Automated lumbar muscles segmentation with segment anything model for adult degenerative scoliosis

This study aims to develop an automatic lumbar muscles segmentation and processing tool based on the Segment Anything Model (SAM), and to explore the association between lumbar muscles degeneration and coronal parameters in patients with adult degenerative scoliosis (ADS).

Fifty-six ADS patients were included. Lumbar muscles (PM: psoas major, MF: multifidus, ES: erector spinae) cross-sectional area (CSA) and fatty infiltration (FI) at L3-L4 were measured via direct, ImageJ-based, and SAM-based (Spyder) methods. We assessed measurement consistency and time efficiency across methods. Correlations between muscle degeneration and ADS coronal parameters were analyzed using Spyder-derived data.

CSA measurement showed acceptable consistency (ICC 0.627–0.881, P < 0.001), FI poor (ICC 0.252–0.442, P < 0.001). Spyder was > 10-fold faster than direct measurement and ~ 2-fold faster than ImageJ (P < 0.001). ADS patients had larger concave-side CSA (PM, ES) and higher concave MF FI (P < 0.05). Lumbar curve angle correlated positively with concave PM FI (r = 0.273, P = 0.042) and negatively with convex MF CSA (r=-0.406, P = 0.002).

The SAM-based automated lumbar muscles tool “Spyder” enables accurate, efficient automated lumbar muscles segmentation in ADS, offering an efficient ADS research and diagnosis tool. Asymmetric lumbar muscles changes may correlate with coronal deformity.