Objectives To investigate deep-learning (DL) model accuracy in quantifying multifidus (MF) and erector spinae (ES) fat fraction (FF) compared to Dixon MRI, and to explore the indirect effect of muscle function between muscle degeneration and disability outcomes in chronic low back pain (CLBP). Materials and methods 96 CLBP and 86 healthy participants underwent 3 T MRI, muscle function assessment, Oswestry Disability Index (ODI), Roland-Morris Disability Questionnaire (RMDQ), and Short Form 36-Health Survey (SF-36). A DL-Otsu thresholding model quantified muscle FF and functional muscle volume from 3D T2_WI images, validated against Dixon-FF. Lin’s concordance correlation coefficient (CCC), Bland–Altman, and Passing-Bablok analyses assessed the concordance between Otsu-FF and Dixon-FF. Partial correlations and mediation analysis examined associations among muscle degeneration, muscle function, and disability outcomes. Results Otsu-FF showed agreement with Dixon-FF (MF: CCC = 0.96, 95% CI: 0.95, 0.97; ES: CCC = 0.95, 95% CI: 0.94, 0.96; bias: MF = 0.009; ES = 0.021). Partial correlations revealed MF and ES FF correlated with disability scores (ODI/RMDQ: r = 0.25 to 0.49; SF-36: r = −0.42, −0.28, p < 0.01). Muscle endurance negatively correlated with ODI (r = −0.57, 95% CI: −0.65, −0.45) and RMDQ (r = −0.49, 95% CI: −0.61, −0.35), positively with SF-36 (r = 0.51, 95% CI: 0.38, 0.63) (p < 0.01). Muscle endurance showed indirect effects on associations between muscle FF and disability outcomes (mediation proportion: 27.12% to 100%). Conclusion DL method accurately quantified muscle FF, closely matching Dixon results. Muscle FF correlated with disability outcomes in CLBP, with muscle endurance demonstrating a statistically indirect association within this relationship. Key Points <UnorderedList Mark="Bullet"> <ItemContent> <Emphasis Type="BoldItalic">Question</Emphasis> What are the associations between the deep learning-derived paraspinal muscle degeneration index, muscle function, and lumbar disability outcomes among patients with chronic low back pain? </ItemContent> <ItemContent> <Emphasis Type="BoldItalic">Findings</Emphasis> In chronic low back pain, deep learning-quantified higher fat fraction of paraspinal muscles correlated with worse lumbar disability outcomes, with muscle endurance demonstrating an indirect effect in this association. </ItemContent> <ItemContent> <Emphasis Type="BoldItalic">Clinical relevance</Emphasis> Incorporating the fat fraction of multifidus and erector spinae muscles and muscle endurance assessment is helpful for targeting rehabilitation training in chronic low back pain, improving disability outcomes. </ItemContent> </UnorderedList> Graphical Abstract

Deep learning-based assessment of paraspinal muscle degeneration and its relationships to muscle function and disability outcomes in chronic low back pain: a prospective study

期刊论文
发表日期：

2025年12月17日

查看全文

Pinzhen Chen,
Ping Ye,
Taotao Yang,
Jun Zhao,
Min He,
Yanmeng Peng,
Fei Luo,
Cunhua Chen,
Wei Chen,
Long Qian,
Jing Li,
Guotai Wang,
Wei Chen

摘要

Objectives

To investigate deep-learning (DL) model accuracy in quantifying multifidus (MF) and erector spinae (ES) fat fraction (FF) compared to Dixon MRI, and to explore the indirect effect of muscle function between muscle degeneration and disability outcomes in chronic low back pain (CLBP).

Materials and methods

96 CLBP and 86 healthy participants underwent 3 T MRI, muscle function assessment, Oswestry Disability Index (ODI), Roland-Morris Disability Questionnaire (RMDQ), and Short Form 36-Health Survey (SF-36). A DL-Otsu thresholding model quantified muscle FF and functional muscle volume from 3D T2_WI images, validated against Dixon-FF. Lin’s concordance correlation coefficient (CCC), Bland–Altman, and Passing-Bablok analyses assessed the concordance between Otsu-FF and Dixon-FF. Partial correlations and mediation analysis examined associations among muscle degeneration, muscle function, and disability outcomes.

Results

Otsu-FF showed agreement with Dixon-FF (MF: CCC = 0.96, 95% CI: 0.95, 0.97; ES: CCC = 0.95, 95% CI: 0.94, 0.96; bias: MF = 0.009; ES = 0.021). Partial correlations revealed MF and ES FF correlated with disability scores (ODI/RMDQ: r = 0.25 to 0.49; SF-36: r = −0.42, −0.28, p < 0.01). Muscle endurance negatively correlated with ODI (r = −0.57, 95% CI: −0.65, −0.45) and RMDQ (r = −0.49, 95% CI: −0.61, −0.35), positively with SF-36 (r = 0.51, 95% CI: 0.38, 0.63) (p < 0.01). Muscle endurance showed indirect effects on associations between muscle FF and disability outcomes (mediation proportion: 27.12% to 100%).

Conclusion

DL method accurately quantified muscle FF, closely matching Dixon results. Muscle FF correlated with disability outcomes in CLBP, with muscle endurance demonstrating a statistically indirect association within this relationship.

Key Points

Question What are the associations between the deep learning-derived paraspinal muscle degeneration index, muscle function, and lumbar disability outcomes among patients with chronic low back pain?

Findings In chronic low back pain, deep learning-quantified higher fat fraction of paraspinal muscles correlated with worse lumbar disability outcomes, with muscle endurance demonstrating an indirect effect in this association.

Clinical relevance Incorporating the fat fraction of multifidus and erector spinae muscles and muscle endurance assessment is helpful for targeting rehabilitation training in chronic low back pain, improving disability outcomes.

Graphical Abstract