AI-based fully automatic image analysis: Optimal abdominal and thoracic segmentation volumes for estimating total...

Area of Science:

• Radiology
• Medical Imaging Analysis
• Sarcopenia Research

Background:

• Sarcopenia evaluation using computed tomography (CT) often relies on single transverse slice skeletal muscle area measurements.
• Automatic segmentation of total muscle volume offers potentially lower variance and improved accuracy compared to single-slice methods.

Purpose of the Study:

• To identify optimal abdominal and thoracic anatomical volumes for predicting total skeletal muscle volume.
• To compare the predictive performance of segmented muscle volumes against traditional single-slice area measurements.

Main Methods:

• Utilized a cloud-based AI tool (recomia.org) for skeletal muscle segmentation across the entire torso in 994 patients.
• Employed linear regression models to compare various anatomical volumes and single-slice areas in predicting total torso muscle volume.

Main Results:

• Abdominal muscle volume from the coccyx to 25 cm cranially best predicted total muscle volume (R² 0.935), outperforming the L3 slice (R² 0.830).
• Thoracic muscle volume from the sternum's top to Th12's lower bound showed the highest correlation (R² 0.892), surpassing the Th12 slice (R² 0.775).
• Adjusting for body height slightly improved correlations but did not alter the ranking of predictive regions.

Conclusions:

• Reliably segmented muscle volumes via automated image analysis provide superior prediction of total muscle volume compared to single-slice areas.
• Identified specific abdominal and thoracic regions as optimal for sarcopenia assessment using automated CT analysis.