This retrospective study developed an automated algorithm for 3D segmentation of adipose tissue and paravertebral muscle on chest CT using artificial intelligence (AI) and assessed its feasibility. The study included patients from the Boston Lung Cancer Study (2000-2011). For adipose tissue quantification, 77 patients were included, while 245 were used for muscle quantification. The data were split into training and test sets, with manual segmentation as the ground truth. Subcutaneous and visceral adipose tissues (SAT and VAT) were segmented separately. Muscle area, mean attenuation value, and intermuscular adipose tissue percentage (IMAT%) were calculated in the paravertebral muscle segmentation. The AI algorithm was trained on the training sets, and its performance was evaluated on the test sets. The AI achieved Dice scores above 0.87 and showed excellent correlations for VAT/SAT ratios, muscle attenuation value, and IMAT% (correlation coefficients > 0.98, p < 0.001). The mean differences between the AI and ground truth were minimal (VAT/SAT ratio: 0.7%; muscle attenuation value: 1 HU; IMAT%: <1%). In conclusion, we developed a feasible AI algorithm for automated 3D segmentation of adipose tissue and paravertebral muscle on chest CT.
Keywords: Adipose tissue; Artificial intelligence; Sarcopenia; Thorax; Tomography; X-ray computed.
© 2024. The Author(s).