Fully automated deep-learning section-based muscle segmentation from CT images for sarcopenia assessment.

Area of Science:

• Radiology
• Artificial Intelligence
• Gerontology

Background:

• Sarcopenia, a key indicator of frailty, is often assessed using muscle area measurements.
• Standard computed tomography (CT) scans of the abdomen present an opportunity for opportunistic sarcopenia screening.
• Current methods for muscle area quantification can be time-consuming and require manual input.

Purpose of the Study:

• To develop and validate a fully automated deep learning (DL) approach for measuring muscle area.
• To enable sarcopenia assessment from standard-of-care abdominal CT images without exclusion criteria.
• To facilitate opportunistic screening for frailty.

Main Methods:

• A retrospective study utilizing 1,070 training and 31 testing abdominal CT images.
• A two-stage fully convolutional neural network (FCNN) pipeline: section detection followed by muscle segmentation.
• Segmentation focused on L3 vertebral level muscles (erector spinae, psoas, rectus abdominus).

Main Results:

• The FCNN pipeline achieved high accuracy in muscle segmentation, with Dice overlaps of 0.96 ± 0.02 for combined muscle area.
• Excellent correlation for muscle attenuation (R² = 0.95) and area (R² = 0.98) on hold-out test cases.
• Fully automated processing completed in under 1 second per CT examination, with no statistical difference from manual segmentation.

Conclusions:

• A FCNN pipeline accurately and efficiently automates muscle segmentation at the L3 level from unselected abdominal CT volumes.
• This automated approach requires no manual processing, making it suitable for opportunistic frailty screening.
• The method shows promise as a generalizable tool for widespread clinical application.