A new study lays the groundwork for AI-based opportunistic screening – the detection of disease using medical images acquired for other indications. In a paper in AJR, researchers show how their homegrown AI algorithm was able to analyze abdominal CT scans and link body composition measurements to the presence of disease.
Opportunistic screening is a sort of holy grail for radiology, with the potential to help radiologists find pathology from scans ordered for other clinical indications.
- Some researchers specifically are focusing on analysis of body composition characteristics derived from CT scans like muscle, fat, and bone that could be biomarkers for hidden pathology – and AI is key because it can process mountains of patient data without getting tired.
In the new paper, researchers from the NIH and the University of Wisconsin tested the concept of AI-based body composition analysis on a massive database of 118k patients who got abdominal CT scans from 2000 to 2021.
- They analyzed the scans with their own internally developed AI tool that measures 13 features of body composition, from volume and attenuation in different organs to area of subcutaneous adipose tissue.
Their goal was to correlate the AI measurements with actual presence of disease, as well as other factors that could affect body composition like age and sex. They found …
- AI-based body composition metrics varied by age and sex, confirming previous studies.
- AI metrics also correlated with the four systemic diseases studied, specifically cancer, cardiovascular disease, diabetes mellitus, and cirrhosis.
- The predictive power of different metrics varied by disease, from a high of 13 measures for diabetes to a low of nine for cancer.
What’s the real-world impact of the study?
- In addition to validating the concept of AI-based opportunistic screening on a broad scale, the findings could be used to establish a set of normal values for body composition that also take into account the impact of systemic disease on these measurements.
The Takeaway
The new study is a bit technical, but it’s an important milestone on the path to opportunistic screening. It not only demonstrates the concept’s feasibility, but also begins to establish the normal values needed to actually implement screening programs in the real world.
