Share：
Share this content in WeChat
X
[Chinese] [PDF] 60 2
• Clinical Article •
Predicting lumbar disc degeneration using interpretable machine learning models based on MAGiC and IDEAL-IQ sequences
BU Qingsong  LIU Xiaofeng  WANG Wenjuan  HU Lei  ZHU Haoyu  WANG Zhiliang  WANG Tao  WANG Lingling  WANG Xiang 

DOI:10.12015/issn.1674-8034.2026.07.011.


[Abstract] Objective To investigate the correlation between multidimensional quantitative parameters derived from the magnetic resonance image compilation (MAGiC) sequence and the iterative decomposition of water and fat with echo asymmetry and least-squares estimation quantitative (IDEAL-IQ) sequence with lumbar disc degeneration (LDD), and to construct an interpretable machine learning model for predicting LDD.Materials and Methods A total of 199 volunteers with chronic low back pain were prospectively recruited and divided into a normal group (n = 83) and a degeneration group (n = 116) based on Pfirrmann grading (> grade Ⅱ defined as LDD) on T2-weighted images. The cohort was randomly split into training and validation sets at a 7∶3 ratio. All participants underwent conventional lumbar MRI, MAGiC, and IDEAL-IQ sequence scans. Collected data included the vertebral bone quality (VBQ) score calculated from conventional T1-weighted images, T1 relaxation time (T1), T2 relaxation time (T2), and proton density (PD) values obtained from the MAGiC sequence, the average fat fraction (FFav) measured by the IDEAL-IQ sequence, as well as general clinical information, laboratory results, and body mass index (BMI). Spearman correlation analysis and logistic regression were used to screen for degeneration-related predictors. Six machine learning algorithms [logistic regression (LR), decision tree (DT), extreme gradient boosting (XGBoost), support vector machine (SVM), k-nearest neighbors (KNN), and light gradient boosting machine (LightGBM)] were employed to build prediction models. Model performance was comprehensively evaluated using the receiver operating characteristic (ROC) curve, area under the curve (AUC), accuracy, sensitivity, specificity, F1-score, positive predictive value (PPV), and negative predictive value (NPV), DeLong test was used to compare the differences in AUC among the models. Clinical net benefit was assessed by decision curve analysis (DCA). The optimal model was interpreted using SHapley Additive exPlanations (SHAP).Results Significant differences were observed between the normal and degeneration groups in age, albumin (ALB), T1, T2, PD, FFav, and VBQ scores (P < 0.05). Disc degeneration showed positive correlations with age, FFav, and VBQ score (r = 0.61, 0.36, 0.39), and negative correlations with ALB, T1, T2, and PD values (r = -0.30, -0.60, -0.70, -0.47). Among the six machine learning models, LightGBM performed best, achieving an AUC of 0.967 (95% CI: 0.942 to 0.993) in the training set, with accuracy of 91.2%, sensitivity of 90.0%, specificity of 93.3%, F1-score of 0.922, PPV of 94.7%, and NPV of 87.5%. DCA indicated a high clinical net benefit within the medium-risk threshold range. SHAP analysis further revealed that T2, T1, and age were the key predictors of LDD.Conclusions Machine learning models built on multidimensional quantitative parameters from MAGiC and IDEAL-IQ sequences can effectively predict LDD, with the LightGBM model demonstrating the best performance. The SHAP method enhances model interpretability, providing a quantitative tool and decision-making reference for early identification and clinical intervention of LDD.
[Keywords] lumbar disc degeneration；machine learning；Pfirrmann grade；magnetic resonance imaging；quantitative magnetic resonance imaging

BU Qingsong   LIU Xiaofeng*   WANG Wenjuan   HU Lei   ZHU Haoyu   WANG Zhiliang   WANG Tao   WANG Lingling   WANG Xiang  

Department of Medical Imaging, Chizhou Hospital Affiliated to Wannan Medical University, Chizhou 247000, China

Corresponding author: LIU X F, E-mail: lxf6364@sina.com

Conflicts of interest   None.

Received  2026-02-04
Accepted  2026-07-10
DOI: 10.12015/issn.1674-8034.2026.07.011
DOI:10.12015/issn.1674-8034.2026.07.011.

[1]
MOHD ISA I L, TEOH S L, MOHD NOR N H, et al. Discogenic low back pain: anatomy, pathophysiology and treatments of intervertebral disc degeneration[J/OL]. Int J Mol Sci, 2022, 24(1): 208 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/36613651/. DOI: 10.3390/ijms24010208.
[2]
COSKUN H, TURAN A, KAPLANOGLU H, et al. Frequency of hypoplasia of the vertebral body at L5, and its relationship with degeneration in patients with low back pain[J]. Turk Neurosurg, 2022, 32(4): 641-648. DOI: 10.5137/1019-5149.JTN.34728-21.2.
[3]
WU L L, LIU L H, RAO S X, et al. Ultrashort time-to-echo T2* and T2* relaxometry for evaluation of lumbar disc degeneration: a comparative study[J/OL]. BMC Musculoskelet Disord, 2022, 23(1): 524 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/35650645/. DOI: 10.1186/s12891-022-05481-9.
[4]
MURTO N, LUOMA K, LUND T, et al. Reliability of T2-weighted signal intensity-based quantitative measurements and visual grading of lumbar disc degeneration on MRI[J]. Acta Radiol, 2023, 64(6): 2145-2151. DOI: 10.1177/02841851231169079.
[5]
SOLLMANN N, FIELDS A J, O'NEILL C, et al. Magnetic resonance imaging of the lumbar spine: recommendations for acquisition and image evaluation from the BACPAC spine imaging working group[J]. Pain Med, 2023, 24(Supplement_1): S81-S94. DOI: 10.1093/pm/pnac130.
[6]
RUSSO F, AMBROSIO L, GIANNARELLI E, et al. Innovative quantitative magnetic resonance tools to detect early intervertebral disc degeneration changes: a systematic review[J]. Spine J, 2023, 23(10): 1435-1450. DOI: 10.1016/j.spinee.2023.05.011.
[7]
YANG L, SUN C, GONG T, et al. T1ρ, T2 and T2* mapping of lumbar intervertebral disc degeneration: a comparison study[J/OL]. BMC Musculoskelet Disord, 2022, 23(1): 1135 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/36575488/. DOI: 10.1186/s12891-022-06040-y.
[8]
BOUHSINA N, DECANTE C, HARDEL J B, et al. Comparison of MRI T1, T2, and T2* mapping with histology for assessment of intervertebral disc degeneration in an ovine model[J/OL]. Sci Rep, 2022, 12(1): 5398 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/35354902/. DOI: 10.1038/s41598-022-09348-w.
[9]
KAPOOR R, RANGANKAR V P, KUMAR D, et al. Apparent diffusion coefficient and T2* mapping on 3T MRI in normal and degenerative lumbar intervertebral discs[J/OL]. Pol J Radiol, 2023, 88: e275-e285 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/37404546/. DOI: 10.5114/pjr.2023.128882.
[10]
STEFANOU D E, VELONAKIS G, KARAVASILIS E, et al. T2 mapping of lumbar intervertebral disc: quantitative evaluation of degeneration in relation to Pfirrmann grading system and a template for intervertebral disc segmentation[J]. Hippokratia, 2023, 27(3): 75-81.
[11]
HSU W H, KO A T, WENG C S, et al. Explainable machine learning model for predicting skeletal muscle loss during surgery and adjuvant chemotherapy in ovarian cancer[J]. J Cachexia Sarcopenia Muscle, 2023, 14(5): 2044-2053. DOI: 10.1002/jcsm.13282.
[12]
ABAS MOHAMED Y, KHOO B EE, SHAHRIMIE MOHD ASAARI M, et al. Decoding the black box: Explainable AI (XAI) for cancer diagnosis, prognosis, and treatment planning-A state-of-the art systematic review[J/OL]. Int J Med Inform, 2025, 193: 105689 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/39522406/. DOI: 10.1016/j.ijmedinf.2024.105689.
[13]
ONWUKA S, BRAVO-MERODIO L, GKOUTOS G V, et al. Explainable AI-prioritized plasma and fecal metabolites in inflammatory bowel disease and their dietary associations[J/OL]. iScience, 2024, 27(7): 110298 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/39040076/. DOI: 10.1016/j.isci.2024.110298.
[14]
LIU D, KADRI A, HERNANDO D, et al. MRI-based vertebral bone quality score: relationship with age and reproducibility[J]. Osteoporos Int, 2023, 34(12): 2077-2086. DOI: 10.1007/s00198-023-06893-6.
[15]
URRUTIA J, BESA P, CAMPOS M, et al. The Pfirrmann classification of lumbar intervertebral disc degeneration: an independent inter- and intra-observer agreement assessment[J]. Eur Spine J, 2016, 25(9): 2728-2733. DOI: 10.1007/s00586-016-4438-z.
[16]
NATALIA F, SUDIRMAN S, RUSLIM D, et al. Lumbar spine MRI annotation with intervertebral disc height and Pfirrmann grade predictions[J/OL]. PLoS One, 2024, 19(5): e0302067[2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/38728318/. DOI: 10.1371/journal.pone.0302067.
[17]
EHRESMAN J, PENNINGTON Z, SCHILLING A, et al. Novel MRI-based score for assessment of bone density in operative spine patients[J]. Spine J, 2020, 20(4): 556-562. DOI: 10.1016/j.spinee.2019.10.018.
[18]
CHEN X L, CUI P, LI Y J, et al. Links among MRI features in paraspinal muscles, inflammatory processes, and related back pain in patients with lumbar disc herniation[J/OL]. Jor Spine, 2024, 7: e1310[2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/38222815/. DOI: 10.1002/jsp2.1310.
[19]
JIANG Y W, YU L, LUO X J, et al. Quantitative synthetic MRI for evaluation of the lumbar intervertebral disk degeneration in patients with chronic low back pain[J/OL]. Eur J Radiol, 2020, 124: 108858 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/32035370/. DOI: 10.1016/j.ejrad.2020.108858.
[20]
ZHU L G, YU C S, ZHANG X F, et al. The treatment of intervertebral disc degeneration using Traditional Chinese Medicine[J/OL]. J Ethnopharmacol, 2020, 263: 113117 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/32738389/. DOI: 10.1016/j.jep.2020.113117.
[21]
NERLICH A G, SCHAAF R, WÄLCHLI B, et al. Temporo-spatial distribution of blood vessels in human lumbar intervertebral discs[J]. Eur Spine J, 2007, 16(4): 547-555. DOI: 10.1007/s00586-006-0213-x.
[22]
GRUNHAGEN T, WILDE G, SOUKANE D M, et al. Nutrient supplyand intervertebral disc metabolism[J]. J Bone Joint Surg Am, 2006, 88(Suppl 2): 30-35. DOI: 10.2106/JBJS.E.01290.
[23]
MARINELLI N L, HAUGHTON V M, MUÑOZ A, et al. T2 relaxation times of intervertebral disc tissue correlated with water content and proteoglycan content[J]. Spine, 2009, 34(5): 520-524. DOI: 10.1097/BRS.0b013e318195dd44.
[24]
FENG F T, WANG W, LIU S W, et al. T2 mapping and q-Dixon for assessment of intervertebral disc degeneration in lower back pain[J/OL]. BMC Musculoskelet Disord, 2025, 26(1): 234 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/40069693/. DOI: 10.1186/s12891-025-08453-x.
[25]
LI W H, ZHAO H, ZHOU S B, et al. Does vertebral osteoporosis delay or accelerate lumbar disc degeneration A systematic review[J]. Osteoporos Int, 2023, 34(12): 1983-2002. DOI: 10.1007/s00198-023-06880-x.
[26]
ZHONG R, WEI F X, WANG L, et al. The effects of intervertebral disc degeneration combined with osteoporosis on vascularization and microarchitecture of the endplate in Rhesus monkeys[J]. Eur Spine J, 2016, 25(9): 2705-2715. DOI: 10.1007/s00586-016-4593-2.
[27]
RACHNER T D, KHOSLA S, HOFBAUER L C. Osteoporosis: now and the future[J]. Lancet, 2011, 377(9773): 1276-1287. DOI: 10.1016/S0140-6736(10)62349-5.
[28]
JUNG M, ROSPLESZCZ S, LÖFFLER M T, et al. Association of lumbar vertebral bone marrow and paraspinal muscle fat composition with intervertebral disc degeneration: 3T quantitative MRI findings from the population-based KORA study[J]. Eur Radiol, 2023, 33(3): 1501-1512. DOI: 10.1007/s00330-022-09140-4.
[29]
LI X W, XIE Y X, LU R, et al. Q-Dixon and GRAPPATINI T2 mapping parameters: a whole spinal assessment of the relationship between osteoporosis and intervertebral disc degeneration[J]. J Magn Reson Imaging, 2022, 55(5): 1536-1546. DOI: 10.1002/jmri.27959.
[30]
CHEN X L, ZHANG A R, ZHAO K C, et al. The role of oxidative stress in intervertebral disc degeneration: Mechanisms and therapeutic implications[J/OL]. Ageing Res Rev, 2024, 98: 102323 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/38734147/. DOI: 10.1016/j.arr.2024.102323.
[31]
WANG Y D, CHENG H G, WANG T, et al. Oxidative stress in intervertebral disc degeneration: Molecular mechanisms, pathogenesis and treatment[J/OL]. Cell Prolif, 2023, 56(9): e13448[2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/36915968/. DOI: 10.1111/cpr.13448.
[32]
DEER T, SAYED D, MICHELS J, et al. A review of lumbar spinal stenosis with intermittent neurogenic claudication: disease and diagnosis[J]. Pain Med, 2019, 20(Suppl 2): S32-S44. DOI: 10.1093/pm/pnz161.
[33]
LIU G H, ZHANG H J, CHEN M C, et al. Causal relationship between intervertebral disc degeneration and osteoporosis: a bidirectional two-sample Mendelian randomization study[J/OL]. Front Endocrinol, 2024, 15: 1298531 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/38745961/. DOI: 10.3389/fendo.2024.1298531.
[34]
BHADOURIA N, HOLGUIN N. Osteoporosis treatments for intervertebral disc degeneration and back pain: a perspective[J/OL]. JBMR Plus, 2024, 8(6): ziae048 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/38706880/. DOI: 10.1093/jbmrpl/ziae048.
[35]
JI Y Y, HONG W F, LIU M Y, et al. Intervertebral disc degeneration associated with vertebral marrow fat, assessed using quantitative magnetic resonance imaging[J]. Skeletal Radiol, 2020, 49(11): 1753-1763. DOI: 10.1007/s00256-020-03419-7.
[36]
MONTAZEL J L, DIVINE M, LEPAGE E, et al. Normal spinal bone marrow in adults: dynamic gadolinium-enhanced MR imaging[J]. Radiology, 2003, 229(3): 703-709. DOI: 10.1148/radiol.2293020747.
[37]
WANG S J, DAI Y T, SHEN J H, et al. Research on expansion and classification of imbalanced data based on SMOTE algorithm[J/OL]. Sci Rep, 2021, 11: 24039 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/34912009/. DOI: 10.1038/s41598-021-03430-5.
[38]
BORRELLI P, KABOTEH R, ENQVIST O, et al. Artificial intelligence-aided CT segmentation for body composition analysis: a validation study[J/OL]. Eur Radiol Exp, 2021, 5(1): 11 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/33694046/. DOI: 10.1186/s41747-021-00210-8.
[39]
IGLESIAS L L, BELLÓN P S, DEL BARRIO A P, et al. A primer on deep learning and convolutional neural networks for clinicians[J/OL]. Insights Imaging, 2021, 12(1): 117 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/34383173/. DOI: 10.1186/s13244-021-01052-z.
[40]
SHEN H T, HUANG J W, ZHENG Q Q, et al. A deep-learning-based, fully automated program to segment and quantify major spinal components on axial lumbar spine magnetic resonance images[J/OL]. Phys Ther, 2021, 101(6): pzab041 [2026-02-03]. https://pubmed.ncbi.nlm.nih.gov/33517461/. DOI: 10.1093/ptj/pzab041.

PREV Analysis of the fat fraction histogram of the brown adipose depot in the supraclavicular region of patients with colorectal cancer based on magnetic resonance mDIXON technology
NEXT Value of multiparametric MR black-blood thrombus imaging in staging diagnosis of lower extremity deep vein thrombosis
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn