Introduction

CVR mapping using T₂*-weighted BOLD MRI with breath-hold technique can be easily applied in clinical settings and can help evaluate disease status and neurovascular uncoupling in patients with brain tumors [1]. Data-driven analysis typically requires segmentation of gray matter and suggests removing lesions for generating the signal model [2,3]. This study aimed to derive relative cerebrovascular reactivity maps from breath-holding challenges and to investigate the impact of lesion inclusion of such approach in brain tumor patients.

Methods

Image acquisition
MR datasets from 18 brain tumor patients were analyzed retrospectively (age [mean/sd] = 47.6/13.6 years, 9 females). Images were collected at a 3 T clinical scanner (GE Healthcare, Waukesha, WI, USA) as part of the presurgical planning protocol. Functional imaging was acquired with a 2D gradient-echo echo-planar imaging (GE-EPI) readout (TR/TE/FA=3000 ms/25 ms/90°, in-plane matrix size = 64x64, 32 slices, resolution = 3.75 x 3.75 x 4 mm³, 70 repetitions). Cerebrovascular reactivity mapping was implemented via a breath-holding (BH) technique. After a 30-s natural breathing baseline, patients were instructed to perform three cycles of breath-holding tasks, each consisting of a 15-s breath-holding followed by a 45-s natural breathing. High-resolution structural images were acquired to provide anatomical landmarks and tissue/lesion segmentations, including pre- and post-contrast T₁-weighted (T₁w) inversion recovery-prepared spoiled gradient echo, T₂-weighted (T₂w) fluid-attenuated inversion recovery (FLAIR), and T₂w fast-spin echo sequence.

Image processing
The streamlined processing pipeline was conducted using FSL v6.0.7.4 (FMRIB, Oxford, UK) and seeVR toolbox v1.6 [3], including motion correction, skull stripping, registration, segmentation, despiking, spatial smoothing (FWHM = 4 mm), nuisance regression, and general linear modeling. The gray matter (GM) masks were calculated from the pre-contrast T₁w images. Tumor lesions were automatically segmented using the cancer imaging phenomics toolkit (CaPTk) [4] and then revised by a neuroradiologist. Anatomical references were aligned to the native EPI space. An EPI-based whole brain (WB) mask was calculated based on the signal and noise levels without prior knowledge of anatomical references. The BOLD time course was normalized to the baseline average. Relative CVR amplitude maps were calculated by correlating BOLD response to averaged reference curves extracted from user-defined masks. Four sets of reference signals were tested based on the following conditions: (1) lesion-excluded GM [GM L(-)], (2) lesion-excluded WB [WB L(-)], (3) GM, and (4) WB. The pairwise correlation was computed using MATLAB 2023a (The MathWorks, Inc., Natick, MA, USA) to test the agreement in the resultant relative CVR amplitudes of the tested conditions. Relative CVR amplitude maps were scaled to corresponding 98^th percentile values for visual comparison.[5]

Results

Figure 1 demonstrates the group mean normalized time courses extracted from the reference tissues. Overall, the reference curves were found to have similar appearances of temporal BOLD response when extracted from different reference tissues. Minor variations were found in signal time curves before spatial and temporal smoothing, whereas the differences were minimized with the processing steps. This implies that a properly handled processing pipeline could reduce the dependency on sources of errors, such as lesion inclusion and misalignment.
Figure 2 shows the four sets of CVR maps that reported low relative CVR in brain tumors. The relative CVR maps indicated negligible differences between the image contrast. The pairwise correlation (Table 1) showed significant correlations (P<.001) between the relative CVR amplitudes generated with the choice of reference signals. The inclusion of the lesion did not alternate the overall appearance of the relative CVR maps. Note that WB masks were derived without information from a separate image. This suggested the feasibility of measuring relative CVR with a data-driven approach when anatomical information is unavailable.

Discussion

This study provided insight into the robustness of relative CVR mapping derived from reference tissue signals when lesions are present. Our results suggested relative CVR mapping can be acquired without additional prior knowledge of anatomical references, respiratory recordings, or end-tidal CO₂. The current approach provided a relative measure to identify the overall appearance of CVR contrast. One potential drawback is that using WB reference signals may degrade the accuracy of detected lag time. While the current findings suggested limited differences from the choice of reference signals, the results may be compromised by partial volume blurring and temporal aliasing from the clinical acquisition.

Acknowledgements

No acknowledgement found.