Sensitivity of Magnetic Resonance Imaging in Detection of Choroidal Metastases Open Access

Once considered a rare entity, choroidal metastasis (CM) is now recognized as the most common intraocular malignancy, with an incidence exceeding that of choroidal melanoma [1‒4]. Indeed, there are an estimated 20,000 patients per year who develop CM in the USA alone, while approximately 3,500 develop choroidal melanoma [1, 2]. Furthermore, most of these patients with CM are never evaluated by an ophthalmologist [1, 3]. It has been proposed that perhaps patients have no visual symptoms to prompt further investigation, although recent epidemiologic studies show that asymptomatic patients manifest CM at an exceedingly low rate between 0 and 2% [4, 5]. It is also possible that visually symptomatic patients and their physicians perceive systemic oncologic problems as being of greater severity and urgency, or that they or their physicians may not attribute visual symptoms to potential CM. Another consideration may be an overdependence on neuroimaging, particularly magnetic resonance imaging (MRI), by non-ophthalmologic specialists to exclude the presence of CM.

MRI plays an essential role in the imaging surveillance of patients with systemic malignancies, and brain MRI is the cornerstone in assessing for brain metastases [6]. Despite this, there exists a paucity of research regarding the sensitivity of brain MRI in detection of CM from systemic primary cancers. Herein, we examine a cohort of patients with clinically confirmed CM and evaluate the CM detection rate by MRI.

A retrospective chart review of all patients seen on the Ocular Oncology Service at Byers Eye Institute (Stanford University, Palo Alto, CA, USA) between January 1, 2018 and March 1, 2022, was performed to identify patients clinically diagnosed with choroidal metastasis (CM) by an ocular oncologist. The study was performed in compliance with the Health Insurance Portability and Accountability Act and Declaration of Helsinki. The Institutional Review Board approval was obtained from Stanford University Hospital.

All included patients underwent comprehensive ophthalmic evaluation and had a preceding or subsequent MRI scan of the brain and/or orbits within 3 months of CM diagnosis. MRI scans were performed at Stanford University Hospital (SUH) with a field strength of 3.0T and typically included pre-contrast axial T₁-weighted, T₂-weighted, and multiplanar post-contrast T₁-weighted scans; fluid-attenuated inversion recovery (FLAIR) imaging was performed post-contrast and with fat suppression. MRI scans were formally interpreted by board-certified neuroradiologists. When multiple MRI scans were performed, the one closest to the date of CM diagnosis was used. In all cases, imaging was obtained following injection of intravenous gadolinium-based contrast. The majority of scans were ordered by the primary oncology team to assess for central nervous system (CNS) spread of disease and consisted mostly of brain protocol scans.

Evaluation of MRI detection of CM was divided into two parts. For the initial part (“standard read”), the presence or absence of CM by MRI was determined solely on the basis of the original radiology report signed by a board-certified neuroradiologist. The second part (“dedicated read”) included the subset of patients for whom all acquisition sequences were performed (i.e., T1-weighted, T2-weighted, post-gadolinium brain volume, 3D FLAIR, and diffusion-weighted imaging [DWI]). For the dedicated read, a board-certified neuroradiologist (N.F.) reevaluated each acquisition sequence, paying particular attention to the radiographic appearance of the globes. This neuroradiologist was blinded to the location and laterality of CM and tasked with providing a “gold standard” interpretation objectively assessing whether the presence or absence of CM could be determined from the MRI exam. A CM was considered detected on the dedicated read if the neuroradiologist could correlate the CM lesion on MR with the lesion identified on clinical ophthalmic exam.

For both standard read and dedicated read parts, the sensitivity of MRI detection of CM was evaluated (Tables 1 and 3, respectively). For both parts, additional collected study data, as depicted in Tables 2 and 4, respectively, included a combination of patient demographics, clinical features of CM, the presence and nature of visual symptoms (i.e., blurry vision, metamorphopsia, flashes,floaters, and/or shadow sensation), clinical and treatment features of the primary systemic malignancy, and patient survival outcomes. Additional statistical analyses were performed for each study feature based on MRI detection versus non-detection groups using Student’s t test, χ² test, and Fisher’s exact test.

This study evaluated CM from 42 eyes of 40 patients. The standard read portion included all 42 eyes of 40 patients (Tables 1, 2), while the dedicated read portion included a subset of 33 eyes of 32 patients who had complete pre- and post-gadolinium MRI scans available (Tables 3, 4). Overall, mean patient age at CM diagnosis was 58 years (median 61, range 33–84 years). Primary cancer diagnosis preceded CM diagnosis in 38 patients by a mean interval of 4 years (median 2, range 0–22 years), while in 2 patients, a diagnosis of CM resulted in subsequent diagnosis of systemic primary malignancy. Most patients were Caucasian (60%) and female (70%), the latter likely reflecting a preponderance of breast metastasis to the choroid. Breast cancer accounted for 48% of the primary systemic malignancies in this study, followed by lung (35%), gastrointestinal (8%), thyroid (5%), and other (4%). CM were predominantly macular (60%), with a mean largest basal diameter of 10.4 mm and ultrasonographic thickness of 2.6 mm. Visual symptoms were present in 35 of 42 (83%; 68–93% 95% CI) eyes.

Based on the standard read, MRI detection of CM occurred in 21 of 42 eyes (50%; 34–66% 95% CI) with clinically confirmed CM (Table 1). There was no statistical difference in CM detection rate between MRI brain and orbit protocols (p = 0.249), likely due to considerable sequence overlap between these two protocols. Comparison of MRI detection versus non-detection groups (Table 2) revealed improved detection with increased tumor basal diameter (12.7 vs. 8.9 mm, respectively; p < 0.001) and ultrasonographic tumor thickness (3.8 vs. 2.1 mm, respectively; p = 0.003). These features were associated with increased time interval from cancer diagnosis to CM diagnosis (5.95 vs. 2.76 years, respectively; p = 0.009) and worse staging of primary cancer (p = 0.018).

Dedicated read was performed on 33 of the original 42 eyes for which all acquisition sequences were performed (i.e., T1-weighted, T2-weighted, post-gadolinium brain volume, 3D FLAIR, and DWI). Of these 33 eyes, the standard read detected 19 of 33 CM (58%). With dedicated read, MRI detection of CM improved to 26 of 33 eyes (76%) (Table 3). Comparison of MRI detection versus non-detection groups (Table 4) again revealed improved detection with increased tumor basal diameter (11.2 vs. 6.4 mm, respectively; p < 0.001) and ultrasonographic tumor thickness (3.0 vs. 1.3 mm, respectively; p < 0.001). The size threshold for detection was notably decreased on dedicated read. Dome-shaped lesion configuration was associated with improved detection compared to flat configuration (76% vs. 28%, p < 0.001), as was the presence of retinal detachment compared to the absence of retinal detachment (58% vs. 0%, p < 0.001).

On dedicated read, the 26 detected lesions were most commonly (88%) visualized using post-gadolinium 3D (volumetric) FLAIR sequence with fat suppression. By comparison, only 42% and 58% of lesions were visualized using conventional pre-gadolinium T1-weighted and T2-weighted imaging, respectively. DWI performed worst, permitting visualization of only 19% of detected lesions, likely reflecting the 2D nature of this sequence such that slice thickness is 3–5 mm, the low resolution of this sequence as typically performed, and the presence of significant distortion and artifact associated with the orbits on routine DWI of the brain.

This study sought to evaluate the sensitivity of MRI for CM in a population of patients with known CM in two parts. In the standard read portion, we evaluated the detection of CM lesions based solely on the original radiology report, thus reflecting real-world sensitivity. In the dedicated read portion, we evaluated whether lesions could be detected on careful reevaluation of the original scans, with particular emphasis on the globe, in order to understand the maximum potential sensitivity of MRI in an ideal setting.

There was indeed a significant discrepancy in MRI sensitivity between the standard read and dedicated read results. While the standard read detected only 50% of confirmed CM, the dedicated read detected 76% (p < 0.008), suggesting that interpretive error might account for the radiographic underdiagnosis of approximately one-third of detectable CM. Interestingly, the complexity of the MRI scans appears to account in part for underdiagnosed CM, with detection of metastatic lesions to the brain or bone potentially serving as a distractor from assessment of the orbits and also giving rise to satisfaction of search. In the standard read portion, brain metastases occurred more frequently in scans without detected CM compared to scans with detected CM (68% vs. 43%). By comparison, on dedicated read, brain metastases were found with similar frequency between both CM undetected and detected groups (43% vs. 54%, p = 0.876). Additional plausible reasons for the lower MRI sensitivity with the standard read group include a failure to include ocular scrutiny in the search pattern of interpretation of an MRI study of the brain, as well as reader fatigue and limited reading time owing to time and efficiency constraints [7, 8].

By comparison, for the dedicated read, the radiologist emphasized assessment of the globe, and no time constraints were imposed. Additionally, the radiologist was aware of the presence of CM despite being blinded to the laterality and location of the CM. We recognize that this approach disregards the practical limitations of real-world practice, and we used it solely to demonstrate the potential upper limit of CM detection with MRI. This approach also points out the importance of providing accurate clinical history when known – providing an indication of a concern for ocular metastasis will give the radiologist a better opportunity to detect and evaluate a potential lesion.

Additionally, it is important to note that most lesions (88%) were identified using a post-gadolinium volumetric FLAIR with fat saturation sequence, and many detected lesions were seen only with this sequence (Fig. 1). While volumetric FLAIR was available for every lesion imaged in the dedicated read portion of this study, an equivalent volumetric fat-suppressed FLAIR sequence may not be routinely acquired in a nonacademic setting, or it may not be performed following contrast administration or with fat suppression. The addition of this sequence to routine brain MRI metastasis protocol, coupled with careful scrutiny of the globes on this sequence, would likely significantly improve the rate of CM detection.

Even in an ideal situation, however, the dedicated read yielded only 76% detection of CM. This reflects the technical limitations of conventional MRI when imaging the globe. Motion artifacts due to blinking and gaze changes, combined with field heterogeneity at the air-cornea interface, can degrade resolution, and standard imaging planes may not be ideal for detecting subtle lesions [9, 10]. Standard reformation planes may also be limiting: while a radiologist can create oblique reformations from 3D volumetric sequences, this is a time-consuming process that would not be done during routine interpretation of a standard MRI study of the brain obtained to assess for metastatic disease. The use of surface coils for imaging the eye may improve resolution, but this is not standard practice and has largely been abandoned outside of the research setting [11]. These limitations are compounded by a slice thickness of up to 5 mm on studies that might be performed without 3D volumetric acquisition, which means that, assuming a mean tumor basal diameter of 10 mm, a CM may only be imaged on 2 slices, and many lesions would not meet this threshold.

Our study shows that several features of CM lesions are associated with improved detection by MRI. Both the standard and dedicated read portions of the study show that, unsurprisingly, larger tumors are associated with improved detection. The differences were more pronounced in the dedicated read portion, and tumor thickness appears to play a more important role in detection than does tumor basal diameter. Intuitively, it follows that CM tumor shape also plays an important role in MRI detection, in that shapes which follow the contour of the globe (i.e., thin and flat) are more likely to be missed, while shapes that induce irregularities in the normal contour of the globe (i.e., dome-shaped) are more likely to be detected. Indeed, 76% of detected CM had a dome-shaped tumor configuration as judged by ultrasound evaluation, compared to only 28% of undetected CM (p < 0.001).

In our patient cohort, the presence of visual symptoms was more sensitive at detection of CM than was MRI in the standard read. There is inherent selection bias, however, as the reason for referral to an ocular oncologist was typically the presence of visual symptoms in a patient with metastatic cancer. Additionally, the specificity of visual symptoms in detection of CM in patients with known systemic cancer cannot be ascertained in this study, as only patients with CM were included. Previous studies have demonstrated that the incidence of CM in asymptomatic patients is exceedingly low. Fenton and Barak separately reported a 0% incidence of CM following ophthalmic screening in a consecutive series of a collective 237 asymptomatic metastatic breast cancer patients, although Barak did find a 2% incidence in patients with metastatic lung cancer [4, 5]. Conversely, patients who do harbor CM are symptomatic in an estimated 80–90% of cases [1, 2, 12‒14], reflecting the frequency seen in our cohort.

These findings, coupled with ours, suggest that routine screening for visual symptoms in oncologic patients, with subsequent referral of positive cases to an ophthalmologist, may be the most sensitive and economical approach to detection of CM. Our findings testify that MRI scans are unreliable in excluding the presence of CM, though MRI scans can play an adjunctive role in this diagnosis. Given the high rate of concomitant CM with presence of CNS metastasis, particular care should be taken to screen patients with known CNS metastases for ocular metastasis, and radiologists should recognize the importance of scrutinizing the globes for CM in patients who are undergoing MRI scans of the brain as part of their oncologic evaluation and management [14, 15].

Initial MRI sensitivity in the detection of CM was 50% in our study based on routine clinical interpretations, but it improved to 76% with increased utilization of fat-suppressed FLAIR imaging and focused ocular scrutiny. MRI scans may be unreliable in independently excluding the presence of CM, but they play an important adjunctive role in making or confirming this diagnosis. In some cases, particularly when lesions are thick and/or dome-shaped, MRI scans may be the first place a diagnosis of CM is made, and the globes should be carefully assessed in MRI studies of the brain in oncology patients. Prompt and careful ophthalmic evaluation, preferably by an ocular oncologist, should be performed in all visually symptomatic patients with known primary cancers.

The study protocol was reviewed and approved by the Stanford University Institutional Review Board (IRB), protocol ID 39760. Written informed consent to participate in the study was waived by the IRB. The study was performed in compliance with the Health Insurance Portability and Accountability Act and the Declaration of Helsinki.

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article.

This study was supported by NIH NEI P30-026777, Research to Prevent Blindness, Irene and Alan Adler Ocular Cancer Initiative Fund. Funding was used to support secure data storage and statistical analysis.

Michael Yu, Sarah Miller, and Hashem Ghoraba: study design, data acquisition/analysis, and drafting manuscript. Luis Sabage: study design, data acquisition, and drafting manuscript. Nancy Fischbein and Prithvi Mruthyunjaya: study design, data interpretation, and reviewing manuscript.

Data are not available due to university data sharing agreements. Further inquiries can be directed to the corresponding author.