ISSN 1004-4140
CN 11-3017/P

The Feasibility of Texture-based Quantification for Evaluating Lumbar Intervertebral Disc Degeneration in Adolescent Idiopathic Scoliosis from Conventional T2-weighted Magnetic Resonance Imaging

WANG Fengxian, WANG Shoufeng, CHANG Ying, ZHOU Jin, CHEN Jing, ZHOU Zhengyang, WANG Dongmei

WANG F X, WANG S F, CHANG Y, et al. The Feasibility of Texture-based Quantification for Evaluating Lumbar Intervertebral Disc Degeneration in Adolescent Idiopathic Scoliosis from Conventional T2-weighted Magnetic Resonance Imaging[J]. CT Theory and Applications, 2023, 32(6): 735-745. DOI: 10.15953/j.ctta.2022.225.
Citation: WANG F X, WANG S F, CHANG Y, et al. The Feasibility of Texture-based Quantification for Evaluating Lumbar Intervertebral Disc Degeneration in Adolescent Idiopathic Scoliosis from Conventional T2-weighted Magnetic Resonance Imaging[J]. CT Theory and Applications, 2023, 32(6): 735-745. DOI: 10.15953/j.ctta.2022.225.
王凤仙, 王守丰, 常莹, 等. 基于MRI T2加权成像纹理分析评估青少年脊柱侧弯患者椎间盘退变的可行性研究[J]. CT理论与应用研究, 2023, 32(6): 735-745. DOI: 10.15953/j.ctta.2022.225.(英).
引用本文: 王凤仙, 王守丰, 常莹, 等. 基于MRI T2加权成像纹理分析评估青少年脊柱侧弯患者椎间盘退变的可行性研究[J]. CT理论与应用研究, 2023, 32(6): 735-745. DOI: 10.15953/j.ctta.2022.225.(英).

The Feasibility of Texture-based Quantification for Evaluating Lumbar Intervertebral Disc Degeneration in Adolescent Idiopathic Scoliosis from Conventional T2-weighted Magnetic Resonance Imaging

More Information
    Corresponding author:

    ZHOU Zhengyang: 男,南京大学医学院附属南京鼓楼医院主任医师、教授,主要研究方向为多排CT与MR的基础和临床应用研究和磁共振淋巴造影对肿瘤转移性淋巴结定性诊断的基础和临床研究,E-mail:zyzhou@nju.edu.cn

    WANG Dongmei: 女,上海中医药大学附属上海市中医院主任医师,主要骨肌影像学的诊断和成像新技术的开发与临床应用研究,E-mail:dongmeiwang9320@163.com

基于MRI T2加权成像纹理分析评估青少年脊柱侧弯患者椎间盘退变的可行性研究

详细信息
  • 中图分类号: R 445.2

  • Abstract:

    Objective: To investigate the utility of texture data based on T2-weighted magnetic resonance imaging (MRI) in determining intervertebral disc degeneration in adolescent idiopathic scoliosis (AIS). Materials and Methods: From October 2016 and March 2020, 122 patients with AIS and 40 volunteers who underwent 3.0T MRI were prospectively included. The following MRI texture data were generated: (1) mean, (2) standard deviation, (3) max, (4) min, (5) the fifth, 10th, 25th, 50th, 75th and 90th percentiles; (6) skewness; (7) kurtosis; and (8) entropy. The Pfirrmann system was used to evaluate the intervertebral discs of all participants. Patients with Pm I were divided into groups 1 and 2. Volunteers were classified into 0. Differences and correlations between the groups were analyzed. Results: The mean, standard deviation, max, entropy and the 5th, 10th, 25th, 50th, 75th, and 90th percentiles in group 2 were significantly lower than those in group 1 and group 0; the min in group 2 was significantly lower than in group 0; the skewness in group 2 was significantly higher than in group 1 and group 0; the kurtosis in group 2 was significantly lower than in group 1; the skewness in group 1 was significantly higher than in group 0 and the standard deviation, min, kurtosis and 5th, 10th, 25th, and 50th percentiles in group 1 were significantly lower than those in group 0. Conclusion: Texture analysis can be used to assess early degenerative changes in the intervertebral discs of patients with AIS.

    摘要:

    目的:探讨MRI T2加权成像纹理分析在青少年脊柱侧弯患者椎间盘退变中的应用价值。材料和方法:从2016年10月至2020年3月,前瞻性纳入122例AIS患者和40名志愿者行3.0 T磁共振成像(MRI)检查并得到患者图像MRI纹理参数值:①平均值,②标准差,③最大值,④最小值,⑤第5、10、25、50、75和90百分位数,⑥偏度,⑦峰度,⑧熵。采用Pfirrmann评分对所有参与者的椎间盘进行评估并分组,AIS患者中,评分为Pm I的患者纳入1组,其余患者纳入2组,志愿者纳入0组;分析组间差异性和相关性。结果:2组的均值、标准差、最大值、熵和第5、10、25、50、75、90百分位均显著低于1组和0组;2组min显著低于0组;2组偏度显著高于1组和0组;2组峰度显著低于1组;1组偏度显著高于0组,1组标准差、最小值、峰度和第5、10、25、50百分位显著低于0组。结论:纹理分析可用于评估AIS患者椎间盘早期退行性改变,且优于常规MRI T2加权成像。

  • Adolescent idiopathic scoliosis (AIS) is the most common form of scoliosis developed in adolescents and can lead to intervertebral disc degeneration[1-3]. Magnetic resonance imaging (MRI) is currently the most important method for the clinical evaluation of intervertebral disc pathology as degenerative changes in the intervertebral discs can manifest as signal changes on T2-weighted MRI[4-5]. Therefore, MRI is widely used to evaluate the degeneration of intervertebral discs.

    The currently recognized disc degeneration assessment system is the Pfirrmann grading system, which is based on MRI of the disc structure[4]. However, this method is limited for the detection of early intervertebral disc degeneration and changes in the microstructure of the degenerated intervertebral disc, and as a qualitative-type analysis is prone to subjectivity. In addition, with the development of emerging treatment technologies such as cell[6] and growth factor therapy[7], it is important to quantitatively evaluate intervertebral disc degeneration and its efficacy.

    Although a variety of functional imaging technologies have emerged to quantitatively evaluate changes in the microstructure of intervertebral discs (such as diffusion weighted (DW) imaging[8-9]), scan time has increased with the development of imaging technology. In addition, the parameters provided are limited and cannot be routinely performed and used in clinical settings, restricting clinical promotion capabilities and applications. Texture analysis refers to a variety of mathematical methods used to evaluate the grey-level intensity and position of pixels within an image to derive parameters automatically. It can not only provide quantitative parameters to reflect changes in the microstructure of the tissue, but also reflect the overall condition of the entire lesion and describe the local and regional relationships among the pixels in the region of interest (ROIs) to better reflect the heterogeneity of the organization[10]. Because of these advantages, texture parameters derived from MRI have been widely used as tools for viewing various diseases, especially in tumor imaging[11].

    Our purpose was to obtain intervertebral disc texture data based on conventional T2 MRI for texture analysis and evaluate intervertebral disc degeneration in patients with AIS.

    This study was approved by our Institutional Ethics Committee. Informed consent was obtained from all participants after explaining the risks involved and the purpose of the study.

    From October 2016 to March 2020, 122 consecutive patients with AIS were included in this trial. The inclusion criteria for patients were as follows: (1) adolescence (ages, 10- to 18-years-old); (2) preliminary diagnosis of idiopathic scoliosis based on a clinical examination and typical standing full-length radiographs[1,12]; (3) major thoracolumbar scoliotic curve; (4) no history of spinal surgery, spinal bracing treatment, or spinal trauma; and (5) ability to complete spine radiology examinations. Clinical and medication histories were collected from patients and their caretakers.

    Forty matched healthy volunteers were recruited as group 0. The inclusion criteria were as follows: (1) age of 10~18 years; (2) no history of spinal trauma, back pain, or spinal surgery; and (3) complete radiography and MRI examinations without spinal abnormalities.

    The Cobb angles were generated by determining and measuring the curve apex using the Cobb method on standard spinal radiographs[1,13]. The curved apexes of patients were located at T12 to L1; the average Cobb angle was (40.00±13.62) degrees (range, 15~77 degrees).

    All MR scans were performed using a 3.0T MR system (Ingenia; Philips Healthcare, Best, Netherlands) with a 16-channel phased-array sensitivity-encoding abdominal coil and a 16-channel torso phased-array body coil. The MRI sequence used a segmented whole-spine scan, a parallel imaging technique, and partial Fourier acquisition (sense=yes, P reduction (AP)=1.6, half-scan factor=0.624), and the lumbar spine was scanned in an oblique sagittal position. The standard MR scan remained the same throughout the study: oblique sagittal images of T1-weighted turbo spin echo (TSE) sequence (repetition time (TR)=450 ms, echo time (TE)=16 ms, field of view (FOV)=350 mm, matrix size=264×389, voxel size=0.7×0.9×3, slice thickness=3 mm, intersection gap=1 mm, echo trains=1, turbo factor=7); oblique sagittal images of T2-weighted TSE sequence (TR=2800 ms, TE=100 ms, FOV=350 mm, matrix size=204×340, voxel size=0.9×1.04×3, slice thickness=3 mm, intersection gap=1 mm, echo trains=1, turbo factor=29); coronal T2-weighted TSE sequence (TR=2800 ms, TE=100 ms, FOV=350 mm, matrix size=204×340, voxel size=0.9×1.04×3 mm, slice thickness=3 mm, intersection gap=1 mm, echo trains=1, turbo factor=29). The MR scan lasted for 8 min and 53 s.

    Image analysis was independently performed by two staff radiologists with different levels of experience (W.F.X and W.D.M with 4 years and 15 years of experience in spine imaging, respectively); neither radiologist had knowledge of the clinical data of the patients. According to the Pfirrmann system[4], which is based on the MRI-evaluated disc structure and includes signal intensity and the distinction between nucleus, disc height, and annulus, two observers evaluated the intervertebral discs (including L3/4, L4/5, and L5/S1) of 122 AIS patients and 40 volunteers. The interverbal discs of all volunteers were Pm I, classified as group 0 (healthy control group) (12 males and 28 females; mean age, (15.05±2.09) years; range, 11 ~ 18 years). For AIS patients, the Pfirrmann grade[4] is assigned based on the evaluation of the intervertebral discs (including L3/4, L4/5, and L5/S1) (previous studies found that lower intervertebral discs generally undergo early degeneration[14-15]). AIS patients with Pm I grade were divided into group 1 (AIS patients without obvious degeneration on MRI) (30 males and 64 females; mean age, (14.12±1.90) years; age range, 11~17 years), and Pm II to V grades were classified into group 2 (AIS patients with degeneration on MRI) (12 males and 16 females; average age, (15.17±2.19) years; age range, 11~18 years).

    Texture analysis was performed on all participants using an internal software (Image Analyzer 2.0, Nanjing, China), as described in our previous studies[16-17]. After downloading patient images and transferring them to our internal software, we were able to manually sketch the regions of interest (ROIs) on the patients’ images. After selecting all ROIs, the volume of interest (VOI) was determined. The software then automatically generated texture features. In all participants, at least two consecutive levels of the intervertebral disc nucleus pulposus (including L3/4, L4/5 and L5/S1) were selected to draw slice-by-slice the ROI and obtain the VOI. The average ROI area was 262.78 mm2 (range, 99.99 ~ 589.75 mm2) and the average VOI was 1166.42 mm3 (range, 439.97 ~ 2874.02 mm3).

    The value of each pixel on the VOI was automatically measured to generate the following histogram features: (1) mean (mean T2 relaxation time); (2) standard deviation (dispersion of a frequency distribution); (3) max (maximum T2 relaxation time); (4) min (minimum T2 relaxation time); (5) the 5th, 10th, 25th, 50th, 75th, and 90th percentiles (cumulative nth percentile of T2 relaxation time histogram); (6) skewness (degree of histogram asymmetry around the mean); (7) kurtosis (statistics on the sharpness of the histogram peak); and (8) entropy (the degree of disorder of T2 relaxation time over the VOI). All images were independently analyzed by two radiologists. In addition to the interobserver agreement analysis, the mean of the values determined by the two radiologists were calculated for statistical analyses.

    MedCalc Statistical Software v.15.2.2 (MedCalc Software, Ostend, Belgium; http://www.medcalc.org; 2015) was used for receiver operating characteristic (ROC) analysis, and SPSS software 18.0, was used for all other statistical analyses. The intraclass correlation coefficient (ICC) was used to estimate the intra- and interobserver agreements of the parameter measurements (poor=0.000 ~ 0.200, fair=0.201 ~ 0.400, moderate=0.301 ~ 0.600, good=0.601 ~ 0.800, and excellent=0.801 ~ 1.000). As most texture parameters did not have a normal distribution, a non-parametric test was used. The Kruskal-Wallis one-way ANOVA (K sample) test was used to compare the differences for all parameters among the three groups. The Spearman correlation coefficient©) was used to evaluate the trends of texture analysis parameters among the three groups (no correlation=0.0 ~ 0.2, weak=0.2 ~ 0.4, moderate=0.4 ~ 0.6, strong=0.6 ~ 0.8, and very strong=0.8 ~ 1.0). ROC analysis (including the area under the ROC curve [AUC]) was used to evaluate the effectiveness of histogram-derived parameters for distinguishing participants among the three groups and to select cut-off values by calculating the maximal Youden index (Youden index=sensitivity+specificity – 1). P<0.05 was considered statistically significant.

    The mean, standard deviation, max, entropy, and the 5th, 10th, 25th, 50th, 75th, and 90th percentiles in group 2 were significantly lower than those in group 1 and group 0; the min in group 2 was significantly lower than that in group 0; the skewness in group 2 was significantly higher than those in group 1 and group 0; the kurtosis in group 2 was significantly lower than that in group 1; the skewness in group 1 was significantly higher than those in group 0 and the standard deviation, min, kurtosis and 5th, 10th, 25th, and 50th percentiles in group 1 were significantly lower than those in group 0 ( Fig.1, Tables 1 ~3).

    Table  1.  Kruskal-Wallis one-way ANOVA (K sample) test between group 0 and group 1
    ParamterGroupP
    0 1 
        mean202.103±74.038180.604±47.4841.000
        standard-deviation62.857±22.63477.961±25.3000.012*
        min28.325±18.9939.819±6.796<0.001*
        max301.825±97.987326.010±97.2480.336
        5th percentile86.000±32.21448.521±18.381<0.001*
        10th percentile106.150±36.77868.202±23.571<0.001*
        25th percentile156.250±61.812116.617±35.0370.001*
        50th percentile217.700±84.736190.276±51.0650.450*
        75th percentile249.500±88.844246.914±67.0771.000
        90th percentile272.250±93.633275.617±77.2671.000
        skewness−0.538±0.246−0.324±0.3020.004*
        kurtosis2.470±0.4562.080±0.335<0.001*
        entropy5.264±0.2165.410±0.3540.136
    NOTE: Mean and percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time; the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree around the mean; kurtosis=measurement of the histogram sharpness; entropy=the distribution of T2 relaxation time levels over the ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiating the groups 0 ~ 1.
    下载: 导出CSV 
    | 显示表格
    Figure  1.  (a) Image of an 18-year-old female volunteer classified as group 0with a mean value of 150.19 ms; (b) Image of a 15-year-old female AIS patients classified as group 1with a mean value of 139.27 ms; (c) Images of a 17-year-old female AIS patients classified as group 2with a mean value of 93.72 ms; (d) Image of the histogram of the three participants in three different groups
    Table  2.  Kruskal-Wallis one-way ANOVA (K sample) test between group 0 and group 2
    ParamterGroupP
    0 2 
        mean202.103±74.03878.691±44.100<0.001*
        standard-deviation62.857±22.63434.065±16.268<0.001*
        min28.325±18.99311.142±17.504<0.001*
        max301.825±97.987163.821±71.492<0.001*
        5th percentile86.000±32.21429.178±28.686<0.001*
        10th percentile106.150±36.77836.178±31.232<0.001*
        25th percentile156.250±61.81251.928±38.613<0.001*
        50th percentile217.700±84.73675.357±44.333<0.001*
        75th percentile249.500±88.844103.892±53.918<0.001*
        90th percentile272.250±93.633126.500±62.421<0.001*
        skewness−0.538±0.2460.233±0.406<0.001*
        kurtosis2.470±0.4562.499±0.5521.000
        entropy5.264±0.2164.636±0.313<0.001*
    NOTE: mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time;the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree aroundthe mean; kurtosis=measurement ofthe histogram sharpness; entropy=the distribution of T2 relaxation time levels overthe ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiatingthe group 0 ~ 2.
    下载: 导出CSV 
    | 显示表格
    Table  3.  Kruskal-Wallis one-way ANOVA (K sample) test between group 1 and group 2
    ParamterGroupP
    1 2 
        mean180.604±47.48478.691±44.100<0.001*
        standard-deviation77.961±25.30034.065±16.268<0.001*
        min9.819±6.79611.142±17.5040.280
        max326.010±97.248163.821±71.492<0.001*
        5th percentile48.521±18.38129.1785±28.686<0.001*
        10th percentile68.202±23.57136.178±31.232<0.001*
        25th percentile116.617±35.03751.928±38.613<0.001*
        50th percentile190.276±51.06575.357±44.333<0.001*
        75th percentile246.914±67.077103.892±53.918<0.001*
        90th percentile275.617±77.267126.5±62.421<0.001*
        skewness−0.324±0.3020.233±0.406<0.001*
        kurtosis2.080±0.3352.499±0.552<0.001*
        entropy5.410±0.3544.636±0.313<0.001*
    NOTE: Mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time;the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree aroundthe mean; kurtosis=measurement ofthe histogram sharpness; entropy=the distribution of T2 relaxation time levels overthe ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiatingthe group 1 ~ 2.
    下载: 导出CSV 
    | 显示表格

    ROC analysis indicated that standard deviation, min, 5th percentile, 10th percentile, 25th percentile, skewness, kurtosis, and entropy differentiated between group 0 and group 1 (all P<0.05), with an AUC of 0.637 ~ 0.865. The 5th percentile had the highest diagnostic efficiency in differentiating between the two groups and performed better than the standard deviation, 25th percentile, skewness, kurtosis, and entropy. With a cut-off value of 61 ms, the sensitivity and specificity of the 5th percentile in differentiating group 0 and group 1 were 0.734 and 0.900, respectively (AUC=0.865).

    ROC analysis indicated that the mean, standard deviation, min, max, skewness, entropy, and 5th, 10th, 25th, 50th, 75th, and 90th percentiles could differentiate between group 0 and group 2, with an AUC of 0.809 ~ 0.967. The 50th percentile had the highest diagnostic efficiency in differentiating between these two groups and performed better than min. With a cut-off value of 142 ms, the sensitivity and specificity of the 50th percentile in differentiating group 0 and group 1 were 0.964 and 0.950, respectively (AUC=0.967).

    ROC analysis indicated that the mean, standard deviation, min, max, skewness, kurtosis, entropy, and 5th, 10th, 25th, 50th, 75th, and 90th percentiles could differentiate between group 1 and group 2, with an AUC of 0.634 ~ 0.960 (P<0.001 ~ 0.042). The 50th percentile had the highest diagnostic efficiency in differentiating between the two groups and performed better than the 5th and 10th percentiles for skewness and kurtosis. With a cut-off value of 127 ms, the sensitivity and specificity of the 50th percentile in differentiating group 0 and group 1 were 0.893 and 0.957, respectively (AUC = 0.960) (Table 4).

    Table  4.  Receiver operating characteristic curves of histogram parameters in distinguishing different groups
     ParameterCut-offSensitivity/%Specificity/%AccuracyAUCP
    Group 0 vs. Group 1 mean142.84028.70095.0000.4850.5480.380
     standard-deviation63.50064.90080.0000.6940.679<0.001*
     min23.00098.90057.5000.8650.796<0.001*
     max345.00044.70087.5000.5740.5990.064
     5th percentile61.00073.40090.0000.7830.865<0.001*
     10th percentile70.00060.60095.0000.7080.823<0.001*
     25th percentile100.00048.90095.0000.6260.727<0.001*
     50th percentile160.00038.30095.0000.5520.5940.072
     75th percentile199.00070.20050.0000.6410.5300.591
     90th percentile286.00041.50080.0000.5290.5400.475
     skewness−0.30052.100100.0000.6640.695<0.001*
     kurtosis1.95048.900100.0000.6410.767<0.001*
     entropy5.45053.20090.0000.6410.6370.005*
    Group 0 vs. Group 2 mean139.30092.90095.0000.9410.954<0.001*
     standard-deviation40.36078.600100.00000.9110.879<0.001*
     min9.00082.10080.0000.8080.809<0.001*
     max194.00082.10092.5000.8820.901<0.001*
     5th percentile52.00089.30095.0000.9260.933<0.001*
     10th percentile70.00092.90095.0000.9410.945<0.001*
     25th percentile92.00092.900100.0000.9700.955<0.001*
     50th percentile142.00096.40095.0000.9550.967<0.001*
     75th percentile134.00085.700100.0000.9410.934<0.001*
     90th percentile168.00085.700100.0000.9410.921<0.001*
     skewness−0.30089.300100.0000.9550.937<0.001*
     kurtosis2.48050.00070.0000.6170.5380.6111
     entropy5.02092.90090.0000.9110.962<0.001*
    Group 1 vs. Group 2 mean96.17085.700100.0000.9670.941<0.001*
     Standard-deviation41.78082.10095.7000.9260.936<0.001*
     min8.00078.60052.1000.5810.6340.042*
     max182.00078.600100.0000.9500.929<0.001*
     5th percentile30.00082.10086.2000.8520.839<0.001*
     10th percentile40.00082.10094.7000.9180.870<0.001*
     25th percentile69.00085.70096.8000.9420.929<0.001*
     50th percentile127.00089.30095.7000.9420.960<0.001*
     75th percentile134.00085.700100.0000.9670.952<0.001*
     90th percentile168.00085.70098.9000.9590.936<0.001*
     skewness0.01075.00088.3000.8520.863<0.001*
     kurtosis2.03085.70064.9000.6960.750<0.001*
     entropy4.89082.10090.4000.8850.937<0.001*
    NOTE: Mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time; 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=degree of histogram asymmetry around the mean; kurtosis=measurement of histogram sharpness; entropy=distribution of T2 relaxation time levels over the ROI. AUC: area under the receiver operating characteristic (ROC) curve. *−P<0.05.
    下载: 导出CSV 
    | 显示表格

    The Spearman correlation test indicated that the mean, standard deviation, min, max, the 5th, 10th, 25th, 50th, 75th, and 90th percentiles and the absolute value of skewness, kurtosis, and entropy correlated negatively from group 0 to group 2 (r=−0.472, P<0.001; r=−0.253, P=0.001; r=−0.448, P<0.001; r=−0.323, P<0.001; r=−0.665, P<0.001; r=−0.652, P<0.001; r=−0.610, P<0.001; r=−0.524; P<0.001; r=−0.405, P<0.001; r=−0.384, P<0.001; r=−0.228, P=0.003; r=−0.070, P=0.379; r=−0.325, P<0.001) (Fig.2).

    Figure  2.  Boxplots of the 5th, 10th, and 25th percentiles. The three parameters can distinguish the differences between the three groups and have a strong negative correlation with g (r=–0.665, –0.652, –0.610)

    Good intra- and interobserver agreements (ICC 0.803 ~ 0.916) were verified for all histogram-related parameters (Table 5).

    Table  5.  Intra- and Interobserver Agreement of All Parameters
    ParameterIntraobserver agreementInterobserver agreementP
    ICC95% CIICC95% CI
      mean0.8520.804, 0.8900.8190.761, 0.864<0.001
      standard deviation0.8690.826, 0.9020.8360.783, 0.877<0.001
      min0.8560.808, 0.8920.8060.745, 0.854<0.001
      max0.8810.841, 0.9110.8380.785, 0.879<0.001
      percentile 50.8460.796, 0.8850.8090.748, 0.856<0.001
      percentile 100.8660.822, 0.9000.8300.775, 0.872<0.001
      percentile 250.8690.826, 0.9020.8100.749, 0.857<0.001
      percentile 500.9000.865, 0.9250.8820.842, 0.912<0.001
      percentile 750.8740.832, 0.9060.8360.783, 0.877<0.001
      percentile 900.9000.865, 0.9250.8820.842, 0.912<0.001
      skewness0.9160.887, 0.9380.8990.865, 0.925<0.001
      kurtosis0.8870.849, 0.9160.8170.759, 0.863<0.001
      entropy0.8280.773, 0.8710.8030.741, 0.852<0.001
    NOTE: mean and all percentile values are in units of ms. Mean, mean T2 relaxation time; standard deviation, spread of distribution; min, minimum T2 relaxation time; max, maximum T2 relaxation time; the 5th, 10th, 25th, 50th, 75th, and 90th percentiles, nth percentile T2 relaxation time of a cumulative histogram; skewness, histogram asymmetry degree around the mean; kurtosis, measurement of the histogram sharpness; entropy the distribution of T2 relaxation time levels over the ROI. Abbreviations: ICC, intraclass correlation coefficient; CI, confidence interval. *-P<0.05.
    下载: 导出CSV 
    | 显示表格

    In this study, we performed T2 texture analysis on the intervertebral discs of adolescent idiopathic scoliosis patients and healthy volunteers to prove the effectiveness and superiority of T2 MRI histogram analysis-derived parameters in distinguishing between patients with scoliosis (with or without intervertebral disc degeneration) and healthy volunteers.

    The mean value is the most commonly used basic parameter and represents the average T2 relaxation time within a voxel. In our study, we found that the mean values in group 2 were significantly lower than those in groups 0 and 1. This is likely due to the dependency of intervertebral disc homeostasis on the interactions of the extracellular matrix, cells, and biomechanical stress[18]. In group 2 patients with intervertebral disc degeneration, this balance was disrupted, and the cells stopped producing proteoglycans, which led to a decrease in hydrostatic pressure and an increase in the shear forces on the cells[19-24]. The increased shear forces further decreased proteoglycan production, leading to further degeneration and dehydration of the intervertebral disc[18], consistent with the results of John et al[25] and Zhang et al[9]. The results of the Spearman correlation analysis indicated that the mean value was significantly negatively correlated from group 0 to group 2, indicating that the T2 relaxation time of the intervertebral discs from group 0 to group 2 showed a downward trend. We hypothesized that the intervertebral discs of group 1 patients had undergone slight histological degeneration, as it was difficult to differentiate with the naked eye the micro differences of the intervertebral discs on conventional images (T2-weighted image) between group 0 and group 1.

    Our results showed that when distinguishing the differences between the three groups, the smaller percentiles (i.e., 5th, 10th, and 25th percentiles) showed better performance and were strongly negatively correlated from group 0 to group 2. The smaller the percentile value, the higher the correlation from group 0 to group 2. Among them, the 5th percentile not only had the highest correlation from group 0 to group 2, but also exhibited the highest efficiency in distinguishing between groups 1 and 0 (AUC=0.865). The mean value between group 1 and group 0 was not significantly different, which could be due to the superiority of the histogram analysis itself. Through the analysis of each pixel in the histogram, it was found that the slight T2 relaxation time changed owing to the slight degeneration of each pixel and the value of the degraded pixel decreased, which induced more significant changes in the smaller percentile values and better reflected the difference between the two groups. The limitation of the mean value may be because it is the average value of the T2 relaxation time of all voxel points, which cannot reflect the change in each pixel and masks subtle but important differences between pixels.

    Our results showed that the minimum value differed significantly between the three groups, especially in the evaluation of group 0 and group 2 and between group 0 and group 1. The maximum value differed significantly in the evaluation of groups 0 and 2 and between groups 1 and 2; both were significantly negatively correlated from group 0 to group 2. The difference in the results between the maximum and minimum values between the different groups may be due to the following: in groups 0 and 1, some of the pixels of the patients in group 1 underwent slight degeneration, and the T2 relaxation time of the degenerated intervertebral disc was reduced; therefore, among the pixels of the intervertebral disc in group 1, the minimum T2 relaxation time value decreased. However, most of the pixels did not undergo a significant degeneration; therefore, the decrease in the min value changed significantly between the two groups. This is consistent with the smaller percentiles having better efficacy in distinguishing patients in groups 0 and 1. In contrast, the significant reduction in the maximum value in group 2 indicates that the intervertebral discs in group 2 underwent extensive degeneration, generally reducing the T2 relaxation time of the intervertebral disc pixels, while the undegraded pixels in group 1 maintained a higher maximum value, which makes the maximum value better in distinguishing between groups 2 and 1.

    Standard deviation is a parameter that reflects the spread of the distribution, entropy reflects the distribution pattern of pixels, and irregularity can be used to measure texture. The higher the entropy value, the more random the grey distribution and stronger the heterogeneity; skewness is a measure of the asymmetry of the distribution, and the greater the absolute value, the greater the asymmetry from the normal distribution[26]. We found that the standard deviation, entropy, and absolute values of skewness showed downward trends in groups 0 to 2, and the histograms shifted to the left and were more symmetrical and gradual in patients in groups 0 to 2. This also shows that as the intervertebral discs from groups to 0 to 2 progressively degenerate, the variability between pixels is reduced, symmetry is increased, and the state of the nucleus pulposus is more uniform, less heterogeneous, and anisotropic, which is in agreement with the results of Antoniou et al[27].

    Our study had some limitations. First, owing to ethical issues in human research, the pathological results of patients with AIS and the volunteers could not be obtained. Second, this was a cross-sectional study, and the role of histogram parameters in long-term follow-up and prognostic prediction remains unclear. Third, this study only examined changes in the nucleus pulposus and did not include the annulus fibrosus. These limitations require further investigation.

    In conclusion, texture analysis can be used to assess early degenerative changes in the intervertebral discs of patients with AIS that are invisible to the naked eye, especially in the smaller percentiles (i.e., 5th, 10th, and 25th percentiles), which can sensitively assess subtle changes in the intervertebral disc.

  • Figure  1.   (a) Image of an 18-year-old female volunteer classified as group 0with a mean value of 150.19 ms; (b) Image of a 15-year-old female AIS patients classified as group 1with a mean value of 139.27 ms; (c) Images of a 17-year-old female AIS patients classified as group 2with a mean value of 93.72 ms; (d) Image of the histogram of the three participants in three different groups

    Figure  2.   Boxplots of the 5th, 10th, and 25th percentiles. The three parameters can distinguish the differences between the three groups and have a strong negative correlation with g (r=–0.665, –0.652, –0.610)

    Table  1   Kruskal-Wallis one-way ANOVA (K sample) test between group 0 and group 1

    ParamterGroupP
    0 1 
        mean202.103±74.038180.604±47.4841.000
        standard-deviation62.857±22.63477.961±25.3000.012*
        min28.325±18.9939.819±6.796<0.001*
        max301.825±97.987326.010±97.2480.336
        5th percentile86.000±32.21448.521±18.381<0.001*
        10th percentile106.150±36.77868.202±23.571<0.001*
        25th percentile156.250±61.812116.617±35.0370.001*
        50th percentile217.700±84.736190.276±51.0650.450*
        75th percentile249.500±88.844246.914±67.0771.000
        90th percentile272.250±93.633275.617±77.2671.000
        skewness−0.538±0.246−0.324±0.3020.004*
        kurtosis2.470±0.4562.080±0.335<0.001*
        entropy5.264±0.2165.410±0.3540.136
    NOTE: Mean and percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time; the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree around the mean; kurtosis=measurement of the histogram sharpness; entropy=the distribution of T2 relaxation time levels over the ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiating the groups 0 ~ 1.
    下载: 导出CSV

    Table  2   Kruskal-Wallis one-way ANOVA (K sample) test between group 0 and group 2

    ParamterGroupP
    0 2 
        mean202.103±74.03878.691±44.100<0.001*
        standard-deviation62.857±22.63434.065±16.268<0.001*
        min28.325±18.99311.142±17.504<0.001*
        max301.825±97.987163.821±71.492<0.001*
        5th percentile86.000±32.21429.178±28.686<0.001*
        10th percentile106.150±36.77836.178±31.232<0.001*
        25th percentile156.250±61.81251.928±38.613<0.001*
        50th percentile217.700±84.73675.357±44.333<0.001*
        75th percentile249.500±88.844103.892±53.918<0.001*
        90th percentile272.250±93.633126.500±62.421<0.001*
        skewness−0.538±0.2460.233±0.406<0.001*
        kurtosis2.470±0.4562.499±0.5521.000
        entropy5.264±0.2164.636±0.313<0.001*
    NOTE: mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time;the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree aroundthe mean; kurtosis=measurement ofthe histogram sharpness; entropy=the distribution of T2 relaxation time levels overthe ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiatingthe group 0 ~ 2.
    下载: 导出CSV

    Table  3   Kruskal-Wallis one-way ANOVA (K sample) test between group 1 and group 2

    ParamterGroupP
    1 2 
        mean180.604±47.48478.691±44.100<0.001*
        standard-deviation77.961±25.30034.065±16.268<0.001*
        min9.819±6.79611.142±17.5040.280
        max326.010±97.248163.821±71.492<0.001*
        5th percentile48.521±18.38129.1785±28.686<0.001*
        10th percentile68.202±23.57136.178±31.232<0.001*
        25th percentile116.617±35.03751.928±38.613<0.001*
        50th percentile190.276±51.06575.357±44.333<0.001*
        75th percentile246.914±67.077103.892±53.918<0.001*
        90th percentile275.617±77.267126.5±62.421<0.001*
        skewness−0.324±0.3020.233±0.406<0.001*
        kurtosis2.080±0.3352.499±0.552<0.001*
        entropy5.410±0.3544.636±0.313<0.001*
    NOTE: Mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time;the 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=histogram asymmetry degree aroundthe mean; kurtosis=measurement ofthe histogram sharpness; entropy=the distribution of T2 relaxation time levels overthe ROI. *−Statistically significant. P<0.05 was considered statistically significant in differentiatingthe group 1 ~ 2.
    下载: 导出CSV

    Table  4   Receiver operating characteristic curves of histogram parameters in distinguishing different groups

     ParameterCut-offSensitivity/%Specificity/%AccuracyAUCP
    Group 0 vs. Group 1 mean142.84028.70095.0000.4850.5480.380
     standard-deviation63.50064.90080.0000.6940.679<0.001*
     min23.00098.90057.5000.8650.796<0.001*
     max345.00044.70087.5000.5740.5990.064
     5th percentile61.00073.40090.0000.7830.865<0.001*
     10th percentile70.00060.60095.0000.7080.823<0.001*
     25th percentile100.00048.90095.0000.6260.727<0.001*
     50th percentile160.00038.30095.0000.5520.5940.072
     75th percentile199.00070.20050.0000.6410.5300.591
     90th percentile286.00041.50080.0000.5290.5400.475
     skewness−0.30052.100100.0000.6640.695<0.001*
     kurtosis1.95048.900100.0000.6410.767<0.001*
     entropy5.45053.20090.0000.6410.6370.005*
    Group 0 vs. Group 2 mean139.30092.90095.0000.9410.954<0.001*
     standard-deviation40.36078.600100.00000.9110.879<0.001*
     min9.00082.10080.0000.8080.809<0.001*
     max194.00082.10092.5000.8820.901<0.001*
     5th percentile52.00089.30095.0000.9260.933<0.001*
     10th percentile70.00092.90095.0000.9410.945<0.001*
     25th percentile92.00092.900100.0000.9700.955<0.001*
     50th percentile142.00096.40095.0000.9550.967<0.001*
     75th percentile134.00085.700100.0000.9410.934<0.001*
     90th percentile168.00085.700100.0000.9410.921<0.001*
     skewness−0.30089.300100.0000.9550.937<0.001*
     kurtosis2.48050.00070.0000.6170.5380.6111
     entropy5.02092.90090.0000.9110.962<0.001*
    Group 1 vs. Group 2 mean96.17085.700100.0000.9670.941<0.001*
     Standard-deviation41.78082.10095.7000.9260.936<0.001*
     min8.00078.60052.1000.5810.6340.042*
     max182.00078.600100.0000.9500.929<0.001*
     5th percentile30.00082.10086.2000.8520.839<0.001*
     10th percentile40.00082.10094.7000.9180.870<0.001*
     25th percentile69.00085.70096.8000.9420.929<0.001*
     50th percentile127.00089.30095.7000.9420.960<0.001*
     75th percentile134.00085.700100.0000.9670.952<0.001*
     90th percentile168.00085.70098.9000.9590.936<0.001*
     skewness0.01075.00088.3000.8520.863<0.001*
     kurtosis2.03085.70064.9000.6960.750<0.001*
     entropy4.89082.10090.4000.8850.937<0.001*
    NOTE: Mean and all percentile values are in units of ms. Mean=mean T2 relaxation time; standard deviation=spread of distribution; min=minimum T2 relaxation time; max=maximum T2 relaxation time; 5th, 10th, 25th, 50th, 75th, and 90th percentiles=nth percentile T2 relaxation time of a cumulative histogram; skewness=degree of histogram asymmetry around the mean; kurtosis=measurement of histogram sharpness; entropy=distribution of T2 relaxation time levels over the ROI. AUC: area under the receiver operating characteristic (ROC) curve. *−P<0.05.
    下载: 导出CSV

    Table  5   Intra- and Interobserver Agreement of All Parameters

    ParameterIntraobserver agreementInterobserver agreementP
    ICC95% CIICC95% CI
      mean0.8520.804, 0.8900.8190.761, 0.864<0.001
      standard deviation0.8690.826, 0.9020.8360.783, 0.877<0.001
      min0.8560.808, 0.8920.8060.745, 0.854<0.001
      max0.8810.841, 0.9110.8380.785, 0.879<0.001
      percentile 50.8460.796, 0.8850.8090.748, 0.856<0.001
      percentile 100.8660.822, 0.9000.8300.775, 0.872<0.001
      percentile 250.8690.826, 0.9020.8100.749, 0.857<0.001
      percentile 500.9000.865, 0.9250.8820.842, 0.912<0.001
      percentile 750.8740.832, 0.9060.8360.783, 0.877<0.001
      percentile 900.9000.865, 0.9250.8820.842, 0.912<0.001
      skewness0.9160.887, 0.9380.8990.865, 0.925<0.001
      kurtosis0.8870.849, 0.9160.8170.759, 0.863<0.001
      entropy0.8280.773, 0.8710.8030.741, 0.852<0.001
    NOTE: mean and all percentile values are in units of ms. Mean, mean T2 relaxation time; standard deviation, spread of distribution; min, minimum T2 relaxation time; max, maximum T2 relaxation time; the 5th, 10th, 25th, 50th, 75th, and 90th percentiles, nth percentile T2 relaxation time of a cumulative histogram; skewness, histogram asymmetry degree around the mean; kurtosis, measurement of the histogram sharpness; entropy the distribution of T2 relaxation time levels over the ROI. Abbreviations: ICC, intraclass correlation coefficient; CI, confidence interval. *-P<0.05.
    下载: 导出CSV
  • [1]

    JANICKI J A, ALMAN B. Scoliosis: Review of diagnosis and treatment[J]. Paediatrics and Child Health, 2007, 12(9): 771−776. doi: 10.1093/pch/12.9.771

    [2]

    KEENAN B E, IZATT M T, ASKIN G N, et al. Sequential magnetic resonance imaging reveals individual level deformities of vertebrae and discs in the growing scoliotic spine[J]. Spine Deformity, 2017, 5(3): 197−207. doi: 10.1016/j.jspd.2016.10.002

    [3]

    NOHARA A, KAWAKAMI N, TSUJI T, et al. Intervertebral disc degeneration during postoperative follow-up more than 10 years after corrective surgery in idiopathic scoliosis: Comparison between patients with and without surgery[J]. Spine, 2018, 43(4): 255−261. doi: 10.1097/BRS.0000000000002319

    [4]

    PFIRRMANN C W, METZDORF A, ZANETTI M, et al. Magnetic resonance classification of lumbar intervertebral disc degeneration[J]. Spine, 2001, 26(17): 1873−1878. doi: 10.1097/00007632-200109010-00011

    [5]

    MODIC M T, MASARYK T J, ROSS J S, et al. Imaging of degenerative disk disease[J]. Radiology, 1988, 168(1): 177−186. doi: 10.1148/radiology.168.1.3289089

    [6]

    BLANQUER S B, GRIJPMA D W, POOT A A. Delivery systems for the treatment of degenerated intervertebral discs[J]. Advanced Drug Delivery Reviews, 2015, 84: 172−187. doi: 10.1016/j.addr.2014.10.024

    [7]

    OROZCO L, SOLER R, MORERA C, et al. Intervertebral disc repair by autologous mesenchymal bone marrow cells: A pilot study[J]. Transplantation, 2011, 92(7): 822−828. doi: 10.1097/TP.0b013e3182298a15

    [8]

    BEATTIE P F, MORGAN P S, PETERS D. Diffusion-weighted magnetic resonance imaging of normal and degenerative lumbar intervertebral discs: A new method to potentially quantify the physiologic effect of physical therapy intervention[J]. Journal of Orthopaedic & Sports Physical Therapy, 2008, 38(2): 42−49.

    [9]

    ZHANG W, MA X, WANG Y, et al. Assessment of apparent diffusion coefficient in lumbar intervertebral disc degeneration[J]. European Spine Journal, 2014, 23(9): 1830−1836. doi: 10.1007/s00586-014-3285-z

    [10]

    YTRE-HAUGE S, DYBVIK J A, LUNDERVOLD A, et al. Preoperative tumor texture analysis on MRI predicts high-risk disease and reduced survival in endometrial cancer[J]. Journal of Magnetic Resonance Imaging, 2018, 48(6): 1637−1647. doi: 10.1002/jmri.26184

    [11]

    SALA E, MEMA E, HIMOTO Y, et al. Unravelling tumour heterogeneity using next-generation imaging: Radiomics, radiogenomics, and habitat imaging[J]. Clinical Radiology, 2017, 72(1): 3−10. DOI: 10.1016/j.crad.2016.09.013.

    [12]

    PANCHMATIA J R, ISAAC A, MUTHUKUMAR T, et al. The 10 key steps for radiographic analysis of adolescent idiopathic scoliosis[J]. Clinical Radiology, 2015, 70(3): 235−242. doi: 10.1016/j.crad.2014.11.013

    [13]

    STUDER D. Clinical investigation and imaging[J]. Journal of Childrens Orthopaedics, 2013, 7(1): 29−35. doi: 10.1007/s11832-012-0461-8

    [14]

    WANG D, WANG S, GAO Y, et al. Diffusion tensor imaging of lumbar vertebras in female adolescent idiopathic scoliosis: Initial findings[J]. Journal of Computer Assisted Tomography, 2018, 42(2): 317−322. doi: 10.1097/RCT.0000000000000667

    [15]

    SIEMIONOW K, AN H, MASUDA K, et al. The effects of age, sex, ethnicity, and spinal level on the rate of intervertebral disc degeneration: A review of 1712 intervertebral discs[J]. Spine, 2011, 36(17): 1333−1339. doi: 10.1097/BRS.0b013e3181f2a177

    [16]

    LIU S, SHI H, JI C, et al. Preoperative CT texture analysis of gastric cancer: Correlations with postoperative TNM staging[J]. Clinical Radiology, 2018, 73(8): 756.e1−756.e9. doi: 10.1016/j.crad.2018.03.005

    [17]

    LIU S, LIU S, JI C, et al. Application of CT texture analysis in predicting histopathological characteristics of gastric cancers[J]. European Radiology, 2017, 27(12): 4951−4959. doi: 10.1007/s00330-017-4881-1

    [18]

    VERGROESEN P P, KINGMA I, EMANUEL K S, et al. Mechanics and biology in intervertebral disc degeneration: A vicious circle[J]. Osteoarthritis and Cartilage, 2015, 23(7): 1057−1070. doi: 10.1016/j.joca.2015.03.028

    [19]

    WEILER C, NERLICH A G, ZIPPERER J, et al. 2002 SSE Award Competition in Basic Science: Expression of major matrix metalloproteinases is associated with intervertebral disc degradation and resorption[J]. European Spine Journal, 2002, 11(4): 308−320. doi: 10.1007/s00586-002-0472-0

    [20]

    ANTONIOU J, STEFFEN T, NELSON F, et al. The human lumbar intervertebral disc: Evidence for changes in the biosynthesis and denaturation of the extracellular matrix with growth, maturation, ageing, and degeneration[J]. Journal of Clinical Investigation, 1996, 98(4): 996−1003. doi: 10.1172/JCI118884

    [21]

    ZHANG Z, CHAN Q, ANTHONY M P, et al. Age-related diffusion patterns in human lumbar intervertebral discs: A pilot study in asymptomatic subjects[J]. Magnetic Resonance Imaging, 2012, 30(2): 181−188. doi: 10.1016/j.mri.2011.09.021

    [22]

    CARTER D R, WONG M. Modelling cartilage mechanobiology[J]. Philosophical Transactions of the Royal Society B-Biological Sciences, 2003, 358(1437): 1461−1471. doi: 10.1098/rstb.2003.1346

    [23]

    SZTROLOVICS R, ALINI M, ROUGHLEY P J, et al. Aggrecan degradation in human intervertebral disc and articular cartilage[J]. Biochemical Journal, 1997, 326(Pt 1): 235−241.

    [24]

    SMITH R L, CARTER D R, SCHURMAN D J. Pressure and shear differentially alter human articular chondrocyte metabolism: A review[J]. Clinical Orthopaedics and Related Research, 2004, (427 Suppl): S89-95.

    [25]

    ANTONIOU J, DEMERS C N, BEAUDOIN G, et al. Apparent diffusion coefficient of intervertebral discs related to matrix composition and integrity[J]. Magnetic Resonance Imaging, 2004, 22(7): 963−972. DOI: 10.1016/j.mri.2004.02.011.

    [26]

    ALOBAIDLI S, MCQUAID S, SOUTH C, et al. The role of texture analysis in imaging as an outcome predictor and potential tool in radiotherapy treatment planning[J]. British Journal of Radiology, 2014, 87(1042): 20140369. doi: 10.1259/bjr.20140369

    [27]

    ANTONIOU J, EPURE L M, MICHALEK A J, et al. Analysis of quantitative magnetic resonance imaging and biomechanical parameters on human discs with different grades of degeneration[J]. Journal of Magnetic Resonance Imaging, 2013, 38(6): 1402−1414. doi: 10.1002/jmri.24120

图(2)  /  表(5)
计量
  • 文章访问数:  181
  • HTML全文浏览量:  41
  • PDF下载量:  16
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-11-13
  • 修回日期:  2023-04-27
  • 录用日期:  2023-05-04
  • 网络出版日期:  2023-06-04
  • 刊出日期:  2023-10-31

目录

/

返回文章
返回
x 关闭 永久关闭