Automated CT Scans to Speed Up Clinical Evaluations

New machine learning model could expand what medical scans can tell us about disease.

Published on Mar. 5, 2026

Got story updates? Submit your updates here. ›

A research team funded by the National Institutes of Health (NIH) has developed a versatile machine learning model called Merlin that could greatly expand what medical scans can tell us about disease. Merlin was trained on a large dataset of abdominal CT scans, radiology reports, and medical diagnosis codes, and was able to outperform specialized models in a variety of tasks including identifying anatomical features, predicting disease onset, and interpreting chest scans despite not being trained on them.

Why it matters

The new Merlin model has the potential to significantly streamline medical workflows and assist in clinical decision-making by automating many tasks currently performed by radiologists. This could help address the growing shortage of physicians in the United States and make medical evaluations more efficient.

The details

The researchers trained Merlin on a unique dataset of over 15,000 3D abdominal CT scans paired with radiology reports and diagnostic codes. They then tested Merlin on more than 50,000 previously unseen scans from four different hospitals. Merlin was able to successfully predict diagnostic codes over 81% of the time on average, and up to 90% for a subset of codes. It also outperformed specialized models in predicting the onset of chronic diseases like diabetes and heart disease based solely on CT scans. Remarkably, Merlin was even able to interpret chest scans, a body part not included in its training data.

  • The research was conducted in 2026.

The players

Merlin

A versatile machine learning model developed by a research team funded by the National Institutes of Health (NIH) that can perform a variety of tasks on medical CT scans.

National Institutes of Health (NIH)

The U.S. government agency that provided funding for the research team that developed the Merlin model.

Stanford University

The university where the research team that developed Merlin is based.

Bruce Tromberg, Ph.D.

The director of NIH's National Institute of Biomedical Imaging and Bioengineering (NIBIB).

Louis Blankemeier, Ph.D.

A co-first author of the study who conducted the research while a graduate student at Stanford University.

Ashwin Kumar

A co-first author of the study and a graduate student at Stanford University.

Akshay Chaudhari, Ph.D.

The senior author of the study and a professor of radiology and biomedical data science at Stanford University.

Got photos? Submit your photos here. ›

What they’re saying

“Rich datasets like this are necessary to push the limits of what artificial intelligence models can accomplish in medicine. This work exemplifies how meticulously crafted training data can enable remarkable insights that significantly streamline workflows and assist in clinical decision-making.”

— Bruce Tromberg, Ph.D., Director of NIH's National Institute of Biomedical Imaging and Bioengineering (NIBIB) (Mirage News)

“With Merlin, you could potentially go beyond traditional radiology and jump straight from imaging to a possible diagnosis. And that's just one potential use.”

— Louis Blankemeier, Ph.D., Co-first author of the study (Mirage News)

“Merlin tackled some tasks, such as predicting diagnosis codes, head-on, while other more complicated tasks, such as drafting radiology reports from scratch or identifying and outlining organs in a 3D space, called for additional training.”

— Ashwin Kumar, Co-first author of the study (Mirage News)

“Our model and the data will provide the community a robust backbone to build upon. From here, the sky's the limit.”

— Akshay Chaudhari, Ph.D., Senior author of the study (Mirage News)

What’s next

The researchers plan to refine Merlin to better handle more complicated challenges, such as writing radiology reports, and encourage users to fine-tune the model with their own data to address their specific needs.

The takeaway

The Merlin model represents a significant advancement in medical imaging technology, with the potential to greatly streamline clinical workflows and assist in disease diagnosis and prognosis. Its ability to outperform specialized models across a wide range of tasks highlights the power of foundation models trained on large, diverse datasets.