Beyond BMI: AI Uses Routine CT to Measure Muscle and Visceral Fat for Personalized Risk
AI-enhanced CT body composition analysis can quantify muscle and fat compartments in seconds, revealing risk profiles that BMI may miss.
Body mass index (BMI) is widely used because it's fast and simple. But BMI cannot distinguish fat from muscle or show where fat is stored—two factors that can strongly influence outcomes in cancer care, cardiac interventions, surgery, critical care, and clinical nutrition. An opinion paper argues that AI-powered analysis of routine CT scans can provide a more clinically informative alternative by directly measuring muscle and fat compartments.
FAST FACTS
• BMI can misclassify risk because it doesn't separate muscle from fat or capture visceral fat distribution.
• AI can extract CT-based body composition measures (SMI, VAT, SAT, fat density) from scans already acquired
for clinical care—often within seconds.
• Case examples show that "underweight" or "overweight" BMI labels can hide very different muscle-and-fat
risk profiles.
• The approach is promising but not fully standardized: measurement methods, scanners, and cut-offs vary.
BMI is simple—but it misses key biology
BMI reduces body size to a single number. What it cannot show includes:
• Sarcopenic obesity (low muscle with excess fat)
• Visceral fat predominance (fat concentrated around organs)
• Fat quality (CT density features linked to tissue characteristics)
These patterns can affect survival, recovery after procedures, and treatment tolerance—yet BMI may group
patients with very different composition profiles together.
A CT + AI alternative: measure what matters
The opinion paper proposes AI-enhanced CT body composition analysis as a more individualized risk tool.
Using a routine CT image at the L3 (lower back) level, AI can quantify:
• SMI (skeletal muscle index)
• VAT (visceral adipose tissue)
• SAT (subcutaneous adipose tissue)
• VAT density (in Hounsfield Units, HU)
Across oncology, cardiology, surgery/critical care, and nutrition, these measures are presented as stronger predictors of outcomes than BMI alone.
Opportunistic workflow: fast, scalable, no additional imaging
Hospitals already perform CT scans for valid clinical reasons. The same images can be analyzed "opportunistically," meaning no extra scan is required in many cases.
The paper contrasts older and newer workflows:
• Manual segmentation: ~15–20 minutes
• AI segmentation: seconds
• Example service reported: ~21 seconds from upload to results display via a web interface
This turns existing images into structured measurements that can fit into routine care if reporting and decision pathways are integrated cleanly.
Results: composition can flip the risk story
The paper describes real-case contrasts illustrating why body composition matters:
• A patient labeled "underweight" by BMI showed very low muscle and higher-risk fat features on CT.
• A patient labeled "overweight" by BMI showed preserved muscle and a lower-risk fat profile.
In transcatheter aortic valve implantation (TAVI), the paper reports CT-based measures that predicted
all-cause mortality beyond BMI, including:
• SMI (HR 0.986)
• VAT density (HR 1.015)
• SAT density (HR 1.014)
It also cites VAT density thresholds reported in prior work as higher-risk flags:
• Men: > −93.27 HU
• Women: > −95.02 HU
Where CT body composition could add value
The authors argue that CT body composition can support more precise risk stratification in:
• Cancer care (treatment tolerance, prognosis, recovery)
• Cardiac procedures (frailty and mortality risk beyond BMI)
• Surgery and critical care (complication risk, recovery trajectories)
• Clinical nutrition (early detection of low muscle reserves)
Implementation: start focused, build standards
The paper recommends stepwise adoption centered on workflow rather than complexity:
• Begin with high-risk groups already receiving CT scans
• Establish local reference values and a concise reporting format
• Integrate results into radiology workflows
• Create decision pathways (screen → comprehensive assessment when needed)
• Train clinicians on interpretation and follow-up actions
Promise and precautions
The field is advancing quickly, but the paper notes key limitations:
• Methods differ across studies and centers
• Scanner settings and image reconstruction can influence measurements
• Universal cut-offs across populations are not established
• Many implementations come from specialized centers
The authors call for multicenter validation, consensus reporting standards, and population-specific
reference values.
More information:
Matej Pekar et al, Beyond BMI: An opinion on the clinical value of AI-powered CT body composition analysis. Biomol Biomed [Internet]. 2025 Jul. 7 [cited 2025 Dec. 22];25(12):2586–2593.
Available from: https://doi.org/10.17305/bb.2025.12774
Journal information: Biomolecules and Biomedicine
Provided by: Association of Basic Medical Sciences of FBIH
Provided by Association of Basic Medical Sciences of FBIH
