Inter-software and inter-threshold reliability of quantitative paraspinal muscle segmentation

Purpose Changes in the cross-sectional area (CSA) and functional cross-sectional area (FCSA) of the lumbar multifidus (MF) and erector spinae muscles (ES) are factors that can contribute to low back pain. For the assessment of muscle CSA and composition there are various software and threshold methods used for tissue segmentation in quantitative analysis. However, there is currently no gold standard for software as well as muscle segmentation. This study aims to analyze the measurement error between different image processing software and different threshold methods for muscle segmentation. Methods Magnetic resonance images (MRI) of 60 patients were evaluated. Muscle CSA and FCSA measurements were acquired from axial T 2-weighted MRI of the MF and ES at L4/L5 and L5/S1. CSA, FCSA, and FCSA/CSA ratio were measured independently by two observers. The MRI images were measured using two different software programs (ImageJ and Amira) and with two threshold methods (Circle/Overlap method) for each software to evaluate FCSA and FCSA/CSA ratio. Results Inter-software comparisons revealed high inter-rater reliability. However, poor inter-rater reliability were obtained with different threshold methods. CSA, FCSA, and FCSA/CSA showed excellent inter-software agreement of 0.75–0.99 regardless of the threshold segmentation method. The inter-rater reliability between the two observers ranged between 0.75 and 0.99. Comparison of the two segmentation methods revealed agreement between 0.19 and 0.84. FCSA and FCSA/CSA measured via the Overlap method were significantly higher than those measured via the Circle method ( P < 0.01). Conclusion The present study showed a high degree of reliability with very good agreement between the two software programs. However, study results based on different threshold methods should not be directly compared.