Can AI Models Improve Detection of Cancer?

Awakend.ai explores the frontier of AI: medical systems catching disease before it strikes, and the broader evolution of AI moving from chatbot to coworker.

Most cancer research assumes catching pancreatic cancer early requires a new scan, a new biomarker, or some breakthrough not yet invented. But today's analysis asks whether the answer was already sitting in the CT scan every patient already gets, just hidden in patterns beyond what the human eye can detect.

This issue also looks at OpenAI turning ChatGPT into an agent that works alongside you, an AI system that rewrote its own code and beat two years of human tuning, and Demis Hassabis's case that we're closer to a turning point than most people think.

Every year, cancer kills nearly 600,000 Americans, making it the second leading cause of death in the US. For 50 years, survival rates for the deadliest types have barely moved, not because treatment failed, but because the cancer is almost never found in time. But Medical AI is now starting to solve that.

Currently, only 14% of cancers in the US are found through recommended screening tests. For many cancers, that means diagnosis comes only after symptoms appear, when treatment is less effective and survival rates have already fallen. No cancer illustrates that gap more starkly than pancreatic cancer. 

It is the only major cancer with a five-year overall survival rate of just 11%, and more than 85% of cases are diagnosed only after the cancer has spread, when surgery is no longer an option. But if it gets caught at the earliest stage, the survival rate can exceed 80%.

The CT scan is the standard imaging tool for the abdomen and sits in every hospital. A specialist reviews the pancreas, sees nothing unusual, and the patient goes home. Meanwhile, the cancer has been developing in the tissue for 10 to 15 years, producing a biological signal that is, in the words of Dr Ajit Goenka, "outside the purview of human detectability."

This is the gap medical AI is closing. An AI model does not directly look for a tumor but first draws a precise border around the pancreas, separating it from the surrounding tissue. Then it reads the texture of that region feature by feature, looking for patterns in tissue architecture that fall outside what the human eye can detect. 

The researchers describe these as imaging biomarkers beyond the human-perceptible range, changes too subtle to see but measurable in the scan's mathematics.

Mayo Clinic's REDMOD was tested on 219 scans that specialists had already reviewed and found nothing. But the model caught 73% of cancers hidden in those same scans, at a median of 16 months before any diagnosis. It found nearly three times as many early cancers as the specialists who had cleared them.

This is not just one study. The PANORAMA trial tested AI against 68 radiologists across 12 countries on standard CT scans. AI achieved 7% more accuracy than the radiologists reading the same images.

Harvard Medical School found that AI reading only a patient's medical records, with no scan at all, could identify who was most likely to develop pancreatic cancer up to three years before diagnosis.

AI pancreatic cancer tools have been receiving FDA approvals since 2025, and the AI-PACED trial launched in March 2026 to test REDMOD in real-world patient care.

Takeaway: For 50 years, a clean CT scan meant the pancreas was healthy, while the cancer was already forming in the tissue long before any expert could see it. That gap is where most patients were lost, and it is now getting thinner. 

The AI-PACED trial is underway, and if the results hold, catching pancreatic cancer before it spreads stops being a matter of luck and becomes a matter of timing, with implications that stretch well beyond the pancreas.

AI Cancer Imaging Is on Track to 7x by 2034

Takeaway:
The market is not growing just because AI got smarter. Medicine is running out of the people it needs to read scans. The UK alone faces a 30% shortfall of radiologists today, projected to reach 40% by 2028. AI imaging is not replacing them—it is helping them keep pace, handling growing scan volumes, catching what human eyes miss, and giving patients a diagnosis before the window for effective treatment closes.

The Timeline Guide:

Each timeline and graph represents the realistic stage of the covered technology, plotted from concept to scale, capturing where it stands today and when broad deployment is likely.

Hit reply and tell us why; we actually read every note.
See you in the next upload