Detection of recurrent brain tumors in children: No significant difference in sensitivity between unenhanced and contrast-enhanced MRI

Abstract

Background

Magnetic resonance imaging (MRI) with a gadolinium injection is currently used in the follow-up of children in remission of cerebral tumors (CTs). Intracerebral gadolinium deposition has been recently reported with unknown risks. The aim of this study was to evaluate the sensitivity of unenhanced brain MRI (U-MRI) in detection of tumor recurrence.

Methods and materials

A set of 58 U-MRIs of children in remission was retrospectively evaluated by three seniors (a neuroradiologist, a pediatric and a general radiologist) and one junior to look for any recurrence. Clinical, tumoral and imaging data were collected. The final diagnosis was anatomopathological when available, or the clinicoradiological evolution. Sensitivity, specificity, predictive values and interobserver agreement were calculated. A Fisher test and Fleiss kappa coefficient were performed.

Results

For the seniors, the U-MRI had a sensitivity of 81% (95% confidence interval (CI): 0.56–0.90), and a negative predictive value (NPV) of 82% (95% CI: 0.63–0.94). The U-MRI sensitivity, regardless of the observer, was not significantly different from the contrast-enhanced MRI sensitivity (86%) according to a Fisher test (p > 0.05). No significant difference in sensitivity within the subgroups was found. The interobserver agreement of seniors was good (κ = 0.68).

Conclusion

U-MRI brain was suboptimal for 80% of patients. Three-dimensional millimetric, fluid-attenuated inversion recovery, and diffusion would constitute helpful sequences in follow-up. Further specific studies depending on each tumor type are still required to determine whether a potential abstention of gadolinium intravenous injection should be discussed for children.

Introduction

After hematological diseases, intracranial tumors (CTs) are the most common neoplasia in children, accounting for approximately 20% of all pediatric malignancies. Brain tumors remain the main causes of cancer death^1,2 in children and represent a very heterogeneous group of diseases, in which primary brain tumors are almost only the ones observed with a majority of infratentorial space.^3,4

Primary neuroepithelial tumors are the most frequent (80%), and among them glial lesions (30% to 50%) and in particular pilocytic astrocytoma.³ Magnetic resonance imaging (MRI) is the best imaging modality in the diagnosis and monitoring of these lesions because of the high resolution of images and its safety advantages over other modalities. MRI is frequently associated with the injection of gadolinium-based contrast agent (GBCA), which has been used worldwide since 1988.⁵ However, current clinical data show gadolinium accumulates in tissues, including the liver, kidneys, muscles, skin and bones.⁶

In 2013, an association between the injection of GBCA and cerebral MRI abnormalities such as deposit was reported for the first time. A positive correlation was found between the T1-signal intensity increase in certain areas of the brain (nucleus dentatus and globus pallidus) and the history of exposure to GBCA, even years after administration of GBCA,⁷ and even in individuals without severe renal dysfunction.⁸ This was first demonstrated for linear-type chelates.⁹ Then in 2017, the same fact was observed with the macrocyclic-type of GBCA, but to a lesser extent.^10,11 Children show a similar, although later, pattern that is accelerated by radiochemotherapy.^12,13 No clinical consequences have been reported to date, but it is important to take into account the potential unknown risks of residual gadolinium in our decisions regarding the injection of GBCA. Patients are exposed to morbidity directly related to neoplasia, and to iatrogeny by irradiation radiotherapy, which continues to be a key element of therapeutic management. Young children also have numerous exposures to radiation with the diagnostic radiological examinations in the context of long-term surveillance.

In the case of brain tumors, GBCA injections are recommended in the diagnostic phase,¹⁴ and they are used in the follow-up of patients in remission. Regarding imaging protocol, there is no clear consensus, but sequence recommendations have been formulated by the Response Assessment in Pediatric Neuro-Oncology Committee for MRI monitoring under treatment, extrapolated to remission surveillance, and to standardize research practices for future recommendations.^15,16 In addition, the diagnostic contribution of GBCA injection is low in children with a normal precontrast brain MRI.¹⁷

Moreover, there are two other consequences of injecting GBCA: the need for a venous access, which is sometimes difficult in pediatric practice, and the need for longer examination time, with the difficulty of immobilization and sometimes the necessity for sedation.¹⁸

The aim of the study was to evaluate the brain tumor recurrence detection sensitivity of unenhanced brain MRI (U-MRI) by different observers.

Materials and methods

Population

We conducted a single-center retrospective study over a 10-year period, between 2007 and 2017, in a university hospital center. A sample of 43 children was created including 58 follow-up examinations with 29 cases of confirmed recurrence and 29 cases of continued remission.

The inclusion criteria were an initial diagnosis of a cerebral malignant tumor before the age of 18 years, diagnosed by imaging, and histologically confirmed.

Remission was defined by the oncology teams from this same center by a macroscopically complete surgical excision during initial surgery or a surgical revision (after initial surgery or after first chemotherapy) associated with the absence of residual tumor visualized at 48 hours postoperative MRI and confirmed by three consecutive MRIs. Children in remission were randomly chosen among our patients to constitute a comparable control population.

An intracranial relapse observed during radioclinical follow-up was defined by a histologically confirmed (or by the death of the patient) of a new intracranial lesion intra- or extraparenchymally including leptomeningeally.

Primary central nervous system (CNS) lymphoma was excluded, as well as brain tumors falling within the framework of a neurofibromatosis-like genetic disorder or intracranial metastases from extracranial primary lesions. Imaging at diagnosis and during therapy was excluded. The absence of gross total resection and the clinical suspicion of recurrence were exclusion criteria.

Stable residual tumors followed by regular imaging were excluded. Patients with malignant CNS tumors involving only the spinal cord at diagnosis were excluded.

A visible relapse on MRI was defined as the detection of a new intracranial lesion intra- or extraparenchymally including leptomeningeally, in comparison with prior MRI.

Clinical data

Clinical characteristics (age at diagnosis, survival), type of tumor and grade according to World Health Organization (WHO) classification, location in the posterior fossa, metastatic character from the outset, and initial enhancement at time of diagnosis were collected, as were characteristics of the recurrence (time to relapse, location).

MRI protocol

MRI studies were performed on a 1.5 Tesla system (General Electric Signa then Siemens Aera, GmbH, Erlangen, Germany). The MRI protocols always included precontrast imaging, with conventional multislices turbo spin echo (TSE) T1- and T2-weighted images, then three-dimensional (3D) from 2014. Each MRI had a fluid-attenuated inversion recovery (FLAIR) sequence. Nine studies contained additional diffusion sequences. The protocols systematically included TSE T1-weighted images after gadoterate meglumine (Dotarem, Guerbet, Roissy-Charles de Gaulle, France) intravenous administration.

Confirmation of recurrence

The reality of the recurrence was histologically affirmed or invalidated, or failing that, by the clinicoradiological evolution observed afterward (e.g. regression of the lesion without treatment, unfavorable evolution, death).

Performances of initial contrast-enhanced MRI (CE-MRI)

Results in the radiological reports at the time of follow-up CE-MRI were reported and compared with the histological confirmation to evaluate the diagnostic performance of the CE-MRI in this series.

Interpretation of U-MRI brain

A retrospective reading of the 58 MRI examinations was performed, which included randomly 29 cases of recurrence and 29 cases of continued remission, by three senior radiologists (a pediatric radiologist with five years’ experience, a neuroradiologist and a general radiologist) and a radiology resident. None of the radiologists were involved in the original interpretation of the MRI scans. All were blinded to the outcome of the patients. Radiologists had to rule on a possible cancer recurrence by concluding “continuation of remission” or “cancer recurrence or suspicion on a recurrence deserving further investigation.”

Only unenhanced MRI sequences, and prior MRIs, were available for analysis.

Statistical analysis

A study of the comparability of populations in recurrence and in continuation of remission was performed using the Wilcoxon, Mann-Whitney and chi-squared tests.

Interobserver agreement was achieved using the kappa test. A kappa coefficient < 0.4 indicated low agreement, in the range 0.4–0.6 average agreement, 0.6–0.8 good agreement, and > 0.8 excellent agreement.

The diagnostic features of U-MRI in the CTs recurrence detection were described by their sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV), with confidence interval (CI) 95%. A Fisher test was performed between the U-MRI and CE-MRI.

Results

Population characteristics

The detailed characteristics of patients with recurrence and patients in continuation of remission are described in Table 1. The two groups were not significantly different except for age at time of diagnosis (p = 0.01). The cancer recurrence population consisted of 14 (61%) boys and nine (39%) girls, and the mean age at diagnosis was 41 months (3 years and 4 months). Seventy-nine percent of the cases involved posterior fossa tumors. Three of the 29 cases of recurrence (9%) involved patients who had secondary lesions from the onset of diagnosis. The average follow-up time for these patients, from the initial diagnosis, was 49.5 months, (median: 34 months, extremities: 3 months, 202 months). Nineteen of the 29 cases of recurrence were anatomopathologically confirmed and 10 confirmed by clinicoradiological evolution, eight of which were by death.

Table 1.

Population characteristics.

	Cancer recurrence population	Remission population
	N = 23 patients, 29 cases	N = 20 patients, 29 cases
Age at time of diagnosis (months)
Average	40.5	62.4
Median	31	48
Range	(4–140)	(7–144)
Sex
Male	14	13
Female	9	7
Duration of follow-up from initial diagnosis (months)
Average	48	58.7
Median	30	43
Range	(3–202)	(12–202)
Initial site in posterior fossa	23	19
Radiotherapy	20	11
Chemotherapy	16	15
Metastatic from the start	3	1
Time to relapse, from remission (months)
Average	25.7
Median	9	10
Range	(1–192)	(2–168)
Age
<6 months	9	8
6 months to 1 year	8	9
1 year to 1 year and a half	2	3
1 year and a half to 2 years	4	4
2 years to 5 years	3	3
>5 years	3	2
Histology of initial tumor
Ependymoma	1	3
Anaplastic ependymoma	10	10
Medulloblastoma	4	4
ATRT	5	4
Unclassifiable embryonic tumor (old PNET)	2	1
Pilocytic astrocytoma	4	3
Pleomorph xanthoastrocytoma	1	1
High-grade glioma	2	1
Choroid plexus carcinoma	0	2
Grade of lesions
I	4	4
II	2	4
III	10	10
IV	13	11

ATRT: atypical teratoid rhabdoid tumor; PNET: primitive neuroectodermal tumor.

MRI studies

There was an enhancement of the initial lesion in 23 of the 29 recurrence cases.

Recurrence tumor characteristics

Histological type and grade

Patients with anaplastic ependymoma accounted for the largest population in the cohort with 10 patients (34%), followed by atypical teratoid rhabdoid tumors (ATRTs) with five patients (17%), and medulloblastomas with four patients (14%).

Grade IV according to the WHO classification was associated with the highest rate of cancer recurrence (45%).

Of the 23 patients, 17 (79%) had a single recurrence and six (21%) had multiple recurrences. Among the six cases of multiple recurrences, three involved anaplastic ependymomas, two medulloblastomas, and one ATRT.

Site

Of the 29 recurrences, 69% (n = 20) involved the initial site in the postsurgical cavity or intraparenchymal in the margin of the cavity, 17% (n = 5) involved distant sites, and 14% (n = 4) combined an initial and distant site. Among the distant recurrences, three were leptomeningeal (including two combined). For two of them the location was sellar or suprasellar, one of which was pituitary (anatomopathologically proved), and in the other a potential secondary lesion to the primary lesion, with reported hypothesis such as a radio-induced cystic meningiomatous lesion or a craniopharyngioma (no pathology today).

Time to recurrence

The average time to recurrence between remission and recurrence was 25.7 months, ranging from one month to 192 months. Recurrences usually occurred in the first year (58%), and almost 80% in the first two years.

Enhancement

Two high-grade gliomas, nine anaplastic ependymomas, five ATRTs, three medulloblastomas, four pilocytic astrocytomas, one pleomorphic xantoastroxytoma and one primitive neuroectodermal tumor (PNET) were enhanced on the initial MRI. One ependymoma and one PNET were not enhanced on the initial MRI.

Twenty-four of the 29 cases of recurrence were characterized by an enhancement (83%).

Results of U-MRI

Sensitivity, specificity, PPV and NPV of senior readers were respectively 81% (95% CI: 0.56–0.90), 89% (95% CI: 0.60–0.92), 88% (95% CI: 0.61–0.92) and 82% (95% CI: 0.63–0.94).

The interobserver agreement of the U-MRI method was good among the resident radiologist, the general radiologist, the pediatric radiologist and the neuroradiologist seniors, with a Fleiss kappa coefficient of 0.71 (95% CI: 0.57–0.82).

There were five to seven false negatives, and one to six false positives, depending on the observer. All observers correctly diagnosed 22 to 24 recurrences out of 29 by U-MRI, a sensitivity ranging from 76% to 83%. Data per reader are detailed in Table 2.

Table 2.

Sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV), per reader, and overall.

	Se	Sp	PPV	NPV
Reader 1 (general radiologist)	76%	97%	96%	80%
95% CI	0.56–0.90	0.82–0.99	0.78–0.99	0.63–0.92
Reader 2 (pediatric radiologist)	83%	90%	89%	84%
95% CI	0.64–0.94	0.73–0.98	0.71–0.98	0.66–0.95
Reader 3 (neuroradiologist)	83%	79%	80%	82%
95% CI	0.64–0.94	0.60–0.92	0.61–0.92	0.63–0.94
Reader 4 (resident)	76%	91%	85%	78%
95% CI	0.56–0.90	0.68–0.96	0.65–0.96	0.60–0.91
Overall	79.25%	89.25%	87.50%	81%
95% CI	0.56–0.90	0.60–0.92	0.61–0.92	0.63–0.94
Senior overall	81%	89%	88%	82%
95% CI	0.56–0.90	0.60–0.92	0.61–0.92	0.63–0.94

CI: confidence interval.

The best sensitivity and negative predictive values were obtained by the pediatric radiologist (reader 2) and neuroradiologist seniors (reader 3). The generalist (reader 1) had the best specificity and positive predictive value. The resident (reader 4) also had better specificity and positive predictive values. An example of a true positive case is reported in Figure 1 with the detection of recurrence by the U-MRI diagnostic method.

Figure 1.

Magnetic resonance images of the follow-up of a child in remission of an anaplastic ependymoma: onset of a lesion at the margin of the excision cavity, visible on (a) postcontrast T1-weighted sequences and (b) fluid-attenuated inversion recovery (FLAIR) images. Reference images: (c) postcontrast T1-weighted sequences and (d) FLAIR.

Recurrent tumor characteristics

Multiple recurrences were present in six patients: Five patients had two recurrences, and one patient had three recurrences.

The recurrences of the two initially unenhanced tumors (PNET and ependymoma) were also nonenhanced. The recurrences of the following initial enhanced tumors (one ATRT, one pilocytic astrocytoma and one anaplastic ependymoma) were not enhanced.

False positives

Specificity and PPV were good in U-MRI. In case of the positivity of the U-MRI, or of diagnostic doubt, an injection imaging would be carried out afterward, and the specificity would then join that of the diagnostic reference method.

In one to three cases per reader, the false positives corresponded to the three false positives of the CE-MRI method, in connection with postradiation appearances.

False negatives

There were five to seven cases of false negatives per reader, three of which were common to the four observers. One of the false negatives common to all observers concerned a recurrence of medulloblastoma in the form of a nodular lesion in the postsurgical cavity, not visible on the T2-weighted sequence because of thick slices not passing through the lesion, and visible with difficulty on FLAIR sequences by its millimetric character (Figure 2).

Figure 2.

Magnetic resonance imaging follow-up of a child in remission of a medulloblastoma: onset of a contrast enhancement on (a) postcontrast T1-weighted sequence in front of a millimeter nodular lesion visible on (b) fluid-attenuated inversion recovery (FLAIR), not visible in T2-weighted (c) because of thick slices not passing through the lesion. Reference image: (d) FLAIR.

Another false negative of the four observers concerned an intraparenchymal lesion at the margin of the excision cavity in a patient in remission of an ATRT, which is detectable only by contrast enhancement on the postcontrast sequences. It was already visible on postoperative imaging, and was diagnosed in connection with a posttherapeutic appearance, gradually increased in size on the following different examinations, without a nodular lesion (Figure 3). The injection of GBCA constituted a real diagnostic contribution here.

Figure 3.

Magnetic resonance imaging follow-up of a child in remission of an atypical teratoid rhabdoid tumor. Increase of intraparenchymal contrast enhancement at the cavity margin: (a) postcontrast T1-weighted sequences and (b) fluid-attenuated inversion recovery (FLAIR). Reference images: (c) postcontrast T1-weighted sequences and (d) FLAIR.

The third false negative common to all observers concerned the onset of a suprasellar lesion in a patient in remission for 144 months of a vermian medulloblastoma. This lesion was visible on all U-MRI sequences, but with difficulty because the 3D FLAIR sequences available for the analysis were not perfectly comparable with the prior two-dimensional FLAIR sequences.

Subgroup analysis

Within the subgroups of histology and grade of tumors, delay and site of recurrences, year of the MR examination, enhanced and nonenhanced recurrence tumors, no significant difference in sensitivity was demonstrated.

Precontrast imaging sensitivity in detection of lesions with enhancement (Se 80%, 95% CI: 0.58–0.93) had no significant difference compared with the detection of lesions without enhancement (Se 75%, 95% CI: 0.05–0.85).

Performances of CE-MRI at time of initial examination

The contrast-enhanced sequences were interpreted by experienced pediatric radiologists with more than five years’ experience, different from the observers of the present study, at the time the MRI were performed. The retrospective analysis of the reports made at the time of the CE-MRI found four cases of false negatives and three cases of false positives compared with the gold standard. The CE-MRI current method in our sample of 29 cases had a sensitivity of 86% (95% CI: 0.68–0.96), a specificity of 90% (95% CI 0.73–0.98), a PPV of 90% (95% CI 0.72–0.98) and an NPV of 87% (95% CI 0.69–0.96).

The sensitivity of U-MRI (79.5%) regardless of the observer is not significantly different from the sensitivity of the CE-MRI (86%) according to a Fisher independence test (p > 0.05).

Discussion

The follow-up imaging aim is the early detection of tumor recurrence, which could allow rapid intervention and a survival rate increase. In our study, the results of seniors in detecting brain cancer recurrence by U-MRI reported an overall frequency of 81% and an 82% NPV. The detection was homogeneous with a good interobserver agreement. The best results were obtained by the seniors, particularly by the specialists. The poorer performance of the resident observer could be explained by the fact that he was still in training and was less experienced.

Few studies on pediatric neuroimaging follow-up have been conducted,^19,20 and most have their assessment limited to medulloblastomas.²¹ To our knowledge, there has been no study concerning the sensitivity of U-MRI examinations in pediatric brain cancer recurrence detection.

Concerning the false negatives, some errors seemed to be related to the absence of a 3D sequence available in the set of precontrast images, in particular in a case in which the recurrence was not enhanced and was visible only by its nodularity. This technical difference may have influenced the interpretation of the FLAIR and TSE T1-weighted sequences, initially in axial slices and secondarily in millimetric 3D. Another possible explanation for errors would be a comparison only with the previous examination. The availability of older examinations would permit better detection of a very slowly evolving nodular lesion.

There have been few studies analyzing the contrast enhancement during recurrences, except for embryonic tumors²² and in perfusion with a pharmacodynamic parameters study.²³ It could be assumed that the recurrence will have a similar enhancement pattern like the primary lesion. In our series, there were as many enhanced initial lesions as there were enhanced recurrence lesions, but this could be partially explained by a selection bias. Nearly half of the unenhanced recurrence lesions had a primary lesion that did not initially enhance.

Whereas the majority of primary embryonic tumors have shown enhancement, the degree of enhancement may be variable, and sometimes absent, with, for example, 11% to 25% of medulloblastomas that do not enhance^24,25 and are variable depending on the molecular subgroup²⁶ (absence in the subset of group IV²⁷), or some ATRTs. On the other hand, previous studies have shown that medulloblastoma metastases tend to have a lower enhancement than their initial primary lesions,²⁸ and cancer recurrence was totally unenhanced in 36% of cases,²⁴ with up to 20% of secondary leptomeningeal diseases missed by conventional MRI but detected by cerebrospinal fluid cytology.²⁹

This may support the hypothesis of the abstention from GBCA injection during the follow-up of embryonic tumors, which has been developed in a study aiming to reinforce the role of diffusion-weighted imaging for this purpose.²⁵ However, the embryonic tumors that belong to WHO grade IV are more at risk of recurrence, with worse prognosis, and thus require increased follow-up.

Another issue of enhancement was the specificity of the enhancement. A variable evolution over time of the enhancement of noncerebellar pilocytic astrocytomas was reported by Gaudino et al.³⁰ without any tumoral or residual tumoral dimensional modification. It suggests that isolated enhancement fluctuations should not be considered an indicator of tumor progression or regression.³⁰

Embryonic tumors have specific recurrence patterns, with leptomeningeal disease of 38% in medulloblastoma, of 15% in ATRT³¹ and cerebromedullary diffuse leptomeningeal disease in pineoblastoma.²² In our series, in agreement with the literature, the rate of recurrence at the initial site was lower for embryonic tumors (eight of the 11 cases, i.e. 72% and only 50% of medulloblastomas). It would be interesting to integrate this particularity in choosing whether to give a GBCA injection. Indeed, studies have reported superiority of enhanced FLAIR sequences’ sensitivity for the diagnosis of leptomeningeal disease vs precontrast FLAIR sequences.^32,33

It would seem appropriate to integrate in the choice of a GBCA injection the different known recurrence risk factors. In our study, the common characteristics of recurrent patients were mainly the histological type (anaplastic ependymoma, embryonic tumors), the WHO grade (III and IV), in agreement with the risk factors reported in the literature, which the histological type (high-grade glial lesions), high grades, incomplete surgical resection, brainstem localization and age younger than 1 year at diagnosis.^34–36 Tumor grade appeared to be the most important prognostic factor, except for the youngest age group, in which complete resection surgery was the most important criteria.³⁶ However, the benefit of a macrocyclic gadolinium injection outweighs the potential risk of gadolinium deposition in children with a brain tumor.

The new WHO classification has established a new type of definition, obtained by a “layered” combination of histological type, WHO grade and molecular information. New genetic subtypes have recently been identified and others are coming up that will have different survivals and will have to be integrated into the decision-making process.³¹

In our study the average time to recurrence occurred in 58% in the first year after the remission date, and for the majority of them (around 80%) in the first two years, in agreement with the literature.³⁶

Our results should be interpreted with the awareness of certain limitations. The low number of patients, the heterogeneity of the MR protocol and the heterogeneity of primary lesions due to the retrospective design of the study represent important limitations that may have affected the results. The recurrence rate of CTs in our center was quite low (29 cases over a 10-year period). We will therefore have to keep in mind an eventual failure of an early detection of a leptomeningeal recurrence on unenhanced brain MRI, especially in the case of a homogeneous thickening, without any nodular lesion that would alert the radiologist. A prospective study is difficult to conduct in this context, but in the future, a multi-center study with a homogeneous MRI protocol and blind reading could be more powerful.

Conclusion

Unenhanced brain MRI enabled a suboptimal diagnostic sensitivity in detecting CT recurrence in children without any significant difference compared with enhanced cerebral MRI, requiring 3D millimetric, FLAIR and diffusion that would constitute three helpful sequences. Further specific studies depending on each tumor type are still required to determine whether a potential abstention of gadolinium intravenous injection could be discussed for patients without any risk factors for recurrence, on condition of having reproducible and complete follow-up examinations.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflict of Interest

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.