<p>Every year, hundreds of millions of abdominal CT scans are performed worldwide — for kidney stones, appendicitis, bowel conditions, and countless other reasons. The radiologist reads the scan for what they're looking for and moves on. An enormous amount of information in those images goes unread.</p>
<p>A team at Stanford University has built an AI system called Merlin that reads it all. And what it finds, it turns out, can predict whether you'll develop diabetes, heart disease, or osteoporosis within the next five years — before any symptoms appear.</p>
<h2>Five Years of Warning, Hiding in Plain Sight</h2>
<p>Merlin was trained on more than 15,000 abdominal CT scans alongside radiology reports and nearly a million diagnostic codes. It was then tested on more than 50,000 scans from hospitals across multiple sites. The results, published in <em>Nature</em> on March 4, 2026, are striking.</p>
<p>The AI correctly predicted which patients would develop chronic conditions within five years with approximately 75% accuracy — substantially better than standard clinical assessments. In some cases, where the accuracy exceeded 90%, it outperformed every other AI system tested on the same task.</p>
<p>The diseases it can flag early include diabetes, cardiovascular disease, and osteoporosis — three of the leading causes of preventable illness and premature death worldwide. All three are far more treatable when caught early. All three currently go undetected in enormous numbers of patients until significant damage has already occurred.</p>
<h2>What Merlin Sees That Radiologists Don't</h2>
<p>The AI appears to detect subtle changes in tissue density, organ composition, and structural features that are not routinely reported in clinical radiology. These features may represent very early biological shifts — years before a disease becomes symptomatic or diagnosable by conventional means.</p>
<p>"Merlin can detect things that are too subtle or too diffuse for the human eye to notice," said one of the researchers involved in the study. "It's not replacing the radiologist — it's adding a layer of analysis that no human could perform at this scale."</p>
<p>The system is classified as a "foundation model" — meaning it's a general-purpose AI capable of performing many tasks, not just one. Beyond prediction, Merlin can diagnose existing conditions with over 81% average accuracy and can generate outputs for more than 200 medical tasks from a single scan.</p>
<h2>Turning Routine Scans Into Preventive Medicine</h2>
<p>The implications are significant. CT scans are already being performed for other reasons on millions of patients who may have pre-disease changes visible in their images. If Merlin — or a system like it — were integrated into clinical workflow, it could flag those patients for early intervention without requiring them to undergo any additional tests.</p>
<p>The NIH funding behind the research reflects a broader push to make preventive medicine more data-driven. Stanford and several partner institutions are now working on refining Merlin for more complex tasks, including generating radiology reports and flagging findings for physician review.</p>
<p>Clinical deployment is still in the future — regulatory approval and prospective validation studies are needed. But the core finding is clear: the information to predict disease years in advance is already there in the scans. We're just beginning to read it.</p>
<p><em>Sources: Nature · Stanford Medicine · NIH · Baptist Health, March 2026</em></p>
