Abstract
Study Design.
Multicenter retrospective observational study.
Objective.
This study aimed to distinguish tuberculous spondylitis (TS) from pyogenic spondylitis (PS) using magnetic resonance imaging (MRI). Further, a novel diagnostic model for differential diagnosis was developed.
Summary of Background Data.
TS and PS are the two most common spinal infections. Distinguishing between these types clinically is challenging. Delayed diagnosis can lead to deficits or kyphosis. Currently, there is a lack of radiology-based diagnostic models for TS and PS.
Methods.
We obtained radiologic images from MRI imaging of patients with TS and PS and applied the least absolute shrinkage and selection operator regression to select the optimal features for a predictive model. Predictive models were built using multiple logistic regression analysis. Clinical utility was determined using decision curve analysis, and internal validation was performed using bootstrap resampling.
Results.
A total of 201 patients with TS (n=105) or PS (n=96) were enrolled. We identified significant differences in MRI features between both groups. We found that noncontiguous multivertebral and single-vertebral body involvement were common in TS and PS, respectively. Vertebral bone lesions were more severe in the TS group than in the PS group (Z=−4.553, P<0.001). The patients in the TS group were also more prone to vertebral intraosseous, epidural, and paraspinal abscesses (P<0.001). A total of 8 predictors were included in the diagnostic model. Analysis of the calibration curve and area under the receiver operating characteristic curve suggested that the model was well-calibrated with high prediction accuracy.
Conclusions.
This is the largest study comparing MRI features in TS and PS and the first to develop an MRI-based nomogram, which may help clinicians distinguish between TS and PS.
Key words: tuberculous spondylitis, pyogenic spondylitis, spinal infection, vertebral bone lesions, intervertebral disk destruction, magnetic resonance imaging, nomogram, vertebral abscess, differential diagnosis, diagnosis
Spinal infections have an annual incidence of 2.4 to 7.2 per 100,000 adults, with a slow upward trend in recent years.1,2 Various microorganisms, including bacteria, mycobacteria, fungi, and parasites, can cause spinal infections.3 The most common types include tuberculous spondylitis (TS) and pyogenic spondylitis (PS). TS is a form of granulomatous spinal infection caused by Mycobacterium tuberculosis.4 PS is a form of nonspecific spinal infection with acute or subacute onset that is most commonly caused by Staphylococcus aureus.5
Distinguishing between TS and PS is challenging, given the similarities in their clinical manifestations, which include fever, spinal pain, and neurological dysfunction. Thus, specific microbiological and pathologic findings are important to confirm the infectious etiology.6 Emerging metagenomic next-generation and whole-genome sequencing technologies have been valuable in spinal infection diagnosis.7–9 Despite advancements in diagnostic methods, a delay in diagnosis of two to four months is not uncommon.1 Disease progression leading to neurological deficits, or kyphosis, can occur as a result.10,11
Radiologic characteristics on x-ray or computerized tomography can be used to inform spinal infection diagnosis.12,13 However, neither can identify the early inflammatory process or differentiate TS from PS in the later stages, thereby delaying disease management.14 Magnetic resonance imaging (MRI) is widely used for spondylodiscitis detection as it is noninvasive, has superior soft tissue contrast resolution with ~92% sensitivity and 96% specificity, and is capable of determining the spatial extent of infection and abscess formation using multidirectional imaging.15 Gadolinium further enhances the clarity of spinal MRI scans and improves diagnostic accuracy.16,17 This study aimed to identify MRI features unique to PS and TS, develop a diagnostic model, and evaluate the clinical significance of these features for differential diagnosis.
MATERIALS AND METHODS
This study was approved by the ethics committee of two study centers (approval no. [2023] LSD [027]H-KY and no. GWLCZXEC2023-72). The requirement for informed consent was waived due to the retrospective nature of the study. Demographic and clinical data were obtained from patients with clinically and/or pathologically proven TS or PS treated at either of the two hospitals between January 2018 and June 2022. Patients were selected using the following inclusion and exclusion criteria. Inclusion criteria were the following: (1) patients with a diagnosis of TS, based on histopathological examination, the presence of acid-fast bacilli, or growth of M. tuberculosis in culture; (2) patients with a diagnosis of PS, based on positive blood culture or histologic examination of percutaneous needle biopsy;18,19 and (3) patients who underwent conventional and contrast-enhanced MRI either of the two hospitals. Exclusion criteria were the following: (1) patients with pre-MRI surgical biopsy or antibiotic treatment; (2) presence of MR image artifacts due to internal fixation or other causes; and (3) patients with poor image quality.
All MRI sequences were acquired using two different GE 3.0T MR scanners (Signa Pioneer and Signa HDxt, GE Healthcare, Waukesha, WI). Pulse sequences included sagittal fast spin-echo T1-weighted imaging (T1w) and T2-weighted short-tau inversion recovery (T2w-STIR). After injecting 0.1 mmol/kg of gadoteric acid meglumine (Hengrui Pharmaceuticals Co., Ltd., Jiangsu, China) through the median cubital vein or dorsal metacarpal vein network, the sagittal and coronal planes were measured with contrast enhancement using fat-suppressed spin-echo T1-weighted imaging. The detailed standardized MR protocols are found in Supplemental Digital Content 1, http://links.lww.com/BRS/C235.
Two blinded musculoskeletal experts reviewed the MR images and collected data. In the multilevel disease setting, the parameters assessed refer to the “most severely affected” vertebral body or intervertebral disk. The evaluation criteria were jointly formulated by two experts and strictly implemented. Disagreements were resolved through consensus. The classifications and definitions of the investigated MRI features are presented in Table 1.
TABLE 1.
Classification and Definition on MR Images and T1w* Contrast-Enhanced Sequences
| Category | Definition |
|---|---|
| Classification and definition of vertebral body destruction on MR images | |
| General parameters | |
| Level of spinal involvement | Classified as cervical, thoracic, T12/L1 vertebrae†, lumbar, or sacral |
| No. involved vertebrae | Classified as 1, 2, or ≥3 vertebral bodies |
| Type of vertebral involvement | Classified as continuous or skip lesion(not-continuous) |
| Morphology of vertebral bone destruction | |
| Extent of vertebral destruction | Classified as 0—only destruction of vertebral endplate without reduction in height, <25% reduction, 25%–50% reduction, 50%–75% reduction, or >75% reduction compared with unaffected adjacent vertebrae body |
| Condition of the marginal endplate | Marginal endplate of the most severely affected vertebral body adjacent to the infected disk was assessed, classified as unchanged, with erosions, or complete destruction (endplate not visible) |
| Vertebral body signal | |
| Strength of vertebral body signal on T1w sequences | Classified as hypointense or heterogeneous |
| Strength of vertebral body signal on T2w-STIR‡ sequences | Classified as hyperintense or heterogeneous |
| Scope of vertebral body signal on T1w sequences | Defined as the percentage of abnormal vertebral signal to all vertebral body signal on T1w sequences, classified as <25%, 25%–50%, 50%–75%, or >75% |
| Scope of vertebral body signal on T2w-STIR sequences | Defined as the percentage of abnormal vertebral signal to all vertebral body signal on T2w-STIR sequences, classified as <25%, 25%–50%, 50%–75%, or >75% |
| T1w contrast-enhanced sequences of vertebral bodies | |
| Type of vertebral body signal enhancement | Classified as homogeneous, heterogeneous, or marginal |
| Scope of vertebral body signal enhancement | Defined as the percentage of abnormal vertebral signal to all vertebral body signal, classified as 25%, 50%, 75%, or 100% |
| Classification and definition of intervertebral disk destruction on MR images | |
| Morphology of intervertebral disk destruction | |
| Extent of intervertebral disk destruction | None/mild—the intervertebral disk height is unchanged and the signal is unchanged or changed; moderate—disk height decreased by <50%, height of the disk is partly maintained and signal with minute fluid-filled area; severe—height of the disk decreased by >50%, or complete destruction—disk abscess, disk structures indistinguishable |
| Intervertebral disk signal | |
| Strength of intervertebral disk signal on T1w sequences | Classified as isointense, hypointense, or hyperintense |
| Strength of intervertebral disk signal on T2w-STIR sequences | Classified as isointense, hypointense, hyperintense, or fluid |
| T1w contrast-enhanced sequences of intervertebral disk | |
| Type of intervertebral disk signal enhancement | Classified as diffuse, focal, marginal, or lack of enhancement |
| Abscess and effusion on T1w contrast-enhanced sequences | |
| Vertebral intraosseous abscess | Classified as yes or not |
| Disk abscess | Classified as yes or not |
| Epidural abscess | Classified as yes or not |
| Paravertebral abscess | Classified as yes or not |
| Paravertebral abscess border | Classified as clear or blur |
| Paravertebral abscess wall | Classified as thin and smooth, or thick and rough |
| Anterior longitudinal ligament enhancement | Classified as yes or not |
T1w: T1-weighted.
T12/L1 vertebrae: 12th thoracic vertebra or first lumbar vertebrae. T12 vertebrae is included in the thoracic count, L1 vertebrae is included in the lumbar count.
T2w-STIR: T2-weighted short-tau inversion recovery.
MR indicates magnetic resonance; STIR, short-tau inversion recovery.
Statistical Analyses
Statistical analyses were performed using SPSS (version 26.0; IBM Corp., Armonk, NY). Continuous variables were compared using an independent-sample t test. Categorical variables were compared using the χ2 test or Fisher exact test. Ranked variables were compared using the rank-sum test. P-values <0.05 were considered statistically significant.
To ensure predictive accuracy, variables not applicable to the whole cohort were omitted. Variables with significant effects were incorporated into RStudio software (version 2022.12.0+353; Posit, PBC, Boston, MA). The least absolute shrinkage and selection operator (LASSO) regression model was used to select the optimal subset of features by shrinking some coefficients to zero. The remaining non-zero variables were analyzed using multiple logistic regression to establish predictive models. Ten-fold cross-validation was applied to select the tuning parameter (λ) in LASSO regression. The calibration level was assessed using the calibration curve, and classification performance was evaluated using receiver operating characteristic curves and the area under the curve (AUC). To assess clinical utility, the net benefit at an acceptable risk threshold consistent with clinical practice was described using decision curve analysis (DCA), while bootstrap resampling (1000 replicates) was used for internal validation.
RESULTS
In total, 235 patients met the inclusion criteria. Among these, 34 patients were excluded. Some patients who were initially “potentially eligible” were ultimately not included in the study, including patients who were initially suspected but not diagnosed with TS or PS after a series of examinations (n=15), as well as patients unable to undergo contrast-enhanced MRI due to internal organ disease (n=4). The 201 patients included consisted of 105 (52.2%) TS and 96 (47.8%) PS cases. A flowchart of patient selection is presented in Figure 1. No significant differences in sex, age, or clinical characteristics were found between the two groups (see Supplemental Digital Content 2, http://links.lww.com/BRS/C236). The most common infectious agents in the PS group were S. aureus (n=44) and E. coli (n=31) (Supplemental Digital Content 3, http://links.lww.com/BRS/C237).
Figure 1.
Flowchart for selecting the study population. MRI indicates magnetic resonance imaging; PS, pyogenic spondylitis; TS, tuberculous spondylitis.
Among the 22 parameters, 16 showed significant differences between TS and PS (P<0.05). The TS group had a higher number of thoracic spines (69.5% vs. 31.3%, χ2=29.40, P<0.001) and T12/L1 vertebrae involvement (46.7% vs. 27.1%, χ2=8.22, P=0.004) than the PS group (Table 2). No significant differences were observed in cervical, lumbar, or sacral involvement. The TS group exhibited more affected vertebrae than the PS group (average 2.13 vs. 1.39), with one case involving six vertebral bodies (Figure 2A). Skip lesions (Figure 2) were observed only in the TS group (28.6% vs. 0, χ2=11.12, P=0.001). Overall, vertebral body destruction was more severe in the TS group than in the PS group (Z=−4.553, P<0.001) (Figures 2A and 3A), and the extent of endplate destruction was not associated with inflammation type (χ2=3.626, P=0.163). The vertebral bodies in both groups generally showed hypointense signals in the T1w sequence (Figures 2B and 3B). The TS group demonstrated higher heterogeneous signal intensities than the PS group (33.3% vs. 6.3%, χ2=22.655, P<0.001). An association between the vertebral body signal type and the infection type was observed in the T2w-STIR sequence, whereby a heterogeneous and a hyperintense signal was typical of TS (81.0%) and PS (87.5%), respectively (χ2=94.108, P<0.001) (Figure 2C and 3C). Furthermore, T1w-STIR (Z=−2.880, P=0.004) and T2w-STIR (Z=−2.985, P=0.003) sequences demonstrated that TS was associated with a wider range of vertebral signal change than PS. Vertebral body contrast enhancement was mostly heterogeneous (65.7%) (Figure 2D, E) and homogeneous (57.3%) (Figure 3D, E) in the TS and PS groups, respectively (χ2=50.919, P<0.001). No significant differences were found between the contrast enhancement range and the inflammation type (χ2=4.025, P=0.253).
TABLE 2.
Vertebral Body Damage and Signal Changes
| Tuberculosis group (n=105) frequencies (%) | Pyogenic group (n=96) frequencies (%) | P | |
|---|---|---|---|
| General parameters of vertebral body destruction | |||
| Level of spinal involvement | |||
| Cervical | 9 (8.6) | 12 (12.5) | 0.363 |
| Thoracic | 73 (69.5) | 30 (31.3) | <0.001 |
| T12/L1 vertebrae* | 49 (46.7) | 26 (27.1) | 0.004 |
| Lumbar | 60 (57.1) | 55 (57.3) | 0.983 |
| Sacral | 11 (10.5) | 7 (7.3) | 0.430 |
| No. involved vertebrae | |||
| 1 vertebral body | 28 (26.7) | 65 (67.7) | <0.001 |
| 2 vertebral bodies | 45 (42.9) | 25 (26.0) | — |
| ≥3 vertebral bodies | 32 (30.5) | 6 (6.3) | — |
| Type of vertebral involvement | (77 cases in total) | (31 cases in total) | — |
| Continuous | 55 (71.4) | 31 (100) | 0.001 |
| Skip lesion | 22 (28.6) | 0 | — |
| Morphology of vertebral bone destruction | |||
| Extent of vertebral destruction | |||
| 0 | 9 (8.6) | 21 (21.9) | <0.001 |
| <25% reduction | 15 (14.3) | 26 (27.1) | — |
| 25%–50% reduction | 23 (21.9) | 22 (22.9) | — |
| 50%–75% reduction | 28 (26.7) | 18 (18.8) | — |
| >75% reduction | 30 (28.6) | 9 (9.4) | — |
| Condition of the marginal endplate | |||
| Unchanged | 0 | 3 (3.1) | 0.163 |
| With erosions | 49 (46.7) | 47 (49.0) | — |
| Complete destruction | 56 (53.3) | 46 (47.9) | — |
| Signal on T1w† sequences and T2w-STIR‡ sequences of vertebral bodies | |||
| Strength of vertebral body signal | |||
| T1w sequences | |||
| Hypointense | 70 (66.7) | 90 (93.8) | <0.001 |
| Heterogeneous | 35 (33.3) | 6 (6.3) | — |
| T2w-STIR sequences | |||
| Hyperintense | 20 (19.0) | 84 (87.5) | <0.001 |
| Heterogeneous | 85 (81.0) | 12 (12.5) | — |
| Scope of vertebral body signal | |||
| T1w sequences | |||
| <25% | 2 (1.9) | 4 (4.2) | 0.004 |
| 25%–50% | 8 (7.6) | 10 (10.4) | — |
| 50%–75% | 11 (10.5) | 24 (25.0) | — |
| >75% | 84 (80.0) | 58 (60.4) | — |
| T2w-STIR sequences | |||
| <25% | 1 (1.0) | 2 (2.1) | 0.003 |
| 25%–50% | 5 (4.8) | 7 (7.3) | — |
| 50–75% | 7 (6.7) | 20 (20.8) | — |
| >75% | 92 (87.6) | 67 (69.8) | — |
| Signal on T1w contrast-enhanced sequences of vertebral bodies | |||
| Type of vertebral body signal enhancement | |||
| Homogeneous | 29 (27.6) | 55 (57.3) | <0.001 |
| Heterogeneous | 69 (65.7) | 16 (16.7) | — |
| Marginal | 7 (6.7) | 25 (26.0) | — |
| Scope of vertebral body signal enhancement | |||
| <25% | 2 (2.0) | 4 (4.1) | P=0.253 |
| 25%–50% | 10 (9.8) | 14 (14) | — |
| 50%–75% | 15 (14.7) | 19 (19.8) | — |
| >75% | 78 (76.6) | 59 (60.8) | — |
T12/L1 vertebrae: 12th thoracic vertebra or first lumbar vertebrae. T12 vertebrae is included in the thoracic count, L1 vertebrae is included in the lumbar count.
T1w: T1-weighted.
T2w-STIR: T2-weighted short-tau inversion recovery.
STIR indicates short-tau inversion recovery.
Figure 2.
Magnetic resonance imaging (MRI) examination obtained in a 71-year-old male patient with tuberculous spondylitis. Include schematic (A), MR images of sagittal T1-weighted (T1w) (B), T2-weighted short-tau inversion recovery (T2w-STIR) (C), contrast-enhanced sagittal T1w (D), and contrast-enhanced coronal T1w (E). Sagittal images revealed bone defects in vertebrae T3, T4, T10, and T11, signal changes in T8 and T9, and intervertebral disk destruction at T3–4 and T10–11. The anterior longitudinal ligament showed heterogeneous high signal. Coronal image displayed clear boundaries of swollen paravertebral soft tissue.
Figure 3.
Magnetic resonance imaging (MRI) examination obtained in a 68-year-old female patient with pyogenic spondylitis. Include schematic (A), MR images of T1w (B), T2w-STIR (C), contrast-enhanced sagittal T1w (D), and contrast-enhanced coronal T1w (E). Sagittal images revealed bone damage in the upper middle portion of the L2 vertebral body. Coronal image showing homogeneous enhancement in the paravertebral soft tissue and in the upper middle portion of the L2 vertical body.
The difference in disk destruction between the TS and PS groups was not significant (χ2=5.028, P=0.170). There were no significant differences in the intervertebral disk signals in the T1w and T2w-STIR sequences between the two groups (χ2=0.387, P=0.849; χ2=7.132, P=0.066) or between intervertebral disk abscess after contrast enhancement and the type of inflammation (χ2=5.512, P=0.138), respectively (Table 3).
TABLE 3.
Intervertebral Disk Damage and Signal Changes
| Tuberculosis group (n=105) frequencies (%) | Pyogenic group (n=96) frequencies (%) | P | |
|---|---|---|---|
| Morphology of intervertebral disk | |||
| Extent of intervertebral disk destruction | |||
| None/mild | 20 (19.0) | 8 (8.3) | 0.170 |
| Moderate | 12 (11.4) | 12 (12.5) | — |
| Severe | 28 (26.7) | 32 (33.3) | — |
| Complete destruction | 45 (42.9) | 44 (45.8) | — |
| Signal intensity on T1w* sequences and T2w-STIR† sequences of intervertebral disk | |||
| Strength of intervertebral disk signal | |||
| T1w sequences | |||
| Isointense | 9 (8.6) | 7 (7.3) | 0.849 |
| Hypointense | 92 (87.6) | 84 (87.5) | — |
| Hyperintense | 4 (3.8) | 5 (5.2) | — |
| T2w-STIR sequences | |||
| Isointense | 29 (27.6) | 13 (13.5) | 0.066 |
| Hypointense | 4 (3.8) | 2 (2.1) | — |
| Hyperintense | 41 (39) | 43 (44.8) | — |
| Fluid | 31 (29.5) | 38 (39.6) | — |
| T1w contrast-enhanced sequences of intervertebral disk | |||
| Type of intervertebral dis signal enhancement | |||
| Diffuse | 5 (4.8) | 10 (10.4) | 0.138 |
| Focal | 40 (38.1) | 33 (34.4) | — |
| Marginal | 44 (41.9) | 46 (47.9) | — |
| Lack of enhancement | 16 (15.2) | 7 (7.3) | — |
T1w: T1-weighted.
T2w-STIR: T2-weighted short-tau inversion recovery.
STIR indicates short-tau inversion recovery.
The following six parameters were more common in the TS group than in the PS group (P<0.001): vertebral intraosseous abscess (67.7% vs. 30.2%, χ2=28.075), epidural abscess (58.1% vs. 7.3%, χ2=57.821), paravertebral abscess (81.9% vs. 29.2%, χ2=56.817), paravertebral clear border (94.2% vs. 32.1%, χ2=8.921), paravertebral abscess wall thin and smooth (95.3% vs. 25%, χ2=61.058), and anterior longitudinal ligament enhancement (72.4% vs. 38.5%, χ2=23.331) (Table 4).
TABLE 4.
T1w Contrast-enhanced Sequences of Abscess and Effusion
| Tuberculosis group (n=105) frequencies (%) | Pyogenic group (n=96) frequencies (%) | P | |
|---|---|---|---|
| Vertebral intraosseous abscess | |||
| Yes | 71 (67.7) | 29 (30.2) | <0.001 |
| Not | 34 (32.4) | 67 (69.8) | — |
| Disk abscess | |||
| Yes | 29 (27.6) | 31 (32.3) | 0.470 |
| Not | 76 (72.4) | 65 (67.7) | — |
| Epidural abscess | |||
| Yes | 61 (58.1) | 7 (7.3) | <0.001 |
| Not | 44 (41.9) | 89 (92.7) | — |
| Paravertebral abscess | |||
| Yes | 86 (81.9) | 28 (29.2) | <0.001 |
| Not | 19 (18.1) | 68 (70.8) | — |
| Paravertebral abscess border | (86 cases in total) | (28 cases in total) | — |
| Clear | 81 (94.2) | 9 (32.1) | <0.001 |
| Blur | 5 (5.8) | 19 (67.9) | — |
| Paravertebral abscess wall | (86 cases in total) | (28 cases in total) | — |
| Thin and smooth | 82 (95.3) | 7 (25.0) | <0.001 |
| Thick and rough | 4 (4.7) | 21 (75.0) | — |
| Anterior longitudinal ligament enhancement | |||
| Yes | 76 (72.4) | 37 (38.5) | <0.001 |
| Not | 29 (27.6) | 59 (61.5) | — |
After omitting the type of vertebral involvement, paravertebral abscess border, and paravertebral abscess wall, which were not applicable to the whole cohort, 13 predictor variables were obtained: thoracic involvement, T12/L1 vertebral involvement, number of involved vertebrae, the extent of vertebral destruction, the strength of the vertebral body signal on T1w sequences, the strength of the vertebral body signal on T2w-STIR sequences, the scope of the vertebral body signal on T1w sequences, scope of the vertical body signal on T2w-STIR sequences, type of vertical body signal enhancement, vertebral intraosseous abscess, epidural abscess, paravertebral abscess, and anterior longitudinal ligament enhancement. Using LASSO, eight of these were selected as the most important for building the diagnostic model and were considered independent predictors of both diseases (Figure 4; Table 5).
Figure 4.
Radiomics feature selection using the least absolute shrinkage and selection operator (LASSO) regression model. A, Tuning parameter (λ) selection in the LASSO model used 10-fold cross-validation based on the minimum criterion. The optimal values of the LASSO tuning parameter (λ) are indicated by the dotted vertical lines, and a value λ of 0.0014 was selected. B, The coefficient profiles of 13 radiomic features against the log (λ) sequence. T1w, T1-weighted imaging; T2w-STIR, T2-weighted short-tau inversion recovery
TABLE 5.
Disease Risk Model Predictors
| Predictive factors | Feature class | LASSO coefficient (β) |
|---|---|---|
| No. involved vertebrae | 2 vertebral body | −1.5595 |
| ≥3 vertebral bodies | −12.7223 | |
| Extent of vertebral destruction | — | −4.3832 |
| Strength of vertebral body signal on T1w* sequences | Heterogeneous | 8.9160 |
| Strength of vertebral body signal on T2w-STIR† sequences | Heterogeneous | −9.2465 |
| Scope of vertebral body signal on T2w-STIR sequences | — | −6.1718 |
| Type of vertebral body signal enhancement | Heterogeneous | −6.5379 |
| Marginal | 5.0175 | |
| Epidural abscess | No | 13.0746 |
| Paravertebral abscess | No | 2.0502 |
T1w: T1-weighted.
T2w-STIR: T2-weighted short-tau inversion recovery.
LASSO indicates least absolute shrinkage and selection operator.
Personalized Predictive Model Development
A model was created using these eight variables for the nomogram (Figure 5A) and calibration curves were plotted. The AUC of the prediction model was 0.992; the prediction accuracy was high (Figure 5B). The apparent and bias-corrected curves of the model demonstrated a good fit, indicating a high discriminative power (Figure 5C). The DCA results indicated that the nomogram provided the greatest clinical net benefit in predicting PS if the line graph was within the threshold range of 0 to 0.9 (Figure 6).
Figure 5.
Nomogram construction and validation. A, Radiomics nomogram combining the individual scores (points) of each characteristic on the left side of the patient, and the sum of the scores of all variables was the total points, and the corresponding risk graph could predict patients with PS. B, The ROC curves of the clinical factors model. The AUC had a cutoff value of −0.270. At this cutoff, the sum of the sensitivity and specificity of the model is the largest at 0.952 and 0.948, respectively. C, Calibration curve of the nomogram model. The black solid line represents the fitting curve of the calibration model after internal sampling using the Bootstrap resampling method (B=1000). The average absolute error is 0.011 (n=201). The thick dashed oblique line represents the perfect prediction of an ideal model, while the thin dashed line represents the apparent prediction of the nomogram. AUC indicates area under the curve; ROC, receiver operating characteristic.
Figure 6.
The figure above is the decision curve analysis (DCA) curve. The blue line is the net benefit of patients using the nomogram model to predict pyogenic spondylitis (PS). The oblique line represents the benefit if all patients are considered to be infected with PS, and the horizontal line represents the benefit if all patients are considered to be free of PS (0). The DCA indicates that the benefits of the nomogram can predict PS at a threshold probability of 0 to 0.9.
DISCUSSION
Since infection can lead to fatality, early diagnosis, and antibiotic treatment are essential to prevent disease progression.20 The differential diagnosis of TS and PS depends on the comprehensive judgment of clinical features, radiology, and laboratory investigations.21 Comprehensive microbiological analysis plays a crucial role in the treatment of spinal infection22; however, only 30% to 57% of patients have positive results of percutaneous biopsy culture, which has a long and difficult culture period and high false-negative rate.23–25 Metagenomic next-generation sequencing and whole-genome sequencing have the advantages of high sensitivity, specificity, and short detection time.26 However, disadvantages include high infrastructure–associated costs and the requirement for skilled operators. MRI is the gold standard in imaging studies to detect spinal infection,15 and is valuable in distinguishing TS from PS.27,28
This study confirmed the effectiveness of MRI in clinically distinguishing 105 patients with TS from 96 patients with PS and established a predictive model to assist in the differential diagnosis of spinal infectious diseases. To our knowledge, this is the first study to use an MRI-based radiologic feature nomogram to distinguish between TS and PS and the largest series comparing MRI features to identify TS and PS.
Patients with TS had a higher prevalence of the following MRI features than patients with PS: thoracic spine or T12/L1 vertebrae involvement, multivertebral involvement, skip lesion (not-continuous), more severe vertebral destruction, vertebral body heterogeneous signal on T1w sequences, T2w-STIR sequences and a wider range of vertebral body signal, wider range of vertebral body signal on T1w contrast-enhanced sequences, vertebral body presence of abscess and paraspinal tissue abscess, paravertebral abscess with clear boundaries, thin and smooth walls, and contrast enhancement of the anterior longitudinal ligament. Patients with PS had the following common MRI features: single-vertebral body involvement, less severe vertebral body destruction, vertebral body hyperintensity on T2w sequences, homogeneous enhancement on T1w contrast-enhanced sequences, and paravertebral tissue abscesses with thick and coarse walls.
Generally, both TS and PS are more common in the thoracic or lumbar spine, and TS is less common in the cervical and sacral spines.4 The most common sites for PS infection are the lumbar, thoracic, cervical (3%–20%), and sacral spines, in that order,29 which is similar to our results. Moreover, skip lesions may be an important MRI feature for defining TS.30 In our study, noncontiguous multifocal spinal involvement was observed in 22 patients (21%) with TS, which was lower than that previously reported.31 We found that multivertebral body involvement was characteristic of TS, whereas single-vertebral body involvement was more common in PS.
Kanna et al 32 reported vertebral bone destruction and decreased vertebral height in ~50% of patients with TS. Similarly, bone destruction was more severe in the TS group in our study, with only 8.6% of patients not showing a decrease in vertebral height. In addition, lesioned vertebral bodies showed low-signal changes on T1w sequences and variable high-signal changes on T2w-STIR sequences. PS generally causes less damage to the vertebral bodies and walls than does TS (31% vs. 70%), which generally does not exceed 50% loss in vertebral body height. Notably, most pathologic changes are limited to the endplate.14,33,34 In our study, 71.9% of patients with PS and 44.8% of patients with TS demonstrated a decrease in vertebral body height of <50%. Vertebral body lesions showed a low signal on T1w sequences and a uniformly high signal on T2w-STIR sequences. Significant diffuse homogeneous enhancement changes were observed in the contrast-enhanced images.
According to our study, ~42.9% of patients with TS had complete intervertebral disk destruction, consistent with previous reports.28,35 Given the proteolytic enzyme-mediated protein degradation of the intervertebral disk structure by PS-associated bacteria,36 the rate of intervertebral disk degeneration in TS is often lower than that of PS, with early intervertebral disk injury and loss of intervertebral space considered as PS indicators.3 Unlike previous studies, no differences were found in the morphology, signal, or intervertebral disk tissue abscess, which could be effective in differentiating TS from PS, and may be attributed to the longer onset time of TS in patients admitted to our clinical center, resulting in more severe intervertebral disk damage.
Consistent with other reports, we found that patients with TS were more prone to intravertebral, epidural, and paravertebral abscesses than those with PS.27,37 Paravertebral abscesses in the TS group were relatively extensive, with thin and smooth walls and clear boundaries, whereas those in most of the PS group were poorly defined and had thick, irregular walls. Contrast enhancement of the anterior longitudinal ligament was more evident in the TS group, suggesting that TS had a stronger invasive force on the epidural soft tissue.
Nomograms provide a better visual and graphical representation of predictive models and can help surgeons quickly estimate outcomes based on readily available information.38 Computerized tomography–based nomograms have been proven to be a promising predictive tool for differentiating TS from PS.39 Among all 22 MRI parameters, we selected the eight most important parameters as independent predictors for both diseases. The MRI-based predictive model we constructed had a high AUC in receiver operating characteristic and a significant net gain in threshold probabilities for DCA. This indicates that the model has excellent discriminative ability.
This study had some limitations. First, because one of our clinical centers is a regional infectious disease referral center, our results may be biased toward patients with potentially more severe diseases. Second, despite the high AUC, our results are only relevant for this population, and external validation is necessary before applying them to clinical practice. Third, our study only involves the differential evaluation of TS and PS and does not represent other spinal infectious diseases, such as brucellar spondylitis or fungal spondylitis. Fourth, our retrospective study was limited to the study of one province; hence, there may be regional bias. Finally, imaging assessment only provides reference values for diagnosis, which must be definitely confirmed by comprehensive judgment.
In conclusion, this is the largest study we are aware of to compare MRI features in TS and PS and the first to develop an MRI-based nomogram. Our prognostic model may help clinicians distinguish TS from PS and select a suitable treatment strategy in the clinical setting.
Key Points
This is the largest study we are aware of to compare MRI features in TS and PS and the first to develop an MRI-based nomogram.
The MRI-based diagnostic model included eight predictive factors, demonstrating good predictive performance.
Our prognostic model may help clinicians distinguish TS from PS and select a suitable treatment strategy in the clinical setting.
Supplementary Material
Footnotes
Q.Z. and Y.K.H. contributed equally to this manuscript as co-corresponding authors.
Supported by the Shandong Provincial Natural Science Foundation [Grant number: ZR2022MH096]; Shandong Province Traditional Chinese Medicine Science and Technology Development Plan [grant number: No.2019-0524]; and Medical and Health Science and Technology Development Project of Shandong Province [grant number: No.2018WS244].
The device(s)/drug(s) is/are FDA-approved or approved by a corresponding national agency for this indication.
The authors report no conflicts of interest.
Supplemental Digital Content is available for this article. Direct URL citations are provided in the HTML and PDF versions of this article on the journal's website, www.spinejournal.com.
Contributor Information
Jin Wang, Email: wangjinsrt@163.com.
Zhaoxin Li, Email: 632247291@qq.com.
Xiansu Chi, Email: chixiansulucky@163.com.
Yungang Chen, Email: chen_yungang@163.com.
Huaxin Wang, Email: whx215@163.com.
Xiaoying Wang, Email: wxwxy123@126.com.
Kaiying Cui, Email: wellxiaoman@126.com.
Qing Wang, Email: tomwq1983@163.com.
Tongxin Lu, Email: 835694787@qq.com.
Jianhu Zheng, Email: 961075431@qq.com.
Qiang Zhang, Email: zq-9999@163.com.
Yanke Hao, Email: haoyanke001@126.com.
References
- 1. Babic M, Simpfendorfer CS. Infections of the spine. Infect Dis Clin North Am. 2017;31:279–97. [DOI] [PubMed] [Google Scholar]
- 2. Giampaolini N, Berdini M, Rotini M, et al. Non-specific spondylodiscitis: a new perspective for surgical treatment. Eur Spine J. 2022;31:461–72. [DOI] [PubMed] [Google Scholar]
- 3. Balériaux DL, Neugroschl C. Spinal and spinal cord infection. Eur Radiol. 2004;14:E72–83. [DOI] [PubMed] [Google Scholar]
- 4. Dunn RN, Husien MB. Spinal tuberculosis: review of current management. Bone Joint J. 2018;100:425–31. [DOI] [PubMed] [Google Scholar]
- 5. Sobottke R, Seifert H, Fätkenheuer G, et al. Current diagnosis and treatment of spondylodiscitis. Dtsch Arztebl Int. 2008;105:181–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Yoon YK, Jo YM, Kwon HH, et al. Differential diagnosis between tuberculous spondylodiscitis and pyogenic spontaneous spondylodiscitis: a multicenter descriptive and comparative study. Spine J. 2015;15:1764–71. [DOI] [PubMed] [Google Scholar]
- 7. Li Y, Yao XW, Tang L, et al. Diagnostic efficiency of metagenomic next-generation sequencing for suspected spinal tuberculosis in China: a multicenter prospective study. Front Microbiol. 2022;13:1018938. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Huang H, Shi J, Zheng M, et al. Pathogen detection in suspected spinal infection: metagenomic next-generation sequencing versus culture. Eur Spine J. 2023:1–9. doi 10.1007/s00586-023-07707-3. Online ahead of print. [DOI] [PubMed] [Google Scholar]
- 9. Ngcelwane M, Omar SV, Said M, et al. New horizons in the diagnosis of tuberculosis of the spine: the role of whole genome sequencing. Asian Spine J. 2023;17:511–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. García-Rodríguez JF, Álvarez-Díaz H, Lorenzo-García MV, et al. Extrapulmonary tuberculosis: epidemiology and risk factors. Enferm Infecc Microbiol Clin. 2011;29:502–9. [DOI] [PubMed] [Google Scholar]
- 11. Khanna K, Sabharwal S. Spinal tuberculosis: a comprehensive review for the modern spine surgeon. Spine J. 2019;19:1858–70. [DOI] [PubMed] [Google Scholar]
- 12. Gao M, Sun J, Jiang Z, et al. Comparison of tuberculous and brucellar spondylitis on magnetic resonance images. Spine. 2017;42:113–21. [DOI] [PubMed] [Google Scholar]
- 13. Hatzenbuehler J, Pulling TJ. Diagnosis and management of osteomyelitis. Am Fam Physician. 2011;84:1027–33. [PubMed] [Google Scholar]
- 14. Frel M, Białecki J, Wieczorek J, et al. magnetic resonance imaging in differentatial diagnosis of pyogenic spondylodiscitis and tuberculous spondylodiscitis. Pol J Radiol. 2017;82:71–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Herren C, Jung N, Pishnamaz M, et al. Spondylodiscitis: diagnosis and treatment options. Dtsch Arztebl Int. 2017;114:875–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Li S, Nguyen IP, Urbanczyk K. Common infectious diseases of the central nervous system-clinical features and imaging characteristics. Quant Imaging Med Surg. 2020;10:2227–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Skaf GS, Kanafani ZA, Araj GF, et al. Non-pyogenic infections of the spine. Int J Antimicrob Agents. 2010;36:99–105. [DOI] [PubMed] [Google Scholar]
- 18. Liu X, Li H, Jin C, et al. Differentiation between brucellar and tuberculous spondylodiscitis in the acute and subacute stages by MRI: a retrospective observational study. Acad Radiol. 2018;25:1183–9. [DOI] [PubMed] [Google Scholar]
- 19. Liu X, Zheng M, Sun J, et al. A diagnostic model for differentiating tuberculous spondylitis from pyogenic spondylitis on computed tomography images. Eur Radiol. 2021;31:7626–36. [DOI] [PubMed] [Google Scholar]
- 20. Guo H, Lan S, He Y, et al. Differentiating brucella spondylitis from tuberculous spondylitis by the conventional MRI and MR T2 mapping: a prospective study. Eur J Med Res. 2021;26:125. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Iwata E, Scarborough M, Bowden G, et al. The role of histology in the diagnosis of spondylodiscitis: correlation with clinical and microbiological findings. Bone Joint J. 2019;101:246–52. [DOI] [PubMed] [Google Scholar]
- 22. Stangenberg M, Mende KC, Mohme M, et al. Influence of microbiological diagnosis on the clinical course of spondylodiscitis. Infection. 2021;49:1017–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Gras G, Buzele R, Parienti JJ, et al. Microbiological diagnosis of vertebral osteomyelitis: relevance of second percutaneous biopsy following initial negative biopsy and limited yield of post-biopsy blood cultures. Eur J Clin Microbiol Infect Dis. 2014;33:371–5. [DOI] [PubMed] [Google Scholar]
- 24. Colmenero JD, Jiménez-Mejías ME, Reguera JM, et al. Tuberculous vertebral osteomyelitis in the new millennium: still a diagnostic and therapeutic challenge. Eur J Clin Microbiol Infect Dis. 2004;23:477–83. [DOI] [PubMed] [Google Scholar]
- 25. Pang Y, An J, Shu W, et al. Epidemiology of extrapulmonary tuberculosis among inpatients, China, 2008-2017. Emerg Infect Dis. 2019;25:457–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Wang G, Long J, Zhuang Y, et al. Application of metagenomic next-generation sequencing in the detection of pathogens in spinal infections. Spine J. 2023;23:859–67. [DOI] [PubMed] [Google Scholar]
- 27. Jung NY, Jee WH, Ha KY, et al. Discrimination of tuberculous spondylitis from pyogenic spondylitis on MRI. AJR Am J Roentgenol. 2004;182:1405–10. [DOI] [PubMed] [Google Scholar]
- 28. Chang MC, Wu HT, Lee CH, et al. Tuberculous spondylitis and pyogenic spondylitis: comparative magnetic resonance imaging features. Spine. 2006;31:782–8. [DOI] [PubMed] [Google Scholar]
- 29. Cheung WY, Luk KD. Pyogenic spondylitis. Int Orthop. 2012;36:397–404. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Thammaroj J, Kitkhuandee A, Sawanyawisuth K, et al. MR findings in spinal tuberculosis in an endemic country. J Med Imaging Radiat Oncol. 2014;58:267–76. [DOI] [PubMed] [Google Scholar]
- 31. Kaila R, Malhi AM, Mahmood B, et al. The incidence of multiple level noncontiguous vertebral tuberculosis detected using whole spine MRI. J Spinal Disord Tech. 2007;20:78–81. [DOI] [PubMed] [Google Scholar]
- 32. Kanna RM, Babu N, Kannan M, et al. Diagnostic accuracy of whole spine magnetic resonance imaging in spinal tuberculosis validated through tissue studies. Eur Spine J. 2019;28:3003–10. [DOI] [PubMed] [Google Scholar]
- 33. Zhang N, Zeng X, He L, et al. The value of MR imaging in comparative analysis of spinal infection in adults: pyogenic versus tuberculous. World Neurosurg. 2019;128:e806–13. [DOI] [PubMed] [Google Scholar]
- 34. Naselli N, Facchini G, Lima GM, et al. MRI in differential diagnosis between tuberculous and pyogenic spondylodiscitis. Eur Spine J. 2022;3:431–41. [DOI] [PubMed] [Google Scholar]
- 35. Lee Y, Kim BJ, Kim SH, et al. comparative analysis of spontaneous infectious spondylitis: pyogenic versus tuberculous. J Korean Neurosurg Soc. 2018;61:81–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Tali ET. Spinal infections. Eur J Radiol. 2004;50:120–33. [DOI] [PubMed] [Google Scholar]
- 37. Bozgeyik Z, Ozdemir H, Demirdag K, et al. Clinical and MRI findings of brucellar spondylodiscitis. Eur J Radiol. 2008;67:153–8. [DOI] [PubMed] [Google Scholar]
- 38. Devin CJ, Bydon M, Alvi MA, et al. A predictive model and nomogram for predicting return to work at 3 months after cervical spine surgery: an analysis from the Quality Outcomes Database. Neurosurg Focus. 2018;45:E9. [DOI] [PubMed] [Google Scholar]
- 39. Wu S, Wei Y, Li H, et al. A predictive clinical-radiomics nomogram for differentiating tuberculous spondylitis from pyogenic spondylitis using CT and clinical risk factors. Infect Drug Resist. 2022;15:7327–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.