Comparing fully automated AI body composition measures derived from thin and thick slice CT image data

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Body composition analysis provides valuable insights into patient health and disease prognosis.
• Automated analysis of CT scans offers a scalable approach to body composition assessment.

Purpose of the Study:

• To compare fully automated artificial intelligence (AI)-derived body composition measures from thin (1.25 mm) and thick (5 mm) slice abdominal CT data.
• To evaluate the feasibility of using both thin and thick slice CT data for AI-based body composition algorithms.

Main Methods:

• A retrospective study analyzed 9882 abdominal CT scans (unenhanced and contrast-enhanced).
• Fully automated AI algorithms quantified bone attenuation, muscle attenuation, muscle area, liver attenuation, liver volume, spleen volume, visceral-to-subcutaneous fat ratio (VSR), and aortic calcium.
• Measures from thin and thick slices were compared using correlation coefficients and Bland-Altman analysis.

Main Results:

• Very strong positive correlations (r² > 0.92) were observed for all soft tissue measures and VSR.
• Moderate correlation was found for bone attenuation (r² = 0.35).
• Bland-Altman analysis demonstrated strong agreement for most soft tissue measures, with mean percentage differences < 5% for VSR, muscle area, liver attenuation, and liver volume.

Conclusions:

• Automated body composition measures derived from thin and thick abdominal CT slices are strongly correlated and show agreement.
• The findings suggest feasibility in using either thin or thick slice CT data for these AI-based body composition algorithms, particularly for soft tissue applications.