CT-based radiomics nomogram for distinguishing Mycoplasma pneumoniae pneumonia from other pneumonias in children with community-acquired pneumonia

Introduction

Community-acquired pneumonia (CAP) is a common disease in children, with a high incidence rate and mortality, thereby seriously threatening children’s health (1). Mycoplasma pneumoniae pneumonia (MPP) accounts for approximately 10–40% of CAP in hospitalized children. And the incidence of Mycoplasma pneumoniae (MP) infection increases year by year (1,2), which is becoming a clinical problem of extensive concern by pediatricians.

MP is one of the most important pathogens of respiratory system infection in not only preschool and school-age children, but also in infants aged 1–3 years in recent years. There are pathological changes of MPP in the respiratory system, and as the disease progresses, it can cause systemic inflammatory response syndrome and multisystem damage. The clinical manifestations of refractory or severe pneumonia are complex and can have various complications, such as pleural effusion, atelectasis, and so on (1). Some problems, such as unreasonable application of antibiotics and lack of pertinence in the selection of examination methods, also exist. Unlike pneumonia caused by bacteria or viruses, the treatment of MPP requires macrolide antibiotics. Early rational use of antibiotics improves disease outcomes and can reduce the length of hospital stay. Thus, early diagnosis is especially important to guide clinical treatment (3). The current etiological diagnosis of MPP primarily depends on serology and nucleic acid detection. Mycoplasma culture is the gold standard for diagnosis, but due to its poor culture conditions and slow growth, it lacks early diagnostic value (4). Children exhibit relatively high levels of difficulty in cooperating during sputum culture and sampling procedures. Compared with etiological-detection methods, although the results of antibody detection are unaffected by antibiotic treatment, time window, and individual differences exist in antibody production. Antibody detection often takes more than 1 week to produce the corresponding immunoglobulin M (IgM) after being infected with pneumonia, and then it lasts for 1–2 months, which is not conducive to the pathogen detection in rapidly progressing pneumonia (5). In a certain proportion of initial infection and reinfection, very young children possibly do not produce IgM (6,7).

X-ray and ultrasound are commonly used imaging examination methods for pneumonia. However, chest computed tomography (CT) is required in the following situations: (I) suspected or confirmed severe pneumonia; (II) poor response to initial empirical treatment, so it is necessary to evaluate the complications; (III) the clinical symptoms are inconsistent with the chest X-ray findings, or the X-ray findings suggest that there are complex lesions that need further evaluation by CT (8,9). The CT manifestations of different causes of pneumonia often overlap with each other, and the imaging manifestations of MPP lack obvious specificity (10,11), which makes it difficult to distinguish MPP from bacterial pneumonia, viral pneumonia or mixed pneumonia solely based on visual assessment in clinical practice (12). This limits the traditional role of CT in pathogen identification. As a result, the qualitative diagnosis of CT manifestations of pneumonia relies on the clinical experience of radiologists and has high requirements for imaging doctors. Therefore, a new method is needed to help clinicians identify the pathogen of pneumonia by tapping into the potential value of CT.

In 2012, Lambin et al. (13,14) formally put forward the concept of radiomics, that is, by extracting a good deal of high-throughput radiomic features from medical imaging. Radiomics can predict the results alone or in combination with additional information to support clinical decision-making and achieve accurate medicine (15). Radiomics has made considerable progress in the differentiation of malignant and benign tumors, risk assessment, prognosis prediction, and so on. Since the breakout of coronavirus disease 2019 (COVID-19), a number of relevant radiomic studies have been reported. Some researchers have used CT-based radiomics to distinguish between COVID-19 and other pneumonias (16-19). They have also used radiomics to predict the severity and prognosis of patients with COVID-19 (20-26). And some researchers have used radiomics respectively to differentiate acute paraquat lung injury from pneumonia, MP from bacterial pneumonia, and primary progressive pulmonary tuberculosis (TB) from CAP of children (27-29). These studies have shown that the radiomics has a certain ability to identify and predict lung inflammation.

The aim of the present research is to establish and test a radiomics nomogram combining radiomics signature (Rad-score) and clinical independent risk factors to distinguish MPP from other pneumonias in children, as well as providing the basis for inchoate clinical diagnosis and accurate treatment. We present this article in accordance with the TRIPOD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0076/rc).

Methods

Patients

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Ethics Committee of Yantai Yuhuangding Hospital (No. 2025-725), and the informed consent requirement was waived because this was a retrospective study using anonymized clinical and imaging data. This study analyzed the clinical and imaging data of children aged 1 month to 14 years who were diagnosed with pneumonia by CT from January 2018 to December 2020 in our hospital. The inclusion criteria were as follows: (I) patients diagnosed with CAP; and (II) definite positive results of etiological examination, as indicated by positive for bacterial culture or positive for bacterial, mycoplasma, or viral nucleic acid in sputum or alveolar lavage fluid, and positive result of the condensation set test or enzyme-linked immunosorbent assay (ELISA) test was positive. The exclusion criteria were as follows: (I) patients with poor CT image quality; (II) patients with previous bronchial asthma, history of uncured severe pneumonia, immunosuppression or deficiency, complicated with pulmonary TB and other lung diseases; (III) the lesion was small or the boundary was unclear, such that it cannot be delineated; and (IV) patients with incomplete clinical data. Children with MP nucleic acid positive or MP result in cold agglutination test with titer >1:160 or antibody titer of MP-IgM in convalescent and acute phase more than four times higher than the initial result were included in the MPP group; otherwise, they were included in other pneumonia groups. All enrolled patients were reviewed at the time of grouping by a 20-year associate professor of pediatrics. Considering that pneumonia may have extensive lesions in the lungs, when multiple lesions were present in the same patient, every unrelated lesion in different lobes was considered as a case of research object. Finally, a total of 387 lesions of 325 patients were incorporated in our study, including 162 lesions with MPP and 225 lesions with other pneumonias. Multiple lesions of the same patient were placed in the same training set or testing set, and then we randomly divided them into the training set (n=268) and testing (n=119) set at a ratio of 7:3, as shown in Figure 1.

Figure 1 Flow chart of patients’ inclusion and grouping. CAP, community-acquired pneumonia; CT, computed tomography.

Clinical characteristics

Clinical characteristics were recorded from the medical-record system, including gender, age in months, white blood cell (WBC) count, neutrophil (N) count and its percentage, lymphocyte (L) count and its percentage, C-reactive protein (CRP), platelet (PLT), procalcitonin (PCT), lactate dehydrogenase (LDH), hemoglobin (Hb), red blood cell (RBC) count and whether there was pleural effusion.

CT image protocols

The CT equipment were Brilliance 64, iCT 256 (PHILIPS, Amsterdam, the Netherlands), Light Speed 64 (GE, Chicago, IL, USA), and SOMATOM Definition AS 128 (Siemens, Munich, Germany). The CT scanning parameters were as follows: tube voltage of 80–100 kV, automatic tube current modulation technique, layer thickness of 5 mm, reconstruction thickness of 1.25 or 1 mm, matrix of 512 mm × 512 mm, lung window width/level of 1,500/−550 Hounsfield units (HU), and mediastinum window width/level of 350/40 HU. During the scanning process, shielding devices were used to protect non-examination areas of children. All CT images were exported in Digital Imaging and Communications in Medicine (DICOM) formats.

Image segmentation and radiomic features extraction

Radiologist 1, who was blinded to the clinical diagnosis and had 5 years of experience in respiratory imaging diagnosis, used the ITK-SNAP software (version 3.8.0) to delineate the volume of interest (VOI) manually along the edge of the lesions. When delineating the VOIs, normal lung tissue, pleura, pleural effusion, and blood vessels outside the lesion were avoided (Figure 2). Prior to radiomic feature extraction, the CT image preprocessing was conducted, including gray level standardization, gray level discretization, and image resampling. Following the Image Biomarker Standardisation Initiative (IBSI) guidelines, radiomic features were extracted using Pyradiomics software, including first-order statistics features, shape features, and texture features. The reproducibility of radiomic feature extraction was evaluated by inter- and intra-correlation coefficients (ICCs). In total, the CT images of 50 research objects were randomly selected. After 2 months, radiologist 1 repeated feature extraction, and radiologist 2, who had 8 years of experience in respiratory imaging diagnosis, used the same method to delineate the VOIs and extract radiomic features. And ICCs >0.75 were considered to have high repeatability.

Figure 2 Examples of VOI segmentation on lung window of CT images. The orange line represents the manually delineated boundary of the VOI. CT, computed tomography; VOI, volume of interest.

Radiomic features selection and Rad-score construction

First, radiomic features with ICCs >0.75 were retained in the training set. Analysis of variance (ANOVA) and least absolute shrinkage and selection operator regression (LASSO) were used to screen the optimal radiomic features in the training set. For the ANOVA algorithm, radiomic features with P<0.05 were retained. For the LASSO algorithm, we determined the optimal LASSO alpha parameter by five-fold cross-validation and finally selected radiomic features with non-zero coefficients in the training set. Then, the radiomics signature (Rad-score) of each lesion was calculated through linear combinations of the screened features weighted by their LASSO coefficients.

Development of the clinical model

Univariate and multivariate logistic regression analyses were applied to screen the independent clinical risk factors in the training set, including gender, month age, CRP, PLT, WBC count, N count and its percentage, L count and its percentage, PCT, LDH, RBC count, Hb, and whether pleural effusion existed. In multivariate logistic regression, characteristics with P<0.05 were considered as clinical independent risk factors. The clinical model, including the clinical independent risk factors, was developed by multivariate logistic regression analysis.

Development and validation of radiomics nomogram

A radiomics nomogram combining Rad-score and clinical independent risk factors was then developed by multivariate logistic regression analysis. Receiver operating characteristic (ROC) curves were used to evaluate the performance of the radiomics nomogram. The area under the curve (AUC), specificity, sensitivity, and accuracy of the model were then calculated. For comparison, the Rad-score model was also developed.

The calibration curve was used to evaluate the consistency between the predicted probability and the observed results of the radiomics nomogram through the Hosmer-Lemeshow test. Decision curve analysis (DCA) was performed to assess the clinical applicability of three prediction models according to the net benefit and the corresponding threshold probability.

Statistical analysis

All statistical analyses were performed in SPSS (version 26), Python (version 3.6), and R software (version 3.4.1). To evaluate the robustness of the radiomics nomogram, we performed internal validation using bootstrap resampling with 1,000 iterations. The three models were developed in the training set, whereas the testing set was utilized only to test the models. We used the independent sample t-test for continuous variables and the Chi-squared test for categorical variables to compare clinical characteristics between the two sets. The DeLong test was employed to compare the differences in AUCs between any two different models. P<0.05 was considered as statistical significance.

Results

Clinical characteristics

Table 1 presents the detailed clinical characteristics of patients between the training and testing sets. The results showed statistical significant differences in month age (P<0.001), PLT (P=0.03), WBC count (P=0.004), N percentage (P<0.001), L count and its percentage (P<0.001), Hb (P=0.01), and whether pleural effusion existed (P<0.001) between MPP and other pneumonias in the training set, However, no significant differences existed in other variables.

Feature selection and radiomics signature construction

A total of 1,409 radiomic features were extracted from each VOI. Between the same radiologist 1 and two different radiologists, the inter- and intra-observer reproducibility of feature extraction reached ICCs >0.75. After the ANOVA algorithm, 852 radiomic features with P<0.05 were retained. After the LASSO algorithm, 14 radiomic features were retained with five-fold cross-validation (Figure 3). Fourteen radiomic features and their coefficients are shown in Table 2. The Rad-score was calculated using 14 features, and the formula was: Rad-score = β₁x₁ + β₂x₂ + ... + β₁₄x₁₄, where β₁–β₁₄ are the coefficients of radiomic features and x₁–x₁₄ are the values of radiomic features.

Figure 3 LASSO algorithm for radiomic features selection. (A) A total of 14 radiomic features with non-zero coefficient were selected. The Y-axis represents coefficient of each feature. (B) Mean square error path using five-fold cross-validation. CV, cross-validation; LASSO, least absolute shrinkage and selection operator.

Development of the radiomics nomogram

In the training set, after univariate logistic regression, the L count and its percentage, PLT, month age, N percentage, WBC count, pleural effusion, and Hb were input into multivariate logistic regression. And in multivariate logistic regression analysis, WBC count and L count were identified as independent risk factors of MPP. The radiomics nomogram was developed by combining WBC count, L count, and Rad-score (Figure 4). For the purpose of evaluating the value of the radiomics nomogram, WBC count and L count were used to establish the clinical model.

Figure 4 Radiomics nomogram with Rad-score, WBC count, and L count incorporated. L, lymphocyte; WBC, white blood cell.

Validation of the radiomics nomogram

The ROC curves of three models are displayed in Figure 5. In the training set, the radiomics nomogram, Rad-score, and clinical model yielded AUCs of 0.874 [95% confidence interval (CI): 0.836–0.908], 0.857 (95% CI: 0.814–0.892), and 0.663 (95% CI: 0.610–0.719), respectively. In the testing set, the radiomics nomogram, Rad-score, and clinical model yielded AUCs of 0.841 (95% CI: 0.771–0.893), 0.825 (95% CI: 0.755–0.882), and 0.597 (95% CI: 0.508–0.670), respectively. The predictive performance of the three models is displayed in Table 3. The radiomics nomogram demonstrated higher specificity and accuracy in distinguishing MPP from other pneumonias compared to the clinical model and Rad-score. The DeLong test demonstrated that statistical significant differences between the radiomics nomogram and clinical model (P<0.001), but no distinct difference between the radiomics nomogram and Rad-score (P=0.16) in the testing set.

Figure 5 ROC curves of radiomics nomogram, Rad-score, and clinical model in the training (A) and testing (B) sets. AUC, area under the curve; ROC, receiver operating characteristic.

Figure 6A,6B illustrate the DCA of three models in the training and testing sets, respectively. The results demonstrated that the radiomics nomogram had more net benefit than the clinical model. The calibration curves are presented in Figure 6C, which demonstrates excellent consistency between the predicted outcome and actual observation in both the training and testing sets.

Figure 6 The DCA and calibration curve in the training and testing sets. (A,B) DCA of three models in the training and testing sets, respectively. The Y-axis represents the net benefits, while the X-axis represents the threshold probability. (C) Calibration curves of radiomics nomogram. The diagonal line represents the perfect prediction between the predicted outcome and actual observation. DCA, decision curve analysis.

Discussion

In this study, we developed and tested a CT-based radiomics nomogram combined with the Rad-score and independent clinical risk factors to distinguish MPP from other pneumonias in children with CAP. The results showed that the prediction efficiency in the radiomics nomogram was significantly better than that in the clinical model. The radiomics nomogram is suitable for identifying MMP and other pneumonias, and can provide a reference of the diagnosis for clinicians.

Polymerase chain reaction (PCR) serves as the gold standard for diagnosing MPP, but it is costly and a significant portion of healthy children have MP as colonizing bacteria in their respiratory tracts (30). The production of serum antibodies has a lag time, and a positive IgM result may indicate an infection that occurred several weeks ago rather than the current infection (31). When the pathogen of pneumonia is unknown, empirical use of antibiotics is the main treatment method, using beta-lactam antibiotics. However, treatment for MPP requires the use of macrolide antibiotics (32). Early use of glucocorticoids and gamma globulin is effective for severe MP (33). Glucocorticoids are usually not used in the treatment of bacterial pneumonia. For children who have undergone CT examinations due to the need of their condition, our research can serve as an auxiliary tool for excluding MPP, to some extent, compensating for the aforementioned shortcomings. Our research achieved acceptable results: the AUC, sensitivity, specificity, and accuracy of the testing set were 0.841, 0.608, 0.941, and 0.798, respectively. From the results, the discriminant ability of the radiomics nomogram was obviously better than that of the clinical model in the testing set (AUC, 0.841 vs. 0.597, P<0.001). However, the relatively low sensitivity (0.608) of this model may lead to the missed diagnosis of some MPP cases, limiting its potential as an independent diagnostic tool. This may be due to the fact that the “other pneumonia” group includes various types, such as bacterial, viral, and mixed infections. Some non-MPP pneumonia cases may exhibit imaging characteristics that highly overlap with MPP, leading the model to correctly classify them as “non-MPP” (high specificity), but it is more conservative in identifying MPP lesions (lower sensitivity). From a clinical perspective, this high specificity remains of great value: excluding MPP helps avoid unnecessary use of macrolide antibiotics.

The independent clinical risk factors included in the radiomics nomogram were WBC count and L count. This consequence was also consistent with the clinical test results of general MP: the WBC count of children with MPP often slightly increases or does not increase (34). The addition of independent clinical risk factors did not significantly improve the performance of the Rad-score and radiomics nomogram in the testing set (AUC, 0.825 vs. 0.841, P=0.16). This finding may be due to the following reasons: (I) we regarded multiple lesions of the same child as multiple subjects, making the clinical information of some subjects consistent; and (II) some children had been treated when they were examined or tested, which might have an impact. In addition, the dataset exhibits a mild class imbalance with MPP accounting for approximately 40% of cases, a proportion that aligns with the epidemiological characteristics of pediatric MPP during the epidemic seasons. This imbalance may contribute to the model’s high specificity and moderate sensitivity, indicating that it is more reliable in excluding MPP than in diagnosing MPP.

Beyond model performance, we further explored which classes of radiomic features contributed most to distinguishing MPP from other pneumonias. Among the 14 selected features, texture-based features predominated, particularly those derived from wavelet filters. These texture features reflect intralesional heterogeneity, indicating that MPP lesions exhibit more complex and non-uniform internal texture patterns compared to other pneumonias. This is consistent with the clinical observation that MPP often presents with peribronchial thickening, nodular opacities, and patchy ground-glass opacities, whereas bacterial pneumonias typically show more homogeneous consolidation. In contrast, first-order statistics features and shape features played a relatively limited role, suggesting that relying solely on the overall lesion density differences between groups and lesion morphological characteristics is insufficient for achieving pathogen differential diagnosis in pediatric CAP.

In recent radiomics studies related to pediatric pneumonia, Wang et al. (28) proposed a CT-based predictive nomogram for the differential diagnosis of pulmonary TB and CAP in children, with an AUC of 0.971 in the validation cohort. Wang et al. (35) used six classifiers to construct models based on CT radiomic features to differentiate MPP from Streptococcal pneumoniae pneumonia in children under 5 years of age. The optimal performance classifier is the random forest (RF), with an AUC of 0.822 in the validation cohort. Studies have shown that the co-infection of viruses, bacteria, and MP is common in children with CAP (36). In a case where determining whether the child had a single-pathogen infection or a mixed infection was impossible, our radiomics nomogram could exclude the pneumonia caused by MP infection, making the radiomics nomogram more suitable in clinical practice. Liu et al. (37) used high-resolution CT-based radiomics to differentiate MPP and mycoplasma co-infection with other pathogens (Co-MPP), yielding an AUC of 0.951. However, the sample size of this study was relatively small, with only 124 pediatric patients included. Furthermore, only the largest layer of the lesion was manually segmented, ignoring the overall heterogeneity of the lesion. Table 4 summarizes and compares recent radiomics studies focusing on pediatric pneumonia.

This retrospective study had several limitations. Firstly, the sample of this study originated from a single medical institution, and its size was not sufficiently large. The expansion of sample size and multi-center cooperation will enable our research results to be more generally applicable. Secondly, the VOIs were manually delineated by a radiologist. Fully automatic or semi-automatic segmentation methods can be used in the future, which is also the main trend of development (38). Thirdly, this study analyzed multiple lesions of the same child as independent samples, which may introduce statistical dependence and limit the direct applicability of the model for patient-level classification. Future research will adopt a patient-centered analysis strategy to validate and generalize the preliminary findings of this study while properly accounting for within-patient correlation.

Conclusions

In conclusion, the CT-based radiomics nomogram can be used as a non-invasive and individualized method of distinguishing MPP from other pneumonias in children with CAP, which can provide evidence for early clinical diagnosis and accurate treatment.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0076/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0076/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0076/prf

Funding: This study was supported by the Shandong Provincial Medical and Health Technology Project (No. 202406010538).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0076/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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Ethics Committee of Yantai Yuhuangding Hospital (No. 2025-725), and the informed consent requirement was waived because this was a retrospective study using anonymized clinical and imaging data.

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. Jain S, Williams DJ, Arnold SR, et al. Community-acquired pneumonia requiring hospitalization among U.S. children. N Engl J Med 2015;372:835-45.
2. Liu WK, Liu Q, Chen DH, et al. Epidemiology of acute respiratory infections in children in Guangzhou: a three-year study. PLoS One 2014;9:e96674. [Crossref] [PubMed]
3. Yang TI, Chang TH, Lu CY, et al. Mycoplasma pneumoniae in pediatric patients: Do macrolide-resistance and/or delayed treatment matter? J Microbiol Immunol Infect 2019;52:329-35. [Crossref] [PubMed]
4. Liu J, Zhao F, Lu J, et al. High Mycoplasma pneumoniae loads and persistent long-term Mycoplasma pneumoniae DNA in lower airway associated with severity of pediatric Mycoplasma pneumoniae pneumonia. BMC Infect Dis 2019;19:1045. [Crossref] [PubMed]
5. Lee SC, Youn YS, Rhim JW, et al. Early Serologic Diagnosis of Mycoplasma pneumoniae Pneumonia: An Observational Study on Changes in Titers of Specific-IgM Antibodies and Cold Agglutinins. Medicine (Baltimore) 2016;95:e3605. [Crossref] [PubMed]
6. Loens K, Goossens H, Ieven M. Acute respiratory infection due to Mycoplasma pneumoniae: current status of diagnostic methods. Eur J Clin Microbiol Infect Dis 2010;29:1055-69. [Crossref] [PubMed]
7. Waites KB, Balish MF, Atkinson TP. New insights into the pathogenesis and detection of Mycoplasma pneumoniae infections. Future Microbiol 2008;3:635-48. [Crossref] [PubMed]
8. de Benedictis FM, Kerem E, Chang AB, et al. Complicated pneumonia in children. Lancet 2020;396:786-98. [Crossref] [PubMed]
9. Jaffe A, Calder AD, Owens CM, et al. Role of routine computed tomography in paediatric pleural empyema. Thorax 2008;63:897-902. [Crossref] [PubMed]
10. Wang J, Xia C, Sharma A, et al. Chest CT Findings and Differential Diagnosis of Mycoplasma pneumoniae Pneumonia and Mycoplasma pneumoniae Combined with Streptococcal Pneumonia in Children. J Healthc Eng 2021;2021:8085530. [Crossref] [PubMed]
11. Reittner P, Müller NL, Heyneman L, et al. Mycoplasma pneumoniae pneumonia: radiographic and high-resolution CT features in 28 patients. AJR Am J Roentgenol 2000;174:37-41. [Crossref] [PubMed]
12. Lee I, Kim TS, Yoon HK. Mycoplasma pneumoniae pneumonia: CT features in 16 patients. Eur Radiol 2006;16:719-25. [Crossref] [PubMed]
13. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441-6. [Crossref] [PubMed]
14. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749-62. [Crossref] [PubMed]
15. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016;278:563-77. [Crossref] [PubMed]
16. Chen Z, Li X, Li J, et al. A COVID-19 risk score combining chest CT radiomics and clinical characteristics to differentiate COVID-19 pneumonia from other viral pneumonias. Aging (Albany NY) 2021;13:9186-224. [Crossref] [PubMed]
17. Lin L, Liu J, Deng Q, et al. Radiomics Is Effective for Distinguishing Coronavirus Disease 2019 Pneumonia From Influenza Virus Pneumonia. Front Public Health 2021;9:663965. [Crossref] [PubMed]
18. Delli Pizzi A, Chiarelli AM, Chiacchiaretta P, et al. Radiomics-based machine learning differentiates “ground-glass” opacities due to COVID-19 from acute non-COVID-19 lung disease. Sci Rep 2021;11:17237. [Crossref] [PubMed]
19. Wang D, Huang C, Bao S, et al. Study on the prognosis predictive model of COVID-19 patients based on CT radiomics. Sci Rep 2021;11:11591. [Crossref] [PubMed]
20. Shi H, Xu Z, Cheng G, et al. CT-based radiomic nomogram for predicting the severity of patients with COVID-19. Eur J Med Res 2022;27:13. [Crossref] [PubMed]
21. Xie Z, Sun H, Wang J, et al. A novel CT-based radiomics in the distinction of severity of coronavirus disease 2019 (COVID-19) pneumonia. BMC Infect Dis 2021;21:608. [Crossref] [PubMed]
22. Zhang M, Zeng X, Huang C, et al. An AI-based radiomics nomogram for disease prognosis in patients with COVID-19 pneumonia using initial CT images and clinical indicators. Int J Med Inform 2021;154:104545. [Crossref] [PubMed]
23. Shiri I, Sorouri M, Geramifar P, et al. Machine learning-based prognostic modeling using clinical data and quantitative radiomic features from chest CT images in COVID-19 patients. Comput Biol Med 2021;132:104304. [Crossref] [PubMed]
24. Cai W, Liu T, Xue X, et al. CT Quantification and Machine-learning Models for Assessment of Disease Severity and Prognosis of COVID-19 Patients. Acad Radiol 2020;27:1665-78. [Crossref] [PubMed]
25. Li L, Wang L, Zeng F, et al. Development and multicenter validation of a CT-based radiomics signature for predicting severe COVID-19 pneumonia. Eur Radiol 2021;31:7901-12. [Crossref] [PubMed]
26. Wu G, Yang P, Xie Y, et al. Development of a clinical decision support system for severity risk prediction and triage of COVID-19 patients at hospital admission: an international multicentre study. Eur Respir J 2020;56:2001104. [Crossref] [PubMed]
27. Li H, Li T, Cai Q, et al. Development and Validation of a Radiomics Nomogram for Differentiating Mycoplasma Pneumonia and Bacterial Pneumonia. Diagnostics (Basel) 2021;11:1330. [Crossref] [PubMed]
28. Wang B, Li M, Ma H, et al. Computed tomography-based predictive nomogram for differentiating primary progressive pulmonary tuberculosis from community-acquired pneumonia in children. BMC Med Imaging 2019;19:63. [Crossref] [PubMed]
29. Yanling W, Duo G, Zuojun G, et al. Radiomics Nomogram Analyses for Differentiating Pneumonia and Acute Paraquat Lung Injury. Sci Rep 2019;9:15029. [Crossref] [PubMed]
30. Chang HY, Chang LY, Shao PL, et al. Comparison of real-time polymerase chain reaction and serological tests for the confirmation of Mycoplasma pneumoniae infection in children with clinical diagnosis of atypical pneumonia. J Microbiol Immunol Infect 2014;47:137-44. [Crossref] [PubMed]
31. Gao L, Sun Y. Laboratory diagnosis and treatment of Mycoplasma pneumoniae infection in children: a review. Ann Med 2024;56:2386636. [Crossref] [PubMed]
32. Wang YS, Zhou YL, Bai GN, et al. Expert consensus on the diagnosis and treatment of macrolide-resistant Mycoplasma pneumoniae pneumonia in children. World J Pediatr 2024;20:901-14. [Crossref] [PubMed]
33. Lee H, Yun KW, Lee HJ, et al. Antimicrobial therapy of macrolide-resistant Mycoplasma pneumoniae pneumonia in children. Expert Rev Anti Infect Ther 2018;16:23-34. [Crossref] [PubMed]
34. Zheng HH, Xiang Y, Wang Y, et al. Clinical value of blood related indexes in the diagnosis of bacterial infectious pneumonia in children. Transl Pediatr 2022;11:114-9. [Crossref] [PubMed]
35. Wang D, Zhao J, Zhang R, et al. The value of CT radiomic in differentiating mycoplasma pneumoniae pneumonia from streptococcus pneumoniae pneumonia with similar consolidation in children under 5 years. Front Pediatr 2022;10:953399. [Crossref] [PubMed]
36. Feng Q, Wang J, Wang X, et al. Clinical epidemiological characteristics of hospitalized pediatric viral community-acquired pneumonia in China. J Infect 2025;90:106450. [Crossref] [PubMed]
37. Liu X, Xu W, Meng L, et al. Development and validation of a machine learning model for predicting co-infection of Mycoplasma pneumonia in pediatric patients. Transl Pediatr 2025;14:1201-12. [Crossref] [PubMed]
38. Mu W, Schabath MB, Gillies RJ. Images Are Data: Challenges and Opportunities in the Clinical Translation of Radiomics. Cancer Res 2022;82:2066-8. [Crossref] [PubMed]