Abstract
Background/Aim: Even though advanced magnetic resonance imaging (MRI) can effectively differentiate between medulloblastoma and ependymoma, it is not readily available throughout the world. This study aimed to investigate the role of simple quantified basic MRI sequences in the differentiation between medulloblastoma and ependymoma in children. Patients and Methods: The institutional review board approved this prospective study. The brain MRI protocol, including sagittal T1-weighted, axial T2-weighted, coronal fluid-attenuated inversion recovery, and axial T1-weighted with contrast enhancement (T1WCE) sequences, was assessed in 26 patients divided into two groups: Medulloblastoma (n=22) and ependymoma (n=4). The quantified region of interest (ROI) values of tumors and their ratios to parenchyma were compared between the two groups. Multivariate logistic regression analysis was utilized to find significant factors influencing the differential diagnosis between the two groups. A generalized estimating equation (GEE) was used to create the predictive model for the discrimination of medulloblastoma from ependymoma. Results: Multivariate logistic regression analysis showed that the T2- and T1WCE-ROI values of tumors and the ratios of T1WCE-ROI values to parenchyma were the most significant factors influencing the diagnosis between these two groups. GEE produced the model: y=exn/(1+exn) with predictor xn=−8.773+0.012x1 − 0.032x2 − 13.228x3, where x1 was the T2-weighted signal intensity (SI) of tumor, x2 the T1WCE SI of tumor, and x3 the T1WCE SI ratio of tumor to parenchyma. The sensitivity, specificity, and area under the curve of the GEE model were 77.3%, 100%, and 92%, respectively. Conclusion: The GEE predictive model can discriminate between medulloblastoma and ependymoma clinically. Further research should be performed to validate these findings.
The most common type of posterior fossa brain tumors in children are medulloblastoma, ependymoma and pilocytic astrocytoma, which have many negative effects on mental and physical development. Among these three groups, the prognosis of medulloblastoma and ependymoma is worse than that of pilocytic astrocytoma (1, 2). The visual characteristics of medulloblastoma and ependymoma are often overlapping, sometimes making the diagnostic process difficult. In addition, the treatment and prognosis of medulloblastoma and ependymoma are different. Therefore, distinguishing between medulloblastoma and ependymoma is essential in the field of pediatric radiology (3-6).
Magnetic resonance imaging (MRI) of the brain is a non-invasive biosecure diagnostic method, utilized as the leading imaging technology for children with a posterior fossa brain tumor (3, 4, 7). MRI is divided into two groups: Basic and advanced MRI sequences. Recent studies have used advanced MRI sequences, including diffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI), in order to improve the ability to differentiate between medulloblastoma and ependymoma (8-10). Although the initial results are very impressive, there are some problematic issues regarding advanced MRI protocols and appropriate analytical tools. These tools are currently not available at all radiology centers around the world, especially not in developing countries (11-13).
Magnetic resonance imaging of a 4-year-old male patient with a tumor in the fourth ventricle confirmed as medulloblastoma post-surgery. A: Sagittal T1-weighted image. B: Axial T2-weighted image. C: Coronal fluid-attenuated inversion recovery image. D: Axial T1-weighted with contrast agent image.
By contrast, conventional MRI techniques are fundamental and can be easily manipulated, without the need for advanced analytical software, at any imaging center in the world. Quantitative MRI has been performed in several previous studies for patients with posterior fossa brain tumors and has yielded some positive results (11-13). However, there is currently no predictive model to differentiate pediatric medulloblastoma from ependymoma.
This clinical study proposes a predictive model to distinguish medulloblastoma and ependymoma based on quantified basic MRI sequence values.
Patients and Methods
The Institutional Review Board of Children's Hospital 02 approved this prospective study (Ref: 352/ND2-CDT). Informed consent was taken from all patients' legal representatives before the MRI procedure. The study was conducted at Children's Hospital 02 for a period of 10 months beginning in February 2019. In this study, 26 patients were included and divided into two groups. Group 1 was children with medulloblastoma (n=22), and group 2 was children with ependymoma (n=4). All patients in this study underwent an MRI followed by either surgery or biopsy for histopathology.
Anesthesia procedure. Induction of anesthesia was by injection of 0.1 mg/kg midazolam (Hameln Pharm GmbH, Germany) followed by 3 mg/kg propofol 1% intravenous infusion (Fresofol, Fresenius Kabi GmbH, Austria).
MRI scan procedure. Pediatric patients were scanned in a supine position using conventional MRI techniques with a 1.5 Tesla MRI machine (Philips, Best, the Netherlands). The conventional MRI techniques consist of the following sequences: T1-weighted (T1W), T2-weighted (T2W), Fluid-attenuated inversion recovery (FLAIR), and T1W with contrast agent (T1WCE). T1WCE was performed using either injection of 0.1 ml/kg Gadovist (Bayer, Germany) or 0.2 ml/kg Dotarem (Guerbet, France). The details of the MRI protocol are presented in Table I.
Collected variables. Conventional MRI was quantified by drawing the region of interest (ROI) on the selected tumoral and parenchymal regions on T1W, T2W, FLAIR, and T1WCE images, providing tumoral and parenchymal signal intensity (SI). The SI ratio of the tumor to the parenchyma was equal to the tumoral SI/parenchymal SI (Figures 1 and 2).
Data analysis. SPSS software version 26 (IBM, Armonk, NY, USA) was used in statistical analysis. Quantitative variables are presented as median and interquartile range. We compared quantitative variables in this study using the Mann–Whitney U-test. Multivariate regression analysis was used to evaluate significant independent indicators in each sequence group to differentiate between medulloblastoma and ependymoma. To build a model of differential diagnosis between the two tumor groups based on significant indicators, the generalized estimating equation (GEE) method was used. Receiver operating characteristic (ROC) curve analysis and Youden index were performed to evaluate the cut-off point, accuracy, sensitivity and specificity of the predictive model. The tests were considered statistically significant when the p-value was less than 0.05.
Magnetic resonance imaging of a 3-year-old male patient with a tumor in the fourth ventricle confirmed as ependymoma post-surgery. A: Sagittal T1-weighted image. B: Axial T2-weighted image. C: Coronal fluid-attenuated inversion recovery image. D: Axial T1-weighted with contrast agent image.
Results
The study comprised 26 children (median age=7.5 years; male/female ratio=17/9), including 22 with medulloblastoma (median age=8 years; male/female ratio=13/9) and four with ependymoma (median age=3.5 years; all male). As shown in Table II, the ratio of FLAIR SI of ependymoma to parenchyma was higher than that of medulloblastoma. In contrast, the T1WCE SI was higher in the medulloblastoma group than in the ependymoma group (p<0.05).
Multivariate logistic regression analyses were performed to identify independent variables (Table III) that significantly and independently influenced the differential diagnosis of medulloblastoma. As shown in Table III, the results of the multivariate logistic regression test showed three statistically significant factors (p<0.05): T2 SI of the tumor, T1WCE SI of the tumor, and T1WCE SI ratio of the tumor to the parenchyma.
The GEE analysis was used to model the prediction of medulloblastoma with the three statistically significant predictors determined from multivariate logistic regression analyses:
where x1 was the T2W SI of tumor, x2 the T1WCE SI of tumor, and x3 the T1WCE SI ratio of tumor to parenchyma.
In the ROC analysis performed in this study, a cut-off linear predictor value of −2.22 (or a cut-off GEE model of 0.100) to predict the diagnosis of medulloblastoma yielded a sensitivity of 77.3%, a specificity of 100%, and an area under the ROC curve (AUC) of 92% (Figure 3).
Discussion
Medulloblastoma is the most common type of posterior fossa tumor in the pediatric population, followed by pilocytic astrocytoma and ependymoma. Medulloblastoma affects slightly older children compared to ependymoma. In our study, the median age was 8 and 3.5 years for medulloblastoma and ependymoma, respectively. Both of these tumors have male predominance. Our results are consistent with previous studies, which estimate the peak age of medulloblastoma before the child reaches the age of 7, while the peak age for ependymoma is between 3 and 5 years old (4, 14).
Basic magnetic resonance imaging protocols.
Comparison of quantitative basic magnetic resonance imaging sequences between medulloblastoma and ependymoma.
Ependymoma and medulloblastoma can be difficult to differentiate based on the imaging features alone as the features may overlap with one another. Therefore, correlation with other factors such as age, prevalence, and imaging-derived values may help in differentiating these tumors. Our study found that medulloblastoma had a higher T1WCE SI value compared to ependymoma. However, ependymoma showed a higher tumor to parenchyma FLAIR SI ratio. These differences may arise because medulloblastoma has a higher cellular density with a higher rate of cellular division, as described by Koeller et al. This is also why this tumor is regarded as the most malignant among posterior fossa brain tumors (15). The smaller extracellular extravascular space of highly dense tumors such as medulloblastoma reduces water content and restricts free water movement. This phenomenon leads to a low-to-moderate signal of medulloblastoma on T2W and FLAIR compared to ependymoma (4, 7, 15, 16). Since medulloblastoma is a highly malignant tumor with massive metabolic needs, it naturally has increased oxygen and nutrient consumption, which results in enhanced perfusion. Thus, the T1WCE SI of medulloblastoma is usually higher than that of tumors with lower malignant potential, such as ependymoma (4, 6, 14-16). Our results are also in concordance with other previous studies (4, 6, 15-18).
Receiver operating characteristic curve of the predictive model in predicting medulloblastoma. The sensitivity, specificity, and area under the curve were 77.3%, 100%, and 92%, respectively.
Multivariate logistic regression analyses performed to identify independent variables that significantly and independently influenced the diagnosis of medulloblastoma.
Several diagnostic models have been proposed over the years to help in differentiating medulloblastoma from ependymoma. Forbes et al. proposed a model based on the T2W SI ratio of tumor to parenchyma as an indirect indicator of abnormal water within tumors. Their study results also showed that all medulloblastomas were hypointense in T2W due to the high cellular density (12). In another study, Forbes et al. used a combination of DWI and T2W SI ratio of tumor to parenchyma to evaluate 50 children with cerebellar tumors. The results showed that the model had a 78% pre-surgical diagnostic accuracy (11). Porto et al. combined the morphology of the tumor and T2W and DWI SI to assess high-grade tumors at 95% and low-grade tumors at 70% accuracy (13). In 2016, Koob et al. used multi-parameter MRI in deriving differential diagnoses and predicting histology in childhood brain tumors. Their combination model included DWI and PWI and showed the highest accuracy of 73.24% in tumor classification (8). Overall, most predictive models have an accuracy ranging from 70% to 95%. In our study, the diagnostic model had a sensitivity, specificity, and AUC of 77.3%, 100%, and 92%, respectively. Thus, the initial results of our diagnostic model are very promising and also comparable with the results from previous studies (8, 11-13). Furthermore, the model introduced in this study only needs quantification of T2W SI, T1WCE SI, and the ratio of the tumor to parenchyma T1WCE SI. All these parameters can be derived from basic MRI sequences, and can easily be applied at most pediatric radiology centers around the world.
The small sample size and single-center study could be regarded as a limitation of this study. In addition, the number of ependymomas was relatively low; however, ependymoma has a low prevalence. We recommend that further studies with larger sample size and multicenter involvement be performed to validate our findings. Further studies of the combination of basic and advanced MRI sequences are essential for improved differential diagnosis among common pediatric posterior fossa brain tumors.
Conclusion
Our study proposes that T2- and T1WCE-ROI values of tumors and the ratio of T1WCE-ROI values to parenchyma derived from basic MRI sequences might be used to help in differentiating between pediatric medulloblastoma and ependymoma. The GEE predictive model might be used clinically to discriminate between medulloblastoma and ependymoma. Further research should be performed to validate these findings.
Acknowledgements
The Authors would like to express their gratitude to Dr. Mai Tan Lien Bang, Dr. Dang Do Thanh Can, and Mr. Nguyen Chanh Thi for their assistance and technical support in completing this research.
Footnotes
↵* These Authors contributed equally to this study.
Author's Contributions
Nguyen Minh Duc and Huynh Quang Huy contributed equally in preparing this article. Nguyen Minh Duc and Huynh Quang Huy made a substantial contribution to data acquisition, analysis, and interpretation. Each of the Authors played their part in preparing the article for drafting and revising it critically for important intellectual content. All Authors gave their final approval of the version to be published and agreed to be accountable for all aspects of the work, ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Conflicts of Interest
There are no conflicts of interest to declare.
- Received March 14, 2020.
- Revision received April 1, 2020.
- Accepted April 2, 2020.
- Copyright© 2020, International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved
