence, it is mathematics. It converts that image into a mathematical representation and extracts those mathematical features.”
The team evaluated the model using a dataset of 2,000 existing CT scans, which were previously collected for medical reasons unrelated to cancer and had all been deemed normal. Approximately one-seventh of these scans belonged to patients who subsequently developed pancreatic cancer.
The model successfully identified 73% of these early-stage cases, and on average, the scans analyzed by the model had been taken 16 months prior to the individual’s actual diagnosis.
“The enhancement in sensitivity over radiologists was nearly twofold across the entire range, and when examining even earlier stages — more than two years before diagnosis — that sensitivity increase was almost threefold,” Goenka remarked. In other words, the AI tool accurately detected cancer cases earlier than radiologists did, and the further back in time one looked, the greater this performance disparity became.
Next steps
Nonetheless, the AI tool still has scope for refinement. “The radiologist was less inclined to incorrectly flag a healthy patient,” Goenka observed. The model correctly identified disease-free patients 81.1% of the time, compared to an average of 92.2% for human radiologists. “Therefore, a collaborative function exists for both, integrating physician expertise with AI augmentation.”
The study was meticulously designed and yielded exceptionally encouraging findings, according to Tatjana Crnogorac-Jurcevic, a professor of molecular pathology and biomarkers at Queen Mary University of London, who was not involved in the research.
“Such early detection would represent a significant shift in the clinical management of patients,” she informed Live Science. “As pancreatic cancer is relatively uncommon, general screening, as currently employed for colon and breast cancer, would not be practical. However, there are identified high-risk cohorts for whom surveillance would be feasible — individuals with a family history of pancreatic cancer, those with other cancer mutations, and patients experiencing new-onset diabetes.”
Related stories
- ‘A second set of eyes’: AI-supported breast cancer screening spots more cancers earlier, landmark trial finds
- Detecting cancer in minutes possible with just a drop of dried blood and new test, study hints
- ‘Fingerprints of cancer’ found after scientists flash infrared light pulses at blood samples
Goenka anticipates that the model could be integrated into routine clinical practice within the next five years, and the team is presently conducting clinical trials to further confirm the practical efficacy of this detection strategy.
Looking ahead, integrating this REDMOD with other diagnostic methodologies could lead to even more substantial advancements in early detection, Crnogorac-Jurcevic suggested.
“We are developing urine-based tests with the identical objective, and combining an AI imaging tool with our body fluid biomarkers would be extremely beneficial,” she stated. “It is highly probable that they will be complementary, thereby vastly enhancing the sensitivity and precision of early detection.”