Guidelines for imaging retinoblastoma: imaging principles and MRI standardization

The human eye, with its superficial position and its fluid-filled structures, is ideally suited for US. Ocular US is usually performed by the ophthalmologist while the child is under general anesthesia, but can also quite easily be performed without sedation. In retinoblastoma, US demonstrates an irregular mass, more echogenic than the vitreous body, with fine calcifications (highly reflective foci mostly with characteristic acoustic shadowing) [41] (Fig. 2). Histologically, there is calcification present in approximately 95% of tumors [42]. Calcification is key to differentiating retinoblastoma from other mass lesions in a young child. US detects calcifications in 92–95% of cases where it is present histopathologically [42, 43]. Retinal detachment may also be observed, which is an important feature to define tumor growth pattern, either endophytic or exophytic, or a combination of both. Endophytic tumors arise from the inner layers of the retina and grow into the vitreous body. Frequently, small clusters of tumor cells detach from an endophytic mass, producing multiple floating tumor islands; this process is known as vitreous seeding. Exophytic tumors originate in the outer layers and grow in the subretinal space, which causes retinal detachment with subretinal exudate and possible subretinal tumor seeding. Tumors with exophytic growth more frequently have choroidal infiltration compared to endophytic tumors [44]. Diffusely infiltrating retinoblastoma is a rare histological form characterized by diffuse infiltration of the retina without a tumor mass [45]. Tumor height and diameter are usually measured at US, as these measurements are used for choice of treatment. Color Doppler can be useful for differentiating a vascularized tumor mass from echogenic effusions and for differentiation against developmental abnormalities such as persistent hyperplastic primary vitreous (PHPV; also known as persistent fetal vasculature, PFV), with the characteristic persisting hyaloid artery.

US is not the imaging modality of choice for direct evaluation of metastatic risk factors. Tumoral calcifications commonly obscure visualization of the optic nerve [41]. Indirect detection of optic nerve invasion by measurement of optic nerve diameter with a 3D-US technique has been reported in a single case report [46], hence its value remains unknown.

On CT, retinoblastoma is typically a mass of high density compared with the vitreous body, usually calcified and moderately enhancing after iodinated contrast medium administration. CT detection of calcifications in retinoblastoma has a sensitivity of 81–96%, and an even higher specificity [47]. However, delineation of intraocular soft-tissue detail is limited. The evidence from surveys suggests that CT is still regarded an obligatory imaging tool for evaluation of leukocoria, primarily because CT is supposed to be the best imaging modality for detection of intraocular calcifications [47‐49]. However, justification of the irradiation of a large group of retinoblastoma patients requires a base of evidence of the procedure’s clinical effectiveness and possibly also radiation-effectiveness [50] for supplying (1) valuable additional information leading toward the diagnosis of retinoblastoma and (2) valuable additional information, compared to non-ionizing radiation modalities in detection of tumor extent.

CT was the first imaging modality used to detect optic nerve invasion [51‐53] and is historically assumed to be precise in detection of tumor extent [54‐56]. However, this assumption is based on conflicting outdated literature, without thorough evidence by radiologic-pathologic correlation studies. The sensitivity of CT in detection of optic nerve invasion is actually very low, even in patients with extensive optic nerve invasion (length of invaded nerve segment > 2 mm) [21, 51, 53]. The specificity, accuracy and negative predictive value of CT remain artificially high because of the relatively low incidence of optic nerve invasion in normal-size nerves. An enlarged nerve due to massive tumor infiltration is rare in developed countries. Assuming retinoblastoma invasion into the optic nerve produces distortion of the anastomotic vascular network in the anterior optic nerve region, Jacquemin and Karcioglu [52] considered that non-visualization of the central retinal vessels is a reliable indicator of optic nerve invasion. However, these results were not confirmed by other studies [21].

Diagnostic MRI evaluation of a suspected retinoblastoma requires much more than performing a routine MR imaging examination of the orbit. High-resolution contrast-enhanced MRI is the technique of choice and should be used whenever possible to answer the key clinical questions (to evaluate an intraocular mass and to determine disease extent; Fig. 3). MRI has proved to be the most sensitive technique for evaluating retinoblastoma, especially regarding tumor infiltration of the optic nerve, extraocular extension and intracranial disease [21, 22, 44, 57, 58]. A major factor influencing the success of MRI is the use of appropriate hardware and optimized pulse sequences with appropriate spatial resolution for ocular MRI [21, 22, 44, 57‐61].

Examination under general anesthesia with fundoscopy and US almost inevitably leads to the diagnosis. As US detects foci of calcification in almost all retinoblastomas, there is now little benefit of routine CT for detection of calcifications in suspected retinoblastoma. Due to technical development, US and MRI are currently almost as accurate as CT for detection of calcifications. Recently, Galluzzi et al. [42] showed that when data from ophthalmoscopy, US and MRI are put together, no calcifications detected on CT were missed. A high-resolution gradient-echo T2-weighted sequence showed promising results regarding detection of calcifications and has been shown to be more effective than spin-echo techniques [42, 44].

The differential diagnosis of retinoblastoma includes several non-neoplastic lesions that also cause leukocoria. After retinoblastoma, which accounts for 47–58% of cases of leukocoria in children, other causes in decreasing order of frequency include PHPV, Coats disease, larval granulomatosis (Toxocara canis), retinopathy of prematurity, and retinal astrocytic hamartoma [48]. Calcification is the most important differentiating feature of retinoblastoma. However, when clinical diagnosis remains uncertain, US and MRI help characterize and differentiate intraocular abnormalities, especially when ophthalmological evaluation is limited due to opaque ocular refractive media, as may occur in all of these conditions. The role of CT in the detection of (sometimes subtle) characteristic findings is limited due to its low soft-tissue contrast [48, 55].

In the past, CT was used to determine tumor size, retro-orbital spread and intracranial growth; however, spread within the optic nerve past the cribriform lamina, and infiltration of choroid and sclera, which are important prognostic factors, are not reliably assessed with CT [51, 53]. Because of its superior soft-tissue contrast, MRI is more sensitive and specific than CT in detection of tumor extent and metastatic risk factors. MR imaging using high-resolution protocols is currently considered to be the most accurate and valuable tool in pretreatment staging of retinoblastoma, without known biological side effects.

Although individual examinations should always be tailored to the specific queries in individual patients (laterality, disease extent, therapy options), there are general recommendations for MRI in retinoblastoma. In the following paragraphs we discuss the minimum requirements for diagnostic evaluation of retinoblastoma or mimicking lesions according to the consensus reached among members of the European Retinoblastoma Imaging Collaboration (ERIC). If these recommendations cannot be followed because of technical limitation, ERIC members recommend to refer the patient to the nearest (or national) reference center for retinoblastoma, where a multidisciplinary team of specialized physicians (ophthalmology, pediatric oncology, radiology, pathology, radiotherapy, clinical genetics, psychology) and specialized nurses will ensure that practice conforms to the best standards of care.

Although the technical success of MRI usually depends on the cooperation of the patient, in retinoblastoma appropriate sedation techniques or general anesthesia are nowadays widely used, with a high yield of diagnostic scans. Nevertheless, in our experience the success rate of intravenous sedation is highly dependent on the presence of trained anesthetists and of the choice of radiofrequency coils. Especially, the use of small surface coils decreases the success rate, because these coils need to be accurately positioned close to the eye. The depth of sedation may be insufficient for accurate patient positioning. Therefore, general anesthesia is recommended for MRI in children with retinoblastoma. Another advantage of general anesthesia is the possibility to ensure that the eyelids are fully closed and to avoid uncontrolled eye movements by putting pads (fixed with tape) on the closed eyelids. Thereby susceptibility artifacts caused by air-tissue interface or by air bubbles under the eyelids is avoided, which is favorable for the image quality in the anterior eye segment. If possible, the MR examination can be combined with fundoscopy and US under the same general anesthetic.

Clinical MRI of orbit and eyes is mostly acquired using the current standard field strengths up to 1.5 T. Since adequate imaging of retinoblastoma requires high spatial resolution, the field strength of the MR system should be at least 1.5 T. The performance of scans at 1.5 T using a head coil is the most practical approach for evaluation of retinoblastoma. However, this combination usually gives insufficient signal-to-noise ratio (SNR) at required in-plane image resolution and section thickness. Therefore, scanning at 1.5 T should always be performed with one or two small surface coils (diameter ≤ 5 cm) to reliably detect small lesions and metastatic risk factors (Fig. 4). Indeed, the use of surface coils in ocular tumors has been reported to increase the diagnostic accuracy [44, 62].

The main advantage of higher field strengths (3 T) is the increased SNR. Publications on ocular 3-T MRI are still limited [63, 64]. At 3 T with multi-channel head coil or surface coils one can achieve high-resolution images similar to those obtained at 1.5 T with surface coils (Fig. 5). We think the diagnostic accuracy for detection of tumor extent might improve, but there is currently no published evidence for this.

MRI protocols vary because of differences in available equipment and individual preferences. However, despite differences, certain basic elements are common to most imaging protocols for retinoblastoma. The minimal requirements for diagnostic evaluation of retinoblastoma or mimicking lesions according to the consensus reached among members of the European Retinoblastoma Imaging Collaboration (ERIC) are presented below and summarized in Table 1. A typical MR imaging protocol for retinoblastoma should always include high-resolution imaging of the affected eye(s) and imaging of the entire brain.

Regardless of laterality, at least one transaxial thin-slice (≤ 2 mm) T2-weighted sequence should cover both orbits. For T2-weighted imaging based on a (fast) spin-echo technique, it is recommended to use a long TE (heavily T2-weighted; TE ≥ 120 ms) for generating the image contrast necessary to provide an optimal differentiation of retinoblastoma and surrounding vitreous or subretinal fluid (Fig. 6). Fat saturation combined with T2-weighted imaging is not recommended. When fat suppression is used, the resulting loss in SNR should be compensated (e.g., by increasing the number of acquisitions).

T2-W spin-echo may be replaced by gradient-echo T2-weighted sequences such as 3D steady-state free precession sequences with slice thickness ≤ 1 mm (Fig. 6; vendor-specific acronyms: CISS [Constructive Interference in Steady State, Siemens], FIESTA [Fast Imaging Employing Steady State Acquisition, GE Healthcare]; DRIVE [Driven Equilibrium, Philips]) (slice thickness, ≤ 1 mm). These pulse sequences provide detailed images of both orbits and eyes, and allow accurate comparison of eye size, anterior chamber depth and laterality. Very small tumors can be depicted with these techniques.

Increased spatial resolution will improve the accuracy of MRI in assessing the anatomical details of the papilla, lamina cribrosa and pre- and postlaminar segments of the optic nerve [65]. The continuous improvement of MR units and the use of small fields-of-view with either multi-channel head coils or surface coils now allows much higher image resolution. High spatial resolution means section thickness ≤ 2 mm and in-plane pixel size ≤ 0.5 × 0.5 mm. For optimal detection of optic nerve invasion, the image plane through the orbit (transaxial and sagittal oblique) should align with the orientation of the distal (1 cm) end of the nerve, just posterior to the lamina cribrosa (Fig. 7). One section in each of these sequences should be precisely aligned within the distal part of the optic nerve at the level of the middle of the optic disk. Although the use of the fat-saturation technique is highly recommended for contrast-enhanced MR imaging in orbital pathology, its use in high-resolution contrast-enhanced T1-weighted MRI in retinoblastoma is declining [44, 58]. In the minimal requirements for diagnostic evaluation of retinoblastoma or mimicking lesions according to the consensus reached among members of the ERIC, the use of fat saturation in contrast-enhanced T1-weighted sequences is no longer recommended.

The recommended protocol for high-resolution MR imaging of the eye(s) and distal optic nerve(s) differs slightly between unilateral or bilateral disease (Table 1, Figs. 8–9). Incidence of metastatic risk factors is highly dependent on tumor location and tumor size. Therefore, a distinction is made between bilateral disease with only one eye strongly affected (high-resolution MRI can be performed in the worst affected eye only) and extensive disease in both eyes (high-resolution MRI of both eyes).

The brain should always be imaged in retinoblastoma patients for analysis of midline structures in order to depict trilateral retinoblastoma (i.e. PNET located mainly in the pineal gland, or more rarely in suprasellar area) or leptomeningeal spread. For a patient presenting with leukocoria suspicious of having retinoblastoma or already diagnosed as retinoblastoma based on clinical findings and US, the baseline evaluation should include an MR imaging of the brain that meets the standardized protocol. Imaging of the brain is performed with (multi-channel) headcoil only and should at least include the following or similar types of sequences:

Compared to the vitreous body, retinoblastoma has moderately higher signal intensity on T1- and lower on T2-weighted images. Increased size of the globe, globe deformation and reduced anterior chamber depth are signs of increased intraocular pressure and are usually associated with buphtalmia (Fig. 10). These signs should be mentioned since they are associated with a higher risk of globe rupture and secondary orbital seeding during enucleation. Laterality and growth pattern should be mentioned as well as the location of the tumor, with respect to the equator of the eye (anterior, posterior or combined) and with respect to the papilla (the optic nerve disk) and the macula. One should in particular identify tumor close to the optic disk, because this may invade the nerve [22] (Fig. 11).

In normal-size optic nerves, the direct radiological criterion used to diagnose postlaminar nerve invasion is the presence of abnormal contrast enhancement (enhancement ≥ 2 mm in diameter) in the distal nerve [21] (Fig. 12). Interruption of the normal linear enhancement at the optic nerve disk (choroidoretinal complex) supports a suggestion of optic nerve invasion [22]. Postlaminar optic nerve or optic nerve menigeal sheath invasion should raise suspicion of leptomeningeal metastases. In such situations, additional contrast-enhanced sagittal T1-weighted imaging of the whole spine is recommended.

Discontinuity of the normal choroidal enhancement is the leading criterion for infiltration [22, 24] (Fig. 13). Massive choroidal invasion presents as focal choroidal thickening (Fig. 13). Increased enhancement and thickening of the entire uveal tract (choroid, ciliary body, iris) is a sign of uveitis, usually secondary to massive (sub) total tumor necrosis [22]. Protrusion of enhancing tissue through the thickened choroid into the (low signal-intensity) sclera or beyond is a sign of scleral invasion or extraocular extension, respectively.

Anterior eye segment enhancement occurs frequently in retinoblastoma and is usually a sign of iris angiogenesis [27, 30]. Tumor invasion into the anterior eye segment (Fig. 11) is an infrequent finding, usually associated with anteriorly located retinoblastoma. Enhancement of the tumor extending into the ciliary body or beyond should raise suspicion of anterior eye segment invasion. Transaxial T2-weighted images of both orbits can be used to depict a decreased anterior chamber depth (Fig. 10).

Careful analysis of midline structures should be performed to depict trilateral retinoblastoma (i.e. PNET located mainly in the pineal gland, or rarely in the suprasellar area) (Fig. 14) or leptomeningeal spread (if patient shows extensive postlaminar optic nerve invasion enhancement). Congenital brain malformations occur mainly in patients with 13q- deletion syndrome [10]. Benign pineal cysts should not be misinterpreted as pineal PNET, even in children with retinoblastoma [10, 66]. Thin-section T2-weighted and contrast-enhanced T1-weighted slices are helpful for differential diagnoses. In doubtful cases, close follow-up with MRI is recommended.