Applying Artificial Intelligence to Quantify Body Composition on Abdominal CTs and Better Predict Kidney...

Area of Science:

• Radiology
• Nephrology
• Artificial Intelligence

Background:

• Abdominal CT scans for kidney transplant candidates offer untapped body composition data.
• Current prognostication models often exclude body composition due to challenges in routine measurement.
• Artificial intelligence (AI) can automate the extraction of this valuable data.

Purpose of the Study:

• To investigate if AI can accurately quantify body composition from abdominal CT scans.
• To determine if AI-derived body composition data can predict 5-year wait-list mortality in kidney transplant candidates.
• To compare the predictive performance of body composition data with the existing Expected Post-Transplant Survival Score (EPTS).

Main Methods:

• Retrospective observational study of 899 kidney transplant candidates (2007-2017).
• Deep learning models quantified body composition (fat, aortic calcification, bone density, muscle mass).
• Logistic regression compared body composition data and EPTS for predicting 5-year wait-list mortality.

Main Results:

• Myosteatosis and increased aortic/abdominal calcification were associated with higher 5-year wait-list mortality.
• AI-derived imaging parameters showed similar predictive performance to EPTS (AUC 0.70 vs. 0.67).
• Combining body composition data with EPTS slightly improved survival prediction (AUC 0.72).

Conclusions:

• Automated quantification of body composition from CT scans is feasible in kidney transplant candidates.
• Myosteatosis and atherosclerosis are significant predictors of 5-year wait-list mortality.
• AI-enhanced body composition analysis can augment existing models for improved prognostication.