The post has been translated automatically. Original language: English
A new study shows you don’t actually need CT to estimate core body-composition metrics. Here’s the data from a 1,118-patient cohort.
Using nothing but:
→ a frontal CXR
→ age, sex, height, weight
The model predicted metrics that normally require an abdominal CT.
The numbers:
-> Subcutaneous fat: r = 0.85
-> Visceral fat: r = 0.76
-> Vertebral bone volume: r = 0.72
No abdominal CT. No dedicated fat windows. Just a 2D chest radiograph.
The performance is surprisingly strong - which suggests we still don’t fully understand how much hidden biological signal plain X-rays actually contain.
The big question:
How do we systematically uncover these signals?
Is it really just brute-forcing patterns from 3D data into 2D?
DOI: 10.1093/radadv/umaf035
A new study shows you don’t actually need CT to estimate core body-composition metrics. Here’s the data from a 1,118-patient cohort.
Using nothing but:
→ a frontal CXR
→ age, sex, height, weight
The model predicted metrics that normally require an abdominal CT.
The numbers:
-> Subcutaneous fat: r = 0.85
-> Visceral fat: r = 0.76
-> Vertebral bone volume: r = 0.72
No abdominal CT. No dedicated fat windows. Just a 2D chest radiograph.
The performance is surprisingly strong - which suggests we still don’t fully understand how much hidden biological signal plain X-rays actually contain.
The big question:
How do we systematically uncover these signals?
Is it really just brute-forcing patterns from 3D data into 2D?
DOI: 10.1093/radadv/umaf035
Share
