Machine learning analysis of CT scans

3 hours ago

March 17, 2026

At a Glance

Researchers developed a machine learning model that can interpret abdominal CT scans to make a diagnosis and possibly predict risk for some chronic diseases.
The model could cut down on the time it takes to analyze and interpret CT scans.

Scientists developed an AI-powered tool that can interpret 3D images from CT scans and diagnose certain abdominal disorders.

YAKOBCHUK VIACHESLAV / Shutterstock

Computed tomography (CT) scans are a common type of 3D medical imaging that can be used to identify and diagnose tumors, infections, and other health problems. But a radiologist must interpret the scans, which can take as long as 20 minutes per scan. As use of CT scans has increased, the number of radiologists has not kept up. Thus, it can be a lengthy process to get a diagnosis, since radiologists have more scans to interpret.

Machine learning models that can analyze images from scans may be able to accurately interpret CT scans. But current models that can interpret medical images are primarily limited to 2D images, like X-rays. An NIH-funded research team, led by Dr. Akshay Chaudhari of Stanford University, developed an AI-powered tool that can interpret 3D images. A description of the tool, called Merlin, along with demonstrations of its capabilities were published on March 4, 2026, in Nature.

The researchers trained Merlin by using more than 15,000 abdominal CT scans, along with their radiology reports, and nearly 1 million associated diagnosis codes. They then tested the tool with more than 50,000 abdominal CT scans from different hospitals and publicly available datasets. Merlin’s performance was evaluated for six different types of tasks.

For diagnosis, the team checked to see if Merlin could reproduce the results from radiologists. On average across hundreds of diagnosis codes, Merlin predicted the correct diagnosis with more than 81% accuracy. For 102 of the codes, Merlin’s accuracy was more than 90%.

The team also tested Merlin’s ability to predict whether healthy patients would develop a chronic disease within five years based on CT scans. Across six chronic diseases, Merlin predicted which patients would develop the diseases in the next five years with 75% accuracy. This suggests that Merlin can detect features in CT scans that are not detectable by the eyes of trained radiologists.

To test Merlin’s generalizability, the researchers used it to analyze CT scans of the chest, which were not used in training the model. Still, Merlin performed as well as or better than existing models that were trained specifically on chest scans.

The team found that Merlin struggled with some more complex tasks. When generating radiology reports based on CT scans, Merlin tended to under-report findings. Merlin also struggled to identify and outline organs in 3D space.

The results suggest that Merlin could assist in interpreting CT scans and reduce demands on radiologists. The researchers hope to get approval to use Merlin for simpler tasks in the clinic. They also plan to refine the model to handle more complicated tasks, such as writing radiology reports. In the meantime, they have made their model, code, and dataset available for other researchers.

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

References

Merlin: a computed tomography vision-language foundation model and dataset. Blankemeier L, Kumar A, Cohen JP, Liu J, Liu L, Van Veen D, Gardezi SJS, Yu H, Paschali M, Chen Z, Delbrouck JB, Reis E, Holland R, Truyts C, Bluethgen C, Wu Y, Lian L, Jensen MEK, Ostmeier S, Varma M, Valanarasu JMJ, Fang Z, Huo Z, Nabulsi Z, Ardila D, Weng WH, Junior EA, Ahuja N, Fries J, Shah NH, Zaharchuk G, Willis M, Yala A, Johnston A, Boutin RD, Wentland A, Langlotz CP, Hom J, Gatidis S, Chaudhari AS. Nature. 2026 Mar 4. doi: 10.1038/s41586-026-10181-8. Epub ahead of print. PMID: 41781626.

Funding

NIH’s National Institute of Biomedical Imaging and Bioengineering (NIBIB), National Heart, Lung, and Blood Institute (NHLBI), National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS); Advanced Research Projects Agency for Health (ARPA-H); Medical Imaging and Data Resource Center; Promedica Foundation.