A comprehensive analysis of magnetic resonance imaging and laboratory features in pyogenic, tuberculous, brucellar, and fungal spondylitis

Introduction

Infectious spondylitis is a serious condition characterized by infection of the vertebrae, intervertebral discs, and surrounding soft tissues (1). The most common forms of infectious spondylitis include pyogenic spondylitis (PS), tuberculous spondylitis (TS), brucellar spondylitis (BS), and fungal spondylitis (FS), and a definitive diagnosis usually depends on microbiologic or histopathologic confirmation by biopsy, which remains the gold standard (2). However, biopsy is an invasive procedure, and delays in obtaining histological confirmation are common in clinical practice (3). As a result, non-invasive diagnostic modalities, such as laboratory studies and magnetic resonance imaging (MRI), play a critical role in the early detection and assessment of spinal infections (3-5). MRI, in particular, is valuable for detecting soft tissue involvement, bone marrow edema, and abscess formation. Despite this, the overlapping imaging features among different types of infectious spondylitis make it challenging to differentiate between them solely based on MRI findings (1,6,7).

Previous studies have predominantly focused on comparing the MRI characteristics and laboratory findings of two or three types of infectious spondylitis, typically contrasting pyogenic and tuberculous forms (8,9). However, to date, no study has systematically examined and compared all four major infectious spondylitides in a single investigation. This gap in the literature is significant. Due to the differing treatment methods for these four diseases, it is crucial to accurately and efficiently distinguish these infections to guide appropriate treatment. For PS, pharmacological treatment generally requires the use of antibiotics that are sensitive to the pathogenic bacteria. TS treatment necessitates long-term combined use of anti-tuberculosis medications. Both BS and PS require the use of appropriate antibiotics, with BS commonly treated with a combination of doxycycline and rifampin or chloramphenicol. FS treatment, on the other hand, requires prolonged use of antifungal medications. Additionally, if PS, TS, BS, or FS result in severe spinal deformities or uncontrollable infections, surgical intervention is often necessary, such as abscess drainage, spinal decompression, or other related surgeries (10,11). The present study aims to fill this gap by offering the first comprehensive comparison of MRI features and laboratory data across all four types of infectious spondylitis. Additionally, to aid clinical decision-making, we propose a decision tree model that can be particularly useful in cases where biopsy is delayed or unfeasible, ultimately facilitating earlier and more targeted interventions to improve patient outcomes. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-605/rc).

Methods

Patients

We selected relevant cases using ICD codes and radiological search terms (including vertebral osteomyelitis, spinal abscess, septic facet, paravertebral abscess, etc.) through the hospital’s computer center. Ultimately, we included 213 cases of suspected infectious spondylitis diagnosed via imaging from January 2019 to December 2023 in the affiliated hospital (Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology), which were used as the training set. Additionally, we identified 86 cases of suspected infectious spondylitis diagnosed via imaging from January 2024 to June 2025 in the affiliated hospital, which were also incorporated into the training set. The inclusion criteria were as follows: (I) pre-surgical MRI imaging, including T1-weighted imaging (T1WI), T2-weighted imaging (T2WI) (sagittal and axial), and contrast-enhanced sequences; (II) confirmation of diagnosis via surgical biopsy of the infected area or computed tomography (CT)-guided puncture biopsy, with all biopsies yielding positive pathogen results. Exclusion criteria included: (I) incomplete laboratory test results or imaging, (II) poor-quality MRI images (e.g., those affected by artifacts), and (III) negative biopsy results. All MRI scans included in this study were acquired using a 3.0T MRI scanner. Imaging sequences consisted of T1WI, T2WI, and T2WI with fat suppression, and gadolinium-enhanced T1-weighted sequences, with or without fat suppression. Imaging was performed in both axial and sagittal planes. The MRIs were consistently analyzed by two musculoskeletal radiologists (G.W. and Z.X.) with 15 and 3 years of experience, respectively. Laboratory data for the included patients were collected and reconciled by two orthopedic surgeons (W.X. and W.Z.) with 20 and 3 years of experience, respectively. When there is a discrepancy between the judgments of two readers, the final decision will be made jointly by these two orthopedic doctors and a more senior doctor (with ≥20 years of experience). Only the consensus decision will be included in the final evaluation. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study has been approved by the Ethics Committee of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (No. TJ-IRB202407106) and individual consent for this retrospective analysis was waived.

Patient information

The time elapsed to the diagnosis of spondylodiscitis was defined as the interval between the onset of symptoms and the performance of MRI examination at our hospital. Neurological symptoms refer to manifestations such as numbness in the lower limbs, radicular pain, or muscle weakness. We defined fever as a condition present at the time of the patient’s visit to our hospital, while the presence of intermittent fever was determined based on whether it had occurred at any point since the onset of the disease.

Imaging evaluation

MRI images were consistently analyzed by two experienced radiologists. A comprehensive evaluation was conducted for each patient, assessing the following MRI features: the specific location of the affected vertebrae, the number of vertebrae involved per patient, the severity of vertebral destruction, the involvement of posterior elements (present or absent), the presence and degree of disc disruption, the presence of subligamentous inflammatory spread, the occurrence of skip lesions, and the presence of intraosseous abscesses, epidural abscesses, and endplate inflammatory reaction line. Additionally, the dimensions of paravertebral abscesses on T1-weighted contrast-enhanced sequences, and the signal intensity of the vertebral body on T2WI images. To minimize potential bias and objectively evaluate the signal intensity of infected vertebrae, we quantified the signal ratio between infected and normal vertebrae on T2WI.

Intervertebral disc destruction was defined as the extension of the infectious lesion through the endplates into the intervertebral disc, resulting in loss of disc height on sagittal images. Mild disc disruption was defined as a decrease in disc height of less than 50% compared to the adjacent normal disc height, while severe disruption was defined as a decrease of more than 50% in disc height (12). Mild vertebral destruction was defined as inflammation destroying the endplate but with vertebral body height loss of less than 25% compared to the adjacent vertebral body, while severe vertebral destruction was defined as a loss of more than 25% of the vertebral body height (13). Subligamentous spread of inflammation was defined as an inflammatory mass or abscess located beneath the anterior longitudinal ligament (7). Skip lesions were defined as spondylitis affecting different vertebral segments that were not contiguous with each other.

In the assessment of paravertebral soft tissues, lesions exhibiting enhanced margins during the contrast-enhancement phase were classified as abscesses (7). An abscess was defined as “large” if its diameter exceeded half the diameter of the vertebral body, whereas it was categorized as “small” if it fell below this threshold.

Statistical analysis

Quantitative data were first assessed for normality. Data conforming to a normal distribution are presented as x¯±s. For data that did not follow a normal distribution, the median and IQR were reported. In the analysis of quantitative data, analysis of variance (ANOVA) was conducted if the data met the assumptions of normality and homogeneity of variance. For pairwise comparisons, Tamhane’s method was employed when homogeneity of variance was not assumed, while the S-N-K test was used when homogeneity of variance was assumed. In cases where the data were not normally distributed, the Kruskal-Wallis (K-W) test was applied. For categorical data, the K-W test was used for multiple ordered variables, and the Chi-square test or Fisher’s exact test was utilized for unordered categorical variables. In the case of multiple sets of binary categorical data, the Chi-square test was applied, with Fisher’s exact test or the Chi-square test (corrected for α value) used for pairwise comparisons.

Model construction

In this study, stratified 5-fold cross-validation was utilized to ensure that the data distribution in the training and validation sets aligns with the original data. Additionally, an external test set was applied for further model verification. We constructed and assessed five models—decision tree, the support vector machine (SVM) model, random forest (RF), XGBoost, and AdaBoost—for their fit. The decision tree was built first, selecting the optimal feature using information gain and Gini index, and recursively splitting the dataset into subsets until stopping conditions were met. The SVM was next, finding the optimal hyperplane to separate classes in high-dimensional space using kernel functions to handle nonlinear data by mapping it to a higher-dimensional space for classification. The RF model is classified by building multiple decision trees, randomly selecting features, and using a voting mechanism. XGBoost performed classification via ensemble learning, optimizing the objective function and performing weighted voting. Lastly, AdaBoost used sequentially trained weak classifiers, adjusted sample weights, and conducted weighted voting for classification. All five models utilized a dataset comprising 20 variables that differed between groups, derived from 117 cases (including 10 binary variables, 2 ternary variables, and 8 numerical variables), with no missing data. To identify the optimal model with the best goodness-of-fit, we subsequently extracted the feature importance ranking from the top-performing model and generated its receiver operating characteristic (ROC) curve for comprehensive performance evaluation. Subsequently, quantitative data were standardized (Z-score normalization) using Python 3.9 (Python Software Foundation, https://www.python.org/), and the optimal model was deployed as a web-based diagnostic application.

Results

Patient characteristics and diagnoses

In this study, according to the established inclusion criteria, we included 117 patients diagnosed with infectious spondylitis from January 2019 to December 2023 at the affiliated hospitals as the test set. Additionally, we included 34 patients diagnosed with infectious spondylitis from January 2024 to June 2025 at the affiliated hospitals as the external test set (Figure 1). The clinical symptoms and laboratory values of the included patients are presented in Table 1.

Figure 1 Patient screening flowchart.

Among the clinical symptoms, there were no significant differences in gender, age, back pain, intermittent fever, and weight across the four groups. However, the time taken to diagnose spondylodiscitis and the presence of fever varied significantly between the groups. Neurological symptoms also differed among the groups. Specifically, in the PS group, spondylodiscitis was diagnosed more quickly in the FS group than in the TS group, with a notable difference. Additionally, the TS group experienced less fever than the BS group.

Although height showed a statistically significant difference (P<0.05), no meaningful variation was observed when comparing the two groups. For this reason, we do not consider height to differ significantly across the four groups, similar to the reasoning explained for the indicator “subligamentous spread”, which will be elaborated upon below.

In terms of laboratory results, white blood cell (WBC), percent monocytes (M%), and percent neutrophils (N%) did not differ significantly across the four groups. However, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), and albumin/globulin ratio (A/G ratio) showed significant differences. Percent lymphocytes (L%) values varied across the groups. Specifically, the PS group had significantly higher CRP, ESR, and A/G ratio levels compared to the TS and BS groups. Furthermore, the other groups had lower L% values compared to the BS group (Table 2).

Comparison between imaging characteristics

In terms of general radiologic findings, inflammation in the cervical spine did not differ significantly across the four groups. However, there were significant differences in inflammation involvement in the thoracic and lumbosacral spine, as well as in the number of vertebrae involved. Specifically, there was a significant between-group difference in thoracic spine involvement between the TS, BS, and FS groups, with the TS group showing a higher likelihood of thoracic spine involvement compared to the BS and FS groups. Regarding lumbosacral vertebrae, significant differences were observed between the BS and PS, TS groups, as well as between the TS and FS groups, with the BS group being more likely to have lumbosacral vertebrae involvement than the PS and TS groups, and the FS group more likely than the TS group (Table 3).

In fluid-sensitive sequences, involvement of posterior elements did not differ among the four groups. However, several other parameters were significantly different across the groups: vertebral body signal on T2WI (Infected vertebrae/normal vertebrae on T2WI), extent of vertebral destruction, presence of skip lesions, and endplate inflammatory reaction line (Figure 2). The degree of disc destruction also varied among the groups. The FS group exhibited significantly lower vertebral body signal intensity on T2WI images compared to the other groups (Figure 2). The incidence of severe vertebral body destruction was notably lower in the BS group compared to the TS and FS groups. The BS group was also more likely to have skip lesions than the PS group, although the incidence of skip lesions did not differ significantly between the TS and BS groups (Figure 3). Additionally, the FS group had a much higher incidence of endplate inflammatory reaction line compared to the other groups, while the PS group showed a higher incidence of this feature than the TS and BS groups (Table 4).

Figure 2 MRI features of FS. (A-C) Endplate inflammatory reaction lines on STIR sequences (white arrows). (D-F) Diffuse low signal intensity in affected vertebrae on T2WI (white arrows). FS, fungal spondylitis; MRI, magnetic resonance imaging; STIR, short tau inversion recovery; T2WI, T2-weighted imaging.

Figure 3 Skip lesions in TS and BS. (A) TS case demonstrating non-contiguous vertebral involvement (white arrows). (B,C) BS cases showing similar skip lesions (white arrows). BS, brucellar spondylitis; TS, tuberculous.

In T1-weighted contrast-enhanced sequences, the occurrence of paravertebral abscesses did not differ significantly across the four groups (Table 5). However, when paravertebral abscesses were present, there was a significant difference among the groups. Epidural abscesses did not show significant differences across the groups. Vertebral intraosseous abscesses, on the other hand, exhibited significant differences among the four groups. In cases where paravertebral abscesses were present, the likelihood of having a large paravertebral abscess was significantly higher in the TS group compared to the other groups (Figure 4). Additionally, vertebral intraosseous abscesses were more prevalent in the TS group than in the other groups (Figure 4).

Figure 4 Abscess characteristics in different spondylitis types. (A-C) Vertebral intraosseous abscesses in TS cases with enhancing walls (white arrows). (D) Large paravertebral abscess in TS. (E,F) Small paravertebral abscesses in FS (E) and PS (F) cases (white arrows). FS, fungal spondylitis; PS, pyogenic spondylitis; TS, tuberculous spondylitis.

Although the P value for subligamentous spread was less than 0.05 when comparing the four groups, we applied the Bonferroni correction to control for the risk of false positives due to multiple comparisons. This correction significantly reduced the significance level for each pairwise comparison. As a result, while there was an overall difference, the relatively small differences between the groups meant that individual pairwise comparisons did not reach the strictly corrected threshold of significance.

Model performance

In this study, we assessed the performance of five machine learning models using both a training set and an external test set, as detailed in Table 6. The metrics evaluated include balance accuracy, recall, precision, and the F1 score. Notably, the RF model achieved the highest balance accuracy of 0.75, with recall and precision values of 0.75 and 0.82, respectively, resulting in an F1 score of 0.73. In contrast, the SVM model exhibited the poorest performance, with a balanced accuracy of 0.48 and an F1 score of 0.42. As illustrated in Figure 5, the RF-derived feature importance analysis identified “signal ratio between infected vertebrae and normal vertebrae in T2WI”, “endplate inflammatory reaction line”, and “vertebral intraosseous abscess” as the top three discriminative features. These results provide a comparative analysis of the models’ predictive capabilities in the context of classifying infectious spondylitis (Figure 6).

Figure 5 Top 7 important features generated by the RF model. CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; m, months; RF, random forest; T2WI, T2-weighted imaging.

Figure 6 Five ML model PRCs. AUPRC, area under precision-recall curve; ML, machine learning; PRC, precision-recall curve; RF, random forest; SVM, support vector machine.

Figure 7 displays the ROC curves for the RF model on the training and test sets. The left panel shows the training set results with a micro-average area under the curve (AUC) of 0.88 and a macro-average AUC of 0.94, indicating strong model performance. The right panel presents the test set results with a micro-average AUC of 0.90 and a macro-average AUC of 0.92 (Figure 7 and Table 7), also reflecting high predictive accuracy. The micro-average AUC considers all classes equally, providing an overall performance metric, while the macro-average AUC calculates the AUC for each class and averages them, offering a class-agnostic measure. Both AUCs demonstrate the RF model’s robustness in distinguishing between classes on both datasets. Despite the high AUC values indicating strong discriminative ability, the relatively lower sensitivity suggests that the model may not be optimizing for balanced classification. This could be due to the model’s tendency to prioritize overall accuracy over precise identification, which is influenced by the chosen decision boundary. Additionally, the RF model, being an ensemble method, excels at capturing complex patterns but might not directly reflect in single-threshold metrics like sensitivity and specificity as effectively as AUC.

Figure 7 ROC curves for RF model’s training and test sets. RF, random forest; ROC, receiver operating characteristic.

To enhance accessibility for patients and clinicians across different centers, we transformed the final model into a user-friendly online prediction application. By visiting the website and entering the predictive indicators, the application can determine the likelihood of having a specific type of infectious spondylitis. The application is available at https://vertebalpredictorapp.streamlit.app/.

Discussion

In this study, we conducted a comprehensive analysis of MRI and laboratory features across pyogenic, tuberculous, brucellar, and fungal spondylitis. Utilizing five machine learning models and evaluating their performance on both training and external test sets, we identified distinct clinical and imaging characteristics for each infection type. The RF model emerged as the most accurate, with AUCs of 0.94 in the training set and 0.92 in the test set, demonstrating its effectiveness in distinguishing between these infections. Additionally, we developed an online prediction application to enhance accessibility and facilitate clinical decision-making.

The common pathogen of PS is Staphylococcus aureus (2), which produces a variety of protein hydrolases, including α-hemolysin and phenol-soluble regulatory proteins. These proteases are crucial for pathogenicity, facilitating the rapid spread of infection and contributing to tissue damage (14). The time to diagnosis for patients with PS is typically shorter than for those with TS, often due to the distinct characteristics of the causative pathogens (6,13). Patients in the TS group generally experience longer diagnostic delays, primarily due to the insidious onset of the disease and the nonspecific nature of the symptoms. In contrast, patients with FS often have a history of immunocompromise, exhibit a rapid disease onset, and are prone to forming small paraspinal abscesses with vertebral destruction, which are typically detected earlier on MRI (7,15). Our finding that patients in the BS group exhibited more pronounced febrile symptoms aligns with the observations of Liang et al., who reported that fever is often an early manifestation of BS (16). This may be attributed to Brucella’s ability to evade the host’s immune defenses by infecting and surviving within macrophages. This endocytosis results in a chronic or subacute inflammatory response, which produces persistent febrile symptoms.

Among the laboratory findings, ESR, CRP, and A/G, which are indicators of acute infection, showed significant differences between the PS group and the TS and BS groups. Since ESR and CRP reflect the acute inflammatory response, they were found to be higher in the PS group, whereas the inflammatory response in the TS and BS groups was generally slower. As a result, the elevation of ESR and CRP in these groups was less pronounced, which aligns with the findings of several studies (17-19). In the case of TS, the local inflammatory response induced by Mycobacterium tuberculosis inhibits macrophage function (20-22), delaying the immune response through modulation of immune pathways (23-27). This combination of factors contributes to the lower ESR and CRP levels observed in patients with tuberculous spondylitis. In the chronic infection process of BS, the pathogen employs immune evasion mechanisms. Lipopolysaccharide (LPS) weakens the activation of toll-like receptor 4 (TLR4), reducing the initiation of pro-inflammatory signaling. Its type IV secretion system (T4SS) evades immune system recognition by directly transferring effector proteins into host cells. Together, these mechanisms synergistically suppress the immune response, leading to a less significant elevation of inflammatory markers (ESR/CRP), which is consistent with the characteristics of chronic infection (28). Furthermore, as mentioned earlier, patients in the FS group often have impaired immune function, which may result in a weaker acute immune response compared to healthy or immunocompetent individuals (29). Consequently, FS patients tend to exhibit lower ESR and CRP levels and a less pronounced systemic inflammatory response.

Similarly, in PS, due to its acute onset, the synthesis of immunoglobulins and other globulins (e.g., complement, acute-phase proteins) is increased, leading to elevated globulin levels (30). The liver prioritizes the synthesis of acute-phase proteins over albumin, resulting in decreased or insufficient albumin production, which significantly lowers the A/G ratio (31-33). In contrast, infections with Mycobacterium tuberculosis and Brucella typically activate cellular immunity (34,35), and the chronic nature of the disease allows for sustained albumin production, which generally does not decrease significantly (36). However, humoral immunity also plays a critical role in combating Mycobacterium tuberculosis and Brucella, with globulin contributing to this defense, as demonstrated in various studies (37-40). This is why the A/G ratio was higher in the BS group compared to the PS group, as observed in the TS group. Regarding lymphocyte ratios, the higher proportion of lymphocytes observed in the BS group can be attributed to the immune escape mechanisms and immunosuppressive effects of Brucella. By evading the host’s immune response, Brucella prevents effective recognition and clearance of the bacteria, leading to a relative increase in circulating lymphocytes (41,42). Similarly, Mycobacterium tuberculosis delays the lymphocyte response through immune evasion and the establishment of an immune tolerance state, which prevents efficient immune activation in the early stages of infection (43,44). These factors may explain the lower percentage of lymphocytes in the TS group compared to the BS group.

In the study by Gupta et al., thoracic spine involvement, posterior element involvement, and spinal deformity were considered independent predictors for the diagnosis of TS (12). However, Batirel et al. analyzed data from a large sample of TS patients and found that lumbar spine involvement was more prevalent than thoracic spine involvement (45), which challenges the notion that thoracic spine involvement is an independent diagnostic factor for TS. Nonetheless, other studies support the findings of Gupta et al. (6,19,46,47). Multiple vertebral body involvement in the TS group has been frequently reported in prior studies and has become a characteristic feature of TS (6,15,48,49). However, the present study unexpectedly found that the percentage of multivertebral involvement in the BS group was significantly higher than that in the PS group. Additionally, the incidence of jumping lesions in the BS group did not differ from that in the TS group, which is an unusual finding not previously reported. In fact, this prevalence of multivertebral involvement in the BS group exceeds the 5–21% range reported in prior studies on Brucella spondylitis (50,51). Previous pathological studies have shown that BS often spreads through the vertebral venous plexus, a pathway also discussed in the context of jumping lesions in TS (52-54). This may provide a potential mechanism for the occurrence of jumping lesions in BS.

Among the parameters observed in the fluid-sensitive sequence, the slight destruction of the vertebral body in the BS group may be attributed to Brucella’s ability to activate and impede the maturation of dendritic cells, as suggested in previous studies (41,42,55). The finding of lesser disc destruction in the TS group compared to the PS group has been reported and explained in several studies, which attribute this to the absence of protein hydrolases in Mycobacterium tuberculosis (47,49). However, our results showed that posterior element involvement was not uniquely present in TS, and this feature did not differ significantly between the four groups. This is a surprising finding, as it contradicts established reports and studies that have noted differences between PS and TS with respect to posterior element involvement (17-19). Our findings suggest that posterior element involvement is not unique to TS and reflects the comprehensive nature of the present study. Additionally, this study for the first time reported the phenomenon that FS patients are prone to low signal intensity on T2WI. This could be related to the paramagnetism of the fungus, which leads to localized iron deposition and results in diffuse low signal on T2WI (55,56). In contrast to Li et al.’s study, we did observe an endplate inflammatory reaction line in the PS group, but the incidence was not as high as the 46% reported in their study (1). Interestingly, a similar phenomenon was observed in the FS group in our study, with a much higher incidence of 71%, compared to the PS group. The occurrence of this high signal may be related to the distribution of arterial blood supply to the vertebral body, although the exact cause remains unclear (55).

The two main features of TS are large paravertebral abscesses and vertebral intraosseous abscesses, which are typically multisegmental. These features differ from other types of infectious spondylitis, which tend to form smaller paravertebral abscesses, with a lower likelihood of developing large paravertebral abscesses (7,8). In this study, the presence or absence of epidural abscesses, subligamentous spread, and paravertebral abscesses were not effective indicators to differentiate the four types of infectious spondylitis. While epidural abscesses and subligamentous spread were considered more common in PS, and subligamentous spread was more frequent in TS, a comparative study by Hammami et al. (6) found epidural abscesses to be more prevalent in the TS group than in other types of infectious spondylitis (18,19). Due to data inconsistencies, no conclusive comparison could be drawn from this study, highlighting the need for a comprehensive study involving the four types of infectious spondylitis. Ultimately, these features were not distinguishable among the four types of infectious spondylitis in our study.

This study has several limitations. First, the sample size is relatively small, and additional studies involving larger cohorts across the four distinct types of infectious spondylitis are needed to validate our findings further. Second, our results differ from those reported by Li et al. (1). Specifically, we found a higher prevalence of endplate inflammatory reaction lines in the FS group compared to the PS group, which presents a discrepancy and warrants confirmation through additional research. Furthermore, our findings suggest that BS is more likely to involve multiple vertebral bodies and that posterior element involvement is not exclusive to TS, contrary to previously established reports. These findings should be verified through more comprehensive studies.

Moreover, the model’s high AUC values indicate strong discriminative ability, but its relatively lower sensitivity suggests it may not be optimizing for balanced classification. This could be due to the model prioritizing overall accuracy over precise identification, particularly at the decision boundary chosen. To address this, future work could involve fine-tuning the decision threshold to better balance sensitivity and specificity, and exploring additional feature engineering to enhance model performance. Additionally, the online application, while enhancing accessibility, should be continuously updated and validated in clinical settings to ensure it captures the full complexity of model performance.

Conclusions

This study comprehensively analyzed the MRI and laboratory features of pyogenic, tuberculous, brucellar, and fungal spondylitis, highlighting distinct characteristics for each type. The RF model demonstrated superior predictive performance, effectively distinguishing between the infections with high accuracy and AUC values. This predictive tool could significantly enhance diagnostic precision and facilitate timely, appropriate interventions. The findings underscore the importance of integrating imaging and clinical features for accurate classification of infectious spondylitis, potentially improving patient outcomes. Future studies with larger cohorts are needed to further validate these results and explore the broader application of this predictive model in clinical practice.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-605/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-605/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-605/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study has been approved by the Ethics Committee of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (No. TJ-IRB202407106) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Li T, Li W, Du Y, Gao M, Liu X, Wang G, Cui H, Jiang Z, Cui X, Sun J. Discrimination of pyogenic spondylitis from brucellar spondylitis on MRI. Medicine (Baltimore) 2018;97:e11195. [Crossref] [PubMed]
2. Zimmerli W. Clinical practice. Vertebral osteomyelitis. N Engl J Med 2010;362:1022-9. [Crossref] [PubMed]
3. Berbari EF, Kanj SS, Kowalski TJ, Darouiche RO, Widmer AF, Schmitt SK, Hendershot EF, Holtom PD, Huddleston PM 3rd, Petermann GW, Osmon DRInfectious Diseases Society of America. 2015 Infectious Diseases Society of America (IDSA) Clinical Practice Guidelines for the Diagnosis and Treatment of Native Vertebral Osteomyelitis in Adults. Clin Infect Dis 2015;61:e26-46. [Crossref] [PubMed]
4. Ahn KS, Kang CH, Hong SJ, Kim BH, Shim E. The correlation between follow-up MRI findings and laboratory results in pyogenic spondylodiscitis. BMC Musculoskelet Disord 2020;21:428. [Crossref] [PubMed]
5. Patel KB, Poplawski MM, Pawha PS, Naidich TP, Tanenbaum LN. Diffusion-weighted MRI “claw sign” improves differentiation of infectious from degenerative modic type 1 signal changes of the spine. AJNR Am J Neuroradiol 2014;35:1647-52. [Crossref] [PubMed]
6. Hammami F, Koubaa M, Feki W, Chakroun A, Rekik K, Smaoui F, Marrakchi C, Mnif Z, Jemaa MB. Tuberculous and Brucellar Spondylodiscitis: Comparative Analysis of Clinical, Laboratory, and Radiological Features. Asian Spine J 2021;15:739-46. [Crossref] [PubMed]
7. Lee SW, Lee SH, Chung HW, Kim MJ, Seo MJ, Shin MJ. Candida spondylitis: Comparison of MRI findings with bacterial and tuberculous causes. AJR Am J Roentgenol 2013;201:872-7. [Crossref] [PubMed]
8. Galhotra RD, Jain T, Sandhu P, Galhotra V. Utility of magnetic resonance imaging in the differential diagnosis of tubercular and pyogenic spondylodiscitis. J Nat Sci Biol Med 2015;6:388-93. [Crossref] [PubMed]
9. Lee Y, Kim BJ, Kim SH, Lee SH, Kim WH, Jin SW. Comparative Analysis of Spontaneous Infectious Spondylitis: Pyogenic versus Tuberculous. J Korean Neurosurg Soc 2018;61:81-8. [Crossref] [PubMed]
10. Qureshi KA, Parvez A, Fahmy NA, Abdel Hady BH, Kumar S, Ganguly A, Atiya A, Elhassan GO, Alfadly SO, Parkkila S, Aspatwar A. Brucellosis: epidemiology, pathogenesis, diagnosis and treatment-a comprehensive review. Ann Med 2023;55:2295398. [Crossref] [PubMed]
11. Xu Z, Zhu W, Zhou S, Zhao Y, Xiang Q, Zhang Y. Aspergillus fumigatus spondylitis in an immunocompetent patient with annular high signal around the intervertebral disks: a case report and literature review. Front Med (Lausanne) 2024;11:1532282. [Crossref] [PubMed]
12. Gupta N, Kadavigere R, Malla S, Bhat SN, Saravu K. Differentiating tubercular from pyogenic causes of spine involvement on Magnetic Resonance Imaging. Infez Med 2022;31:62-9. [Crossref] [PubMed]
13. Frel M, Białecki J, Wieczorek J, Paluch Ł, Dąbrowska-Thing A, Walecki J. Magnetic Resonance Imaging in Differentatial Diagnosis of Pyogenic Spondylodiscitis and Tuberculous Spondylodiscitis. Pol J Radiol 2017;82:71-87. [Crossref] [PubMed]
14. Ahmad-Mansour N, Loubet P, Pouget C, Dunyach-Remy C, Sotto A, Lavigne JP, Molle V. Staphylococcus aureus Toxins: An Update on Their Pathogenic Properties and Potential Treatments. Toxins (Basel) 2021;13:677. [Crossref] [PubMed]
15. Adelhoefer SJ, Gonzalez MR, Bedi A, Kienzle A, Bäcker HC, Andronic O, Karczewski D. Candida spondylodiscitis: a systematic review and meta-analysis of seventy two studies. Int Orthop 2024;48:5-20. [Crossref] [PubMed]
16. Liang C, Wei W, Liang X, De E, Zheng B. Spinal brucellosis in Hulunbuir, China, 2011-2016. Infect Drug Resist 2019;12:1565-71. [Crossref] [PubMed]
17. Liu YX, Lei F, Zheng LP, Yuan H, Zhou QZ, Feng DX. A diagnostic model for differentiating tuberculous spondylitis from pyogenic spondylitis: a retrospective case-control study. Sci Rep 2023;13:10337. [Crossref] [PubMed]
18. Strauss SB, Gordon SR, Burns J, Bello JA, Slasky SE. Differentiation between Tuberculous and Pyogenic Spondylodiscitis: The Role of the Anterior Meningovertebral Ligament in Patients with Anterior Epidural Abscess. AJNR Am J Neuroradiol 2020;41:364-8. [Crossref] [PubMed]
19. Yoon YK, Jo YM, Kwon HH, Yoon HJ, Lee EJ, Park SY, Park SY, Choo EJ, Ryu SY, Lee MS, Yang KS, Kim SW. Differential diagnosis between tuberculous spondylodiscitis and pyogenic spontaneous spondylodiscitis: a multicenter descriptive and comparative study. Spine J 2015;15:1764-71. [Crossref] [PubMed]
20. Kim YS, Lee HM, Kim JK, Yang CS, Kim TS, Jung M, Jin HS, Kim S, Jang J, Oh GT, Kim JM, Jo EK. PPAR-α Activation Mediates Innate Host Defense through Induction of TFEB and Lipid Catabolism. J Immunol 2017;198:3283-95. [Crossref] [PubMed]
21. Knight M, Braverman J, Asfaha K, Gronert K, Stanley S. Lipid droplet formation in Mycobacterium tuberculosis infected macrophages requires IFN-γ/HIF-1α signaling and supports host defense. PLoS Pathog 2018;14:e1006874. [Crossref] [PubMed]
22. Peyron P, Vaubourgeix J, Poquet Y, Levillain F, Botanch C, Bardou F, Daffé M, Emile JF, Marchou B, Cardona PJ, de Chastellier C, Altare F. Foamy macrophages from tuberculous patients’ granulomas constitute a nutrient-rich reservoir for M. tuberculosis persistence. PLoS Pathog 2008;4:e1000204. [Crossref] [PubMed]
23. Buter J, Cheng TY, Ghanem M, Grootemaat AE, Raman S, Feng X, et al. Mycobacterium tuberculosis releases an antacid that remodels phagosomes. Nat Chem Biol 2019;15:889-99. [Crossref] [PubMed]
24. Chandra P, He L, Zimmerman M, Yang G, Köster S, Ouimet M, Wang H, Moore KJ, Dartois V, Schilling JD, Philips JA. Inhibition of Fatty Acid Oxidation Promotes Macrophage Control of Mycobacterium tuberculosis. mBio 2020;11:e01139-20. [Crossref] [PubMed]
25. Lerner TR, Queval CJ, Lai RP, Russell MR, Fearns A, Greenwood DJ, Collinson L, Wilkinson RJ, Gutierrez MG. Mycobacterium tuberculosis cords within lymphatic endothelial cells to evade host immunity. JCI Insight 2020;5:136937. [Crossref] [PubMed]
26. Saini NK, Baena A, Ng TW, Venkataswamy MM, Kennedy SC, Kunnath-Velayudhan S, Carreño LJ, Xu J, Chan J, Larsen MH, Jacobs WR Jr, Porcelli SA. Suppression of autophagy and antigen presentation by Mycobacterium tuberculosis PE_PGRS47. Nat Microbiol 2016;1:16133. [Crossref] [PubMed]
27. Strong EJ, Jurcic Smith KL, Saini NK, Ng TW, Porcelli SA, Lee S. Identification of Autophagy-Inhibiting Factors of Mycobacterium tuberculosis by High-Throughput Loss-of-Function Screening. Infect Immun 2020;88:e00269-20. [Crossref] [PubMed]
28. de Figueiredo P, Ficht TA, Rice-Ficht A, Rossetti CA, Adams LG. Pathogenesis and immunobiology of brucellosis: review of Brucella-host interactions. Am J Pathol 2015;185:1505-17. [Crossref] [PubMed]
29. Badiee P, Ghasemi F, Jafarian H. Role of biomarkers in the diagnosis of invasive aspergillosis in immunocompromised patients. Ann Clin Microbiol Antimicrob 2022;21:44. [Crossref] [PubMed]
30. Plebani M. Why C-reactive protein is one of the most requested tests in clinical laboratories? Clin Chem Lab Med 2023;61:1540-5. [Crossref] [PubMed]
31. Beamer N, Coull BM, Sexton G, de Garmo P, Knox R, Seaman G. Fibrinogen and the albumin-globulin ratio in recurrent stroke. Stroke 1993;24:1133-9. [Crossref] [PubMed]
32. Choe H, Kamono E, Abe K, Hieda Y, Ike H, Kumagai K, Kobayashi N, Inaba Y. Accuracy of Albumin, Globulin, and Albumin-Globulin Ratio for Diagnosing Periprosthetic Joint Infection: A Systematic Review and Meta-Analysis. J Clin Med 2023;12:7512. [Crossref] [PubMed]
33. Hill LA, Bodnar TS, Weinberg J, Hammond GL. Corticosteroid-binding globulin is a biomarker of inflammation onset and severity in female rats. J Endocrinol 2016;230:215-25. [Crossref] [PubMed]
34. Cooper AM. Cell-mediated immune responses in tuberculosis. Annu Rev Immunol 2009;27:393-422. [Crossref] [PubMed]
35. Martirosyan A, Gorvel JP. Brucella evasion of adaptive immunity. Future Microbiol 2013;8:147-54. [Crossref] [PubMed]
36. Quinlan GJ, Martin GS, Evans TW. Albumin: biochemical properties and therapeutic potential. Hepatology 2005;41:1211-9. [Crossref] [PubMed]
37. Achkar JM, Casadevall A. Antibody-mediated immunity against tuberculosis: implications for vaccine development. Cell Host Microbe 2013;13:250-62. [Crossref] [PubMed]
38. Achkar JM, Chan J, Casadevall A. B cells and antibodies in the defense against Mycobacterium tuberculosis infection. Immunol Rev 2015;264:167-81. [Crossref] [PubMed]
39. Baltierra-Uribe SL, Chanona-Pérez JJ, Méndez-Méndez JV, Perea-Flores MJ, Sánchez-Chávez AC, García-Pérez BE, Moreno-Lafont MC, López-Santiago R. Detection of Brucella abortus by a platform functionalized with protein A and specific antibodies IgG. Microsc Res Tech 2019;82:586-95. [Crossref] [PubMed]
40. Osoba AO, Balkhy H, Memish Z, Khan MY, Al-Thagafi A, Al Shareef B, Al Mowallad A, Oni GA. Diagnostic value of Brucella ELISA IgG and IgM in bacteremic and non-bacteremic patients with brucellosis. J Chemother 2001;13:54-9. [Crossref] [PubMed]
41. Oliveira SC, de Oliveira FS, Macedo GC, de Almeida LA, Carvalho NB. The role of innate immune receptors in the control of Brucella abortus infection: toll-like receptors and beyond. Microbes Infect 2008;10:1005-9. [Crossref] [PubMed]
42. Weiss DS, Takeda K, Akira S, Zychlinsky A, Moreno E. MyD88, but not toll-like receptors 4 and 2, is required for efficient clearance of Brucella abortus. Infect Immun 2005;73:5137-43. [Crossref] [PubMed]
43. Augenstreich J, Briken V. Host Cell Targets of Released Lipid and Secreted Protein Effectors of Mycobacterium tuberculosis. Front Cell Infect Microbiol 2020;10:595029. [Crossref] [PubMed]
44. Srivastava S, Ernst JD. Cutting edge: Direct recognition of infected cells by CD4 T cells is required for control of intracellular Mycobacterium tuberculosis in vivo. J Immunol 2013;191:1016-20. [Crossref] [PubMed]
45. Batirel A, Erdem H, Sengoz G, Pehlivanoglu F, Ramosaco E, Gülsün S, et al. The course of spinal tuberculosis (Pott disease): results of the multinational, multicentre Backbone-2 study. Clin Microbiol Infect 2015;21:1008.e9-1008.e18. [Crossref] [PubMed]
46. Zhang N, Zeng X, He L, Liu Z, Liu J, Zhang Z, Chen X, Shu Y. The Value of MR Imaging in Comparative Analysis of Spinal Infection in Adults: Pyogenic Versus Tuberculous. World Neurosurg 2019;128:e806-13. [Crossref] [PubMed]
47. Liu X, Zheng M, Sun J, Cui X. A diagnostic model for differentiating tuberculous spondylitis from pyogenic spondylitis on computed tomography images. Eur Radiol 2021;31:7626-36. [Crossref] [PubMed]
48. Liu X, Zheng M, Jiang Z, Wang G, Li T, Sun J, Cui X. Computed tomography imaging characteristics help to differentiate pyogenic spondylitis from brucellar spondylitis. Eur Spine J 2020;29:1490-8. [Crossref] [PubMed]
49. Liu X, Li H, Jin C, Niu G, Guo B, Chen Y, Yang J. Differentiation Between Brucellar and Tuberculous Spondylodiscitis in the Acute and Subacute Stages by MRI: A Retrospective Observational Study. Acad Radiol 2018;25:1183-9. [Crossref] [PubMed]
50. Colmenero JD, Jiménez-Mejías ME, Sánchez-Lora FJ, Reguera JM, Palomino-Nicás J, Martos F, García de las Heras J, Pachón J. Pyogenic, tuberculous, and brucellar vertebral osteomyelitis: a descriptive and comparative study of 219 cases. Ann Rheum Dis 1997;56:709-15. [Crossref] [PubMed]
51. Colmenero JD, Reguera JM, Martos F, Sánchez-De-Mora D, Delgado M, Causse M, Martín-Farfán A, Juárez C. Complications associated with Brucella melitensis infection: a study of 530 cases. Medicine (Baltimore) 1996;75:195-211. [Crossref] [PubMed]
52. Hong SH, Choi JY, Lee JW, Kim NR, Choi JA, Kang HS. MR imaging assessment of the spine: infection or an imitation? Radiographics 2009;29:599-612. [Crossref] [PubMed]
53. Dagirmanjian A, Schils J, McHenry MC. MR imaging of spinal infections. Magn Reson Imaging Clin N Am 1999;7:525-38.
54. Pintor IA, Pereira F, Cavadas S, Lopes P. Pott’s disease (tuberculous spondylitis). Int J Mycobacteriol 2022;11:113-5. [Crossref] [PubMed]
55. Zhu W, Zhou S, Zhang J, Li L, Liu P, Xiong W. Differentiation of Native Vertebral Osteomyelitis: A Comprehensive Review of Imaging Techniques and Future Applications. Med Sci Monit 2024;30:e943168. [Crossref] [PubMed]
56. Smith EE, Rosand J, Greenberg SM. Hemorrhagic stroke. Neuroimaging Clin N Am 2005;15:259-72. ix. [Crossref] [PubMed]