Sensitivity of different MRI sequences in the early detection of melanoma brain metastases

Background After the emergence of new MRI techniques such as susceptibility- and diffusion-weighted imaging (SWI and DWI) and because of specific imaging characteristics of melanoma brain metastases (MBM), it is unclear which MRI sequences are most beneficial for detection of MBM. This study was performed to investigate the sensitivity of six clinical MRI sequences in the early detection of MBM. Methods Medical records of all melanoma patients referred to our center between November 2005 and December 2016 were reviewed for presence of MBM. Analysis encompassed six MRI sequences at the time of initial diagnosis of first or new MBM, including non-enhanced T1-weighted (T1w), contrast-enhanced T1w (ceT1w), T2-weighted (T2w), T2w-FLAIR, susceptibility-weighted (SWI) and diffusion-weighted (DWI) MRI. Each lesion was rated with respect to its conspicuity (score from 0—not detectable to 3—clearly visible). Results Of 1210 patients, 217 with MBM were included in the analysis and up to 5 lesions per patient were evaluated. A total of 720 metastases were assessed and all six sequences were available for 425 MBM. Sensitivity (conspicuity ≥2) was 99.7% for ceT1w, 77.0% for FLAIR, 64.7% for SWI, 61.0% for T2w, 56.7% for T1w, and 48.4% for DWI. Thirty-one (7.3%) of 425 lesions were only detectable by ceT1w but no other sequence. Conclusions Contrast-enhanced T1-weighting is more sensitive than all other sequences for detection of MBM. Disruption of the blood-brain-barrier is consistently an earlier sign in MBM than perifocal edema, signal loss on SWI or diffusion restriction.


Introduction
Brain metastases are the most frequent brain tumors [1], most often caused by breast cancer, lung cancer, and by melanoma [2]. In the assessment of patients with cerebral metastases, MR imaging is the most sensitive diagnostic technique and an essential part of the follow-up in patients with metastatic melanoma. Approximately 40-50% of stage IV melanoma patients develop clinical manifestations of melanoma brain metastasis (MBM) [3,4], while autopsy studies suggest an even higher prevalence of 55-75% [3,5,6].
Since the introduction of new therapeutic possibilities in melanoma [7][8][9][10][11][12], early detection of metastases has an increasingly important role in managing patients with MBM since their detection could be crucial for timely change of therapy.
The imaging hallmark of brain metastases has been contrast enhancement [13][14][15][16]. Previous imaging studies on MBM have additionally focused on MRI signal intensity characteristics, e.g. high signal intensity relative to cortex on unenhanced T1-weighted (T1w) images [17,18] as well as T2 Ã -weighted signal intensity loss [19]. Both of these imaging findings could conceivably be caused by intratumoral melanin content as well as hemorrhage [17][18][19][20]. Given the high sensitivity of SWI for susceptibility artifacts and their frequent presence in MBM, it is conceivable that SWI might be the most sensitive technique for detection of at least a subset of MBM.
Further, an increasing number of patients and also referring physicians have become reluctant in accepting gadolinium based contrast agents due to their potential retention in the brain [21,22], impairment of renal function or possible contrast agent allergy and would prefer non contrast agent examinations, especially given the frequency of screening cranial MRI (cMRI) at stage IIIC and IV of melanoma.
It is therefore a relevant and hitherto unresolved issue which MRI sequence allows for reliable, early detection of new MBM and whether contrast administration must be considered mandatory. The aim of this study was to identify the MRI sequences with highest diagnostic sensitivity for reliable early detection of MBM.

Patients and ethics
This retrospective study was approved and patients´consent was waived by the ethics committee of the University of Heidelberg. We reviewed the medical record and neuroimaging studies of all patients who were referred to our imaging center between November 2005 and December 2016 with a diagnosis of melanoma. Patients with brain metastasis were included in the study and cranial MRIs (cMRI) at the time of first diagnosis of new MBM (hereinafter referred to as first diagnosis MRI) were analyzed. Patients without available cMRI at initial diagnosis had to be excluded.

Imaging protocol
The majority of cMRI scans were performed at our institution (72.0%, 223/310) on a Siemens Magnetom Symphony 1.5 Tesla MRI system (67.3%, 150/223). For detailed information of the used MRI systems and contrast media see Table 1.

Image analysis
Image analysis was performed on a per-patient and per-lesion basis, with a maximum of five lesions per patient. If a patient had more than five newly developed MBM, the five smallest MBM were analyzed. If available, MRI examinations prior to the first diagnosis MRI (pre-diagnosis MRI) were consulted to evaluate on which sequences MBM were retrospectively detectable at very early stages. Reading of the MBM was performed by one of two radiologists with eight or one year of experience in neuroimaging on our picture archiving and communication system (Centricity PACS, GE Healthcare Integrated IT Solutions, Barrington, Illinois). A prereading session was conducted to reach agreement between readers with regard to the different Conspicuity Scores (CS). Sequence order of analysis was random. Image analysis was carried out with readers blinded to the patients' data and according to the following description: • In each patient, up to five lesions were rated with regard to lesion conspicuity (4 point ordinal scale: CS = 0, not detectable; CS = 1, pathological correlate detectable with prior knowledge of existence and localization; CS = 2, detectable, CS = 3, clearly visible, see Fig 1). • Additionally, lesion size for each lesion was determined (product of the longest and the orthogonal diameter) and signal intensity on non-enhanced T1-weighted and susceptibility-weighted imaging was evaluated relative to the contralateral cortex (hyper-, iso-, or hypointense).
• Those MBM with an imaging appearance compatible with leptomeningeal or pachymeningeal localization were also recorded.

Reference standard
In every patient the diagnosis of the primary tumor of malignant melanoma was confirmed by histopathology. Final diagnosis of MBM for every brain lesion was based on comparison with prior examinations, histopathological data, and most importantly by clinical and MRI follow-up.

Statistical analysis
Descriptive statistics were carried out using Microsoft Excel 2016. The mean CS with standard deviation (SD) was calculated and a subgroup analysis was performed for: • MBM of melanotic and amelanotic primary melanoma, • MBM with lesion size up to 5 mm in diameter, • lesions found in the MRI scan prior to the one that led to diagnosis (see Fig 2), and • MBM sited lepto-or pachymeningeally.
Statistical testing was performed with the software package R (R Foundation for Statistical Computing, version 3.1.1, 2014-07-10). Shapiro-Wilk test was used to test for standard distribution of the data and Wilcoxon signed-rank test was performed to compare the CS of the two most sensitive sequences. Significance was set at the p < 0.05 level.
The ceT1w sequence had a significantly higher CS than the second best sequence (paired Wilcoxon-test, p-value < 0.001). Notably, not a single metastasis (0/712) was rated with a CS  Table 3 shows detailed CS results. Since lesions with a CS of 2 and 3 are generally detectable by attentive radiological reading, the sensitivity in the 720 lesions of our patient group was 99.7% (

Relevance of sequences in addition to contrast-enhanced imaging
In 24 of 712 (3.4%) lesions, at least one sequence was rated with a higher CS than the ceT1w sequence. Eleven of these lesions were displayed more clearly by one other sequence, while in 13 lesions there were multiple superior sequences. In total, SWI was 17 times superior to ceT1w, FLAIR 10 times, T1w 6 times and T2w as well as DWI 4 times each.
The two of 712 lesions that were detectable on ceT1w only with prior knowledge (corresponding to a CS of 1) were better depicted on SWI (CS 3) in one case, and on FLAIR (CS 3) in the other.

Assessment of pre-diagnosis cMRIs
A pre-diagnosis MRI was available for 456 of 720 lesions (63.3%). These pre-diagnosis MRIs were assessed regarding the presence of MBM with knowledge of the MBM in the first diagnosis MRI. The average time between first diagnosis MRI and pre-diagnosis MRI was 12.3 weeks (±4.2 weeks). Retrospectively, 33 MBM already had a correlate in the pre-diagnosis cMRI (an example is shown in Fig 2), with 18 well detectable (CS !2 in any sequence) and the others only detectable retrospectively (CS 1 in any sequence). All of these lesions (33/33) had a correlate on ceT1w (100%). In no case (0/33) was another sequence superior to contrast-enhanced imaging in depicting the MBM. Out of the 18 lesions with a CS !2 on contrast-enhanced MRI (18/33, 54.5%), 3 were also detectable on SWI (3/23; 13.0%) and 4 on FLAIR (4/32, 12.5%).

Discussion
As a principal finding, our results demonstrate the highest sensitivity of contrast-enhanced T1-weighted imaging for the detection of brain metastases from malignant melanoma beyond all other sequences. While the usefulness of contrast-enhanced sequences for brain metastases was never questioned, the clear superiority of ceT1w over all other sequences and its exquisite sensitivity was not expected to this degree and has implications for differential diagnosis of unclear suspicious findings in melanoma patients.
It is well known that MRI contrast agents are an essential aid in the early detection of intracranial metastases [24][25][26]. Nevertheless, it has been assumed that MBM may behave differently. Gaviani et al. found hypointense lesions on T2 Ã -weighted images without correlates on ceT1w images and other unenhanced sequences in patients with malignant melanoma, whereas those lesions in a comparison group with patients with lung cancer were not found. The authors hypothesized that in patients with malignant melanoma those lesions could represent early stages of melanoma metastases [20]. Our results are in contrast to this hypothesis and confirm results of Gramsch et al. who showed that 29 cases of isolated focal areas of susceptibility artifacts in brain parenchyma of 18 melanoma patients were not found to develop as metastases in their follow-up period of 6-46 months [27]. Although approaches similar to ours have been described for metastases of other locations and for different investigations so far [28][29][30][31][32], to our knowledge, our study is the first to compare the visibility of brain metastases across commonly used MRI sequences. Different MRI sequences reflect different aspects of tumor pathophysiology. Signal characteristics are highly variable across metastases and tumor entities and also change in the progression of the disease. In many higher-grade primary central nervous system tumors, signal increase in T2-weighted and FLAIR images or diffusion restriction most frequently indicate initial diagnosis or tumor progression [33]. In contrast, melanoma brain metastases apparently generally first manifest with disruption of the blood-brain-barrier before signal or diffusion abnormalities due to high cellularity or perilesional edema are detectable.
The high sensitivity of contrast-enhanced imaging for detection of MBM in our study group is supported by three main findings. First, nearly all lesions were conspicuous on ceT1w at the time of first diagnosis. All MBM had a ceT1w correlate, and only two of 712 lesions were detectable in ceT1w only with prior knowledge. Second, only 24 of 712 lesions were considered more conspicuous in a sequence other than ceT1w on the MRI of first diagnosis. And third, only contrast-enhanced imaging depicted a correlate for all lesions that were retrospectively detectable on the pre-diagnosis MRI. Taken together, we believe that these results establish contrast enhancement as the most sensitive early sign for MBM. This is of considerable importance, because many patients and physicians are increasingly critical of serial injections of gadolinium based contrast agents (GBCAs) due to the unclear effects of the intracerebral deposition of some GBCAs as well as potential contrast agent allergy or renal impairment.
Our study has limitations. First, only intracerebral metastases were assessed. Extracerebral metastases, such as bone metastases of the skull, were not included in the formal analysis even though they are similarly important for staging or therapy decisions. Second, an observerexpectancy bias cannot be precluded, because the ceT1w scan quickly turned out to be superior compared with the other sequences. From then on, the identification of potential lesions might have been guided by the contrast-enhanced scan. Third, only the sensitivity of different sequences in depicting melanoma brain metastases was assessed, whereas the specificity could not be assessed. Moreover, the potential impact of previous therapies such as immuno-or targeted therapies as well as cranial radiotherapy on the depiction of MBM on the different sequences was not evaluated. With FLAIR hyperintensities and susceptibility artifacts being frequent but unspecific findings in the general population and even more so in treated patients, contrast-enhanced imaging can be considered even more relevant.
Fourth, scan parameters varied at different MRI scanners during the included study period (2005-2016), leading to slightly different image quality and appearance: Contrast-to-noise ratio is significantly higher at 3 Tesla than at 1.5 Tesla [34] and high resolution isotropic 3D imaging is most accurate in depiction of small lesions. In that respect, contrast enhanced T1-weighting might have had an advantage over the other sequences as it is frequently acquired with the highest resolution.
Finally, type and dose of contrast media as well as time-to-acquisition differed between contrast-enhanced images. These factors have a well-known impact on lesion conspicuity: Signalto-noise ratio increases with escalation of dose and time-delay between contrast media application and image acquisition [35][36][37][38][39][40].
Nonetheless, all scans were treated equally in the assessment, indicating that ceT1w imaging is superior to other sequences in the detection of MBM independently from contrast-media dependent parameters.
Our results suggest the optimization of contrast-enhanced MRI for even more precise detection of small MBM. In this study, mainly 1.5 Tesla MRI systems and 2D contrastenhanced scans were used. It is well known that higher magnetic field strength and 3D MR images, such as MPRAGE and SPACE, are superior for detecting small lesions because of higher tumor-to-brain-contrast [34], less partial volume effect and less blood flow artifacts [14,41,42]. Several investigators have already reported the increase in lesion detection rate using SPACE imaging compared with MPRAGE imaging for evaluating brain metastases [43][44][45]. These studies have mentioned possible mechanisms, including higher contrast enhancement effect, greater magnetization transfer effect, and the effect of vascular signal suppression on SPACE [46][47][48].

Conclusion
In patients with metastatic melanoma contrast-enhanced T1-weighting is the most sensitive MRI sequence for early detection of brain metastases. No other sequence including T2w-FLAIR or susceptibility-weighted imaging can fully replace the application of contrast media for highest sensitivity. It is therefore likely beneficial to optimize image quality and assessment of the contrast-enhanced sequence to precisely detect small MBM.