Diagnostic Confidence of Contrast-Enhanced T1-Weighted MRI for the Detection of Brain Metastases: 3D FSE versus 3D GRE–Based Sequences

DISCUSSION

CE 3D T1-weighted images are crucial to the accurate early detection of brain metastasis, with widespread implications for patient care. The ability to correctly detect the number of metastases is paramount, because it greatly affects the prognosis and choice of therapy, deciding between local therapy such as stereotactic radiosurgery or use of whole-brain radiation or systemic therapy. This study found that CE SPACE detected more lesions than 3D FLASH. This improved ability to detect CE lesions can, in part, be attributed to a higher CNR, better image quality, and fewer image artifacts on SPACE versus 3D FLASH, as shown in Table 2. There is abundant literature demonstrating that 3D FSE-based T1-weighted sequences, including SPACE, have superior lesion conspicuity and CNR compared with 3D GRE-based sequences, mostly MPRAGE.^6⇓–8,13 Our findings mirror previously published data, including the results of a 2016 meta-analysis¹⁴ that found superior detection of brain metastasis with 3D FSE-based T1 sequences compared with MPRAGE in 5 published studies.^6,7,12,15,16

While the MPRAGE sequence offers superior gray matter–white matter contrast as a result of the IR preparation, its detection of CE lesions may be compromised, especially in those with a low gadolinium concentration^11,17 or enhancing lesions located within the bright white matter.¹⁰ Early simulation studies demonstrated higher lesion/tissue CNR with 3D FLASH than with MPRAGE, though they were both lower than a spin-echo sequence at a low gadolinium concentration, which may be improved with sequence optimization.¹¹

There are limited studies comparing non–IR-prepared 3D GRE versus MPRAGE or SPACE for the detection of brain metastasis. Wetzel et al¹⁸ compared VIBE with MPRAGE at 1.5T and found a significantly higher SNR and CNR with VIBE in 30 patients with focal brain lesions, including 10 with suspicion of metastatic disease or CNS lymphoma. Majigsuren et al¹⁹ compared T1-Cube (GE Healthcare) and 3D fast-spoiled gradient-recalled sequences, which are equivalent to the T1 SPACE versus 3D FLASH sequences in this study but on a 3T GE scanner and found significantly higher CNRs with T1-Cube in 9 patients with brain metastasis. More recently, Danieli et al⁹ compared MPRAGE, SPACE, and VIBE in 54 contrast-enhancing tumors, including 16 metastases. Although both SPACE and VIBE provided significantly higher CNRs than MPRAGE, there was no significant difference between SPACE and VIBE. Note that VIBE is a similar sequence to 3D FLASH, with the main difference being the use of asymmetric k-space sampling and interpolation for acceleration.²⁰

The present study is the first to directly demonstrate the superior performance of SPACE over a non-IR-prepared 3D GRE sequence for the detection of CE metastatic lesions in a sizable patient population (n = 72). An interesting finding in our study was that although the CNR result from the qualitative assessment significantly differed between the 2 sequences, the difference in the quantitative assessment did not reach statistical significance. This outcome may be attributed to the small sample size (n = 16) in our quantitative CNR calculation (due to the inclusion criterion of a solidly-enhancing lesion size of >1 cm) or to the small CNR difference between the 2 sequences, similar to the finding in Danieli et al.⁹

In this study, we found fewer indeterminate lesions using SPACE, likely due to the suppression of vascular enhancement. This metric can be used as a surrogate measure of how confident the interpreting radiologist is in the imaging findings. Small bright vessels can be difficult to differentiate from metastases, particularly in the basal ganglia or along the superficial aspect of the gyri, contributing to the decreased confidence of the interpreting radiologists. The SPACE sequence has inherent flow suppression from the intravoxel dephasing due to the uncompensated gradient moments in the echo-train and the stimulated echoes from variable flip angle radiofrequency pulses.¹⁰ Thus, enhancement can more confidently be attributed to metastatic disease. The importance of flow suppression has been further realized by improved metastasis detection using the black-blood vascular suppression (delay alternating with nutation for tailored excitation [DANTE]) version of SPACE.²¹ Most recently, Chkili et al²² compared the use of SPACE in combination with VIBE with SPACE alone and found that SPACE alone performed like SPACE and VIBE combined.

Improved margin delineation was seen on SPACE versus 3D FLASH (Fig 2B). CE 3D T1-weighted images are used for stereotactic radiosurgery treatment planning, in which not only lesion detection but also precise margin delineation is paramount. This improved margin delineation may result in improved targeting of radiation therapy, with downstream clinical implications. Furthermore, accurate margin delineation may improve the accuracy of lesion measurement and thus the accuracy of treatment-response assessment. A significant difference in volume measurement on SPACE versus MPRAGE was demonstrated by Danieli et al,⁹ with SPACE generating larger volumes. The consensus recommendation for standardized brain tumor imaging for clinical trials in brain metastases recommends using SPACE at 3T.¹⁰

We found no statistically significant differences in reading time between the 2 sequences. This finding is clinically meaningful when implementing SPACE into a busy clinical practice and is helpful for physician buy-in. The trend of slightly longer interpretation time for SPACE may be attributable to unfamiliarity with the sequence by the interpreting radiologists and possibly to more lesions being detected on SPACE, resulting in a longer time spent interpreting the study.

The use of SPACE for the detection of brain metastases has potential drawbacks related to reduced gray-white matter differentiation and the lack of vascular enhancement. The absence of enhancement of superficial venous structures may be undesirable for surgical planning.

This study has some limitations. This was a retrospective single-institution study that aimed for quality improvement. Therefore, the control condition, ie, the 3D FLASH sequence, was limited by our standard of care protocol with a different acquisition plane (axial) and voxel size (0.9 × 0.9 × 1.4 mm³) than the SPACE sequence (sagittal and 1-mm isotropic, respectively). The image review was primarily performed using axial images, which, in the case of SPACE, were reconstructed from sagittal acquisitions, while the axial 3D FLASH images represented source images. Due to the different appearances of SPACE and 3D FLASH images, it was not possible to make the qualitative assessment a blind experiment. Therefore, we implemented a 2-week separation between reviewing the 2 sequences from the same patient to at least avoid memory bias. Our study evaluated only sequence performance at 3T. A lower SNR on SPACE at 1.5T may prove challenging and will require additional studies. The detection number of metastases and indeterminate lesions was not validated with follow-up imaging to establish the accuracy of lesion detection. Without follow-up imaging, it is possible that some of the lesions included represented other etiologies such as subacute infarcts. Lesion size was not recorded, so the effect of lesion size on the performance of the sequences was not included in the primary analysis. However, the subsequent evaluation found that all except 1 lesion not detected on either of the sequences were small, <5 mm. Finally, lesion segmentation and morphometric analyses were not performed in this study, but they are important perspectives in comparing contrast-enhancing sequences in addition to the lesion detectability.