Scholarlink

Purpose <p>To evaluate the predictive value of preoperative variables on postoperative outcomes, specifically by comparing the performance of clinical variables, disc radiomics, and paraspinal muscle radiomics for predicting poor disability improvement after single-level lumbar surgery.</p> Methods <p>In this retrospective single-centre cohort, we included 213 adults who underwent elective L4/5 decompression and fusion for degenerative lumbar disease between January 2020 and June 2024 and had preoperative lumbar MRI and 6–12-month Oswestry Disability Index (ODI) follow-up. Poor disability improvement was defined a priori as &lt; 20-point reduction in ODI. Sagittal and axial T2-weighted images at the index level were used for semi-automatic disc segmentation and fully automatic deep-learning-based segmentation of bilateral multifidus and erector spinae muscles. Radiomics features were extracted according to Image Biomarker Standardisation Initiative recommendations. Five L1-penalised logistic regression models with nested cross-validation were developed: (1) clinical variables (age, sex, BMI, symptom duration, baseline ODI and VAS); (2) disc radiomics; (3) muscle conventional metrics (volume, cross-sectional area, fat infiltration and Goutallier asymmetry); (4) muscle radiomics (conventional metrics plus radiomics features); and (5) combined feature-union model (clinical + disc-radiomics + muscle-radiomics). Discrimination, calibration and net benefit were assessed using area under the receiver-operating characteristic curve (AUC), Brier score and decision curve analysis.</p> Results <p>Poor disability improvement occurred in 76/213 patients (35.7%). The clinical model showed moderate discrimination (AUC 0.75), whereas the disc radiomics model performed close to chance (AUC 0.51). The muscle conventional model achieved an AUC of 0.70. The muscle radiomics model provided the highest discrimination (AUC 0.86) with good calibration and the lowest Brier score, and yielded the greatest net benefit across clinically relevant threshold probabilities. The combined model showed no significant incremental improvement and had a slightly lower AUC (AUC 0.81).</p> Conclusion <p>Paraspinal muscle MRI radiomics at the index level provide substantial incremental prognostic information for poor postoperative disability improvement beyond clinical variables and disc radiomics alone. Incorporating muscle-focused imaging biomarkers into preoperative risk stratification may refine patient counselling and rehabilitation planning for single-level lumbar surgery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Paraspinal muscle MRI radiomics outperform disc radiomics for predicting poor disability improvement after L4/5 single-level lumbar surgery

  • 期刊论文

  • 发表日期：

    2026年2月13日

查看全文
  • Machao Guo,
  • Xiangyu Li,
  • Yuxi Liu,
  • Shibao Lu

摘要

Purpose

To evaluate the predictive value of preoperative variables on postoperative outcomes, specifically by comparing the performance of clinical variables, disc radiomics, and paraspinal muscle radiomics for predicting poor disability improvement after single-level lumbar surgery.

Methods

In this retrospective single-centre cohort, we included 213 adults who underwent elective L4/5 decompression and fusion for degenerative lumbar disease between January 2020 and June 2024 and had preoperative lumbar MRI and 6–12-month Oswestry Disability Index (ODI) follow-up. Poor disability improvement was defined a priori as < 20-point reduction in ODI. Sagittal and axial T2-weighted images at the index level were used for semi-automatic disc segmentation and fully automatic deep-learning-based segmentation of bilateral multifidus and erector spinae muscles. Radiomics features were extracted according to Image Biomarker Standardisation Initiative recommendations. Five L1-penalised logistic regression models with nested cross-validation were developed: (1) clinical variables (age, sex, BMI, symptom duration, baseline ODI and VAS); (2) disc radiomics; (3) muscle conventional metrics (volume, cross-sectional area, fat infiltration and Goutallier asymmetry); (4) muscle radiomics (conventional metrics plus radiomics features); and (5) combined feature-union model (clinical + disc-radiomics + muscle-radiomics). Discrimination, calibration and net benefit were assessed using area under the receiver-operating characteristic curve (AUC), Brier score and decision curve analysis.

Results

Poor disability improvement occurred in 76/213 patients (35.7%). The clinical model showed moderate discrimination (AUC 0.75), whereas the disc radiomics model performed close to chance (AUC 0.51). The muscle conventional model achieved an AUC of 0.70. The muscle radiomics model provided the highest discrimination (AUC 0.86) with good calibration and the lowest Brier score, and yielded the greatest net benefit across clinically relevant threshold probabilities. The combined model showed no significant incremental improvement and had a slightly lower AUC (AUC 0.81).

Conclusion

Paraspinal muscle MRI radiomics at the index level provide substantial incremental prognostic information for poor postoperative disability improvement beyond clinical variables and disc radiomics alone. Incorporating muscle-focused imaging biomarkers into preoperative risk stratification may refine patient counselling and rehabilitation planning for single-level lumbar surgery.