Artificial intelligence is no longer just promising in cancer diagnostics — it is demonstrating, in peer-reviewed clinical studies, the ability to detect cancers earlier, more accurately, and from more challenging tissue presentations than human experts alone. Two major studies published in April and May 2026 have pushed the boundaries of what AI-assisted cancer detection can accomplish and offer a glimpse at how the field of oncology may be transformed within the next decade.
The first, published in Nature Cancer by a team at the Hong Kong University of Science and Technology, introduced PRET — a "plug-and-play" AI pathology system capable of recognizing 18 distinct cancer types from just a handful of tissue slides, without requiring any additional training specific to the institution or patient population where it is deployed. The second, published in Gut by Mayo Clinic researchers, demonstrated an AI system that detected pancreatic cancer from routine CT scans up to three years before clinical diagnosis — nearly doubling the detection rate achieved by specialist radiologists reviewing the same scans without AI assistance.
PRET: Recognizing 18 Cancer Types Without Additional Training
The shortage of pathologists is a global crisis with direct consequences for cancer outcomes. Pathological examination of tissue samples is required for virtually all cancer diagnoses, yet there is a severe and growing deficit of trained pathologists capable of analyzing the increasing volume of samples generated by aging populations worldwide.
PRET addresses this shortage by enabling precise pan-cancer recognition from minimal tissue input without requiring institution-specific training data. In evaluations reported in the Nature Cancer paper, the system achieved remarkable accuracy across multiple cancer types: an AUC of 100 percent in colorectal cancer screening and 99.54 percent in esophageal squamous cell carcinoma tumor segmentation. In the challenging task of lymph node metastasis detection — identifying whether cancer has spread to lymph nodes, a critical staging question — PRET also achieved high accuracy.
The system's "plug-and-play" characteristic is its most practically important feature. Existing AI pathology systems typically require extensive training on local datasets to achieve acceptable performance, limiting their deployment in lower-resource settings that lack the infrastructure to generate and curate large training datasets. PRET's ability to perform accurately on new populations without retraining substantially reduces this barrier.
The research team indicated plans to expand PRET's applications to genetic mutation prediction from tissue slides — a capability that could enable more precise matching of patients to targeted therapies without requiring separate molecular testing — and to patient prognosis assessment.
Mayo Clinic's AI: Seeing Cancer Before It Looks Like Cancer
The challenge of pancreatic cancer diagnosis is one of oncology's most urgent unsolved problems. Pancreatic cancer is on course to be the second-leading cause of cancer-related death in the United States by 2030, primarily because 85 percent of cases are not diagnosed until the cancer has spread beyond the pancreas, when surgical cure is no longer possible. At the localized stage, the five-year survival rate exceeds 40 percent; at the metastatic stage, it falls below 3 percent.
The Mayo Clinic AI system, described in research published in Gut on April 28, 2026, analyzes routine CT scan images to identify subtle tissue patterns associated with pancreatic cancer that are invisible to the human eye at the time of imaging but that, in retrospect, precede clinical diagnosis by months to years. In a landmark validation study, the system detected pancreatic cancer in nearly three out of four cases approximately 16 months before clinical diagnosis — nearly double the detection rate of specialist radiologists reviewing the same scans independently.
In some cases, the system identified suspicious patterns more than two years before diagnosis, and the Mayo research team believes detection windows of up to three years may be achievable. Crucially, the AI's predictions were stable across multiple scans of the same patient taken months apart — a critical validation that the system is detecting real biological patterns rather than random noise.
"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," said Dr. Ajit Goenka, lead researcher at Mayo Clinic. The team is advancing the AI into prospective clinical testing through the AI-PACED study, which will evaluate how clinicians can integrate AI-guided detection into care for patients at elevated risk.
What This Means for Patients — and the Path to Clinical Deployment
For individual patients, the near-term implications of these AI systems are not yet fully realized. PRET is a research system, not yet cleared by the FDA for clinical use in the United States, though the publication of strong accuracy data in Nature Cancer will support future regulatory submissions. The Mayo pancreatic cancer AI is entering prospective clinical testing rather than broad clinical deployment — validating its performance in a real-world setting before it becomes a standard clinical tool.
However, the trajectory is clear. AI systems are demonstrating performance that rivals or exceeds specialist humans in specific cancer detection tasks, and the FDA has already cleared a number of AI diagnostic algorithms for use in radiology, dermatology, and ophthalmology. As these systems mature, the vision of AI-assisted early cancer detection — catching malignancies when they are most curable, from routine imaging or tissue samples that would otherwise be evaluated without AI assistance — is becoming measurably closer to clinical reality.
Frequently Asked Questions
Q: Can AI actually detect cancer before symptoms appear?
A: Yes, in specific contexts. The Mayo Clinic AI detected pancreatic cancer from CT scans up to 3 years before clinical diagnosis in approximately 75% of cases in a validation study.
Q: What is PRET and what types of cancer can it detect?
A: PRET is an AI pathology system published in Nature Cancer in April 2026 that can recognize 18 cancer types from tissue slides without requiring institution-specific training. It demonstrated 100% accuracy in colorectal cancer screening in evaluations.
Q: When will these AI systems be available for clinical use?
A: The Mayo pancreatic cancer AI is entering prospective clinical trials as of 2026. PRET is in research validation. Clinical deployment would require FDA clearance and clinical validation, typically taking 2-5 years after initial research publication.
Q: Does AI pathology replace human pathologists?
A: Current evidence suggests AI works best alongside pathologists, augmenting human diagnostic accuracy rather than replacing it. The AUC data from PRET suggests it can handle triage and screening tasks at high accuracy, potentially reducing the workload burden on pathologists.
Q: What should people at elevated risk for pancreatic cancer do?
A: People with a family history of pancreatic cancer, certain genetic mutations (BRCA2, Lynch syndrome, others), chronic pancreatitis, or new-onset diabetes after age 50 should discuss surveillance options with a gastroenterologist or genetic counselor.
