The same week that a new philanthropic push pledged $500 million to build artificial intelligence (AI) systems that model disease at the cellular level, two separate research teams published evidence that AI can already see what radiologists cannot.

Mayo Clinic’s Model Reads a Normal Scan Differently

The numbers behind Mayo Clinic’s new AI are striking.

A Mayo Clinic-developed model called REDMOD can detect pancreatic cancer on routine abdominal CT scans up to three years before a clinical diagnosis, identifying subtle signs of disease before tumors are visible, when curative treatment may still be possible, Mayo Clinic News Network reported Wednesday (April 29).

Researchers analyzed nearly 2,000 CT scans, including scans from patients later diagnosed with pancreatic cancer that were originally interpreted as normal. REDMOD identified 73% of those prediagnostic cancers at a median of about 16 months before diagnosis, nearly double the detection rate of specialists reviewing the same scans without AI assistance.

The advantage compounded over time. In scans obtained more than two years before diagnosis, the AI identified nearly three times as many early cancers that would otherwise go undetected, Mayo Clinic News Network reported.

The stakes are real. More than 85% of pancreatic cancer patients receive a diagnosis after the disease has already spread, and five-year survival rates remain below 15%, according to the National Cancer Institute. Projections show pancreatic cancer will become the second-leading cause of cancer-related death in the U.S. by 2030.

REDMOD works by measuring quantitative imaging features that describe tissue texture and structure, capturing faint biological changes before any visible mass forms. The model runs automatically without time-intensive manual preparation and was validated across CT scans from multiple institutions, imaging systems and protocols, Mayo Clinic News Network reported. It is designed to analyze scans already obtained for other reasons, particularly in high-risk patients such as those with new-onset diabetes.

Alibaba’s Tool Outpaces Radiologists on a Different Cancer

The same week, Alibaba’s research arm Damo Academy published results for a parallel effort targeting colorectal cancer. Its Coca AI model accurately identified five previously missed cases of colorectal cancer from the non-contrast CT scans of more than 27,000 people, achieving a sensitivity of 86.6% and a specificity of 99.8%.

Damo Academy said Coca outperformed 10 radiologists of varying experience levels by 20.4% on sensitivity, the South China Morning Post reported. The research was developed with Chinese institutions including Guangdong General Hospital and published in the Annals of Oncology. Current diagnostic methods for colorectal cancer, including colonoscopy and CT colonography, are invasive and create discomfort for patients, whereas Coca has the potential to be a non-invasive, cost-effective and scalable tool, according to Damo researchers.

Biohub Bets $500 Million on the Deeper Problem

The detection breakthroughs matter. But the infrastructure bet announced the same week signals where the field is heading.

The Chan Zuckerberg Biohub committed $500 million over five years to AI-driven biology, with $400 million going toward its own work and $100 million aimed at spurring others, Axios reported Wednesday. According to the report, Zuckerberg said last year that Biohub’s long-term goal is to cure all human disease through the intersection of AI and biology.

The stated goal is not incremental. Biohub chief Alex Rives told Axios that the usefulness of an AI model’s prediction increases exponentially as the scale of the data grows. Most current datasets cover about a billion cells. The aim is to reach an order of magnitude or more beyond that, Rives said.

Biohub is focused on frontier AI and frontier biology, using large-scale models for virtual cells, immune reprogramming and disease prediction, Fortune reported. The ambition sits upstream of what REDMOD and Coca are doing. Scan-reading AI catches disease once the body has already started changing. A virtual cell model, if it works as theorized, could predict which patients will reach that point at all.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.