Automated Assessment of Cerebral Arterial Perforator Function on 7T MRI

Abstract

Background: Blood flow velocity and pulsatility of small cerebral perforating arteries can be measured using 7T quantitative 2D phase contrast (PC) MRI. However, ghosting artifacts arising from subject movement and pulsating large arteries cause false positives when applying a previously published perforator detection method.

Purpose: To develop a robust, automated method to exclude perforators located in ghosting artifacts.

Study type: Retrospective.

Subjects: Fifteen patients with vascular cognitive impairment or carotid occlusive disease and 10 healthy controls.

Field strength/sequence: 7T/cardiac-gated 2D PC MRI.

Assessment: Perforators were automatically excluded from ghosting regions, which were defined as bands in the phase-encoding direction of large arteries. As reference, perforators were manually excluded by two raters (T.A., J.J.M.Z.), based on perforator location with respect to visible ghosting artifacts. The performance of both censoring methods was assessed for the number of (N_included ), mean velocity (V_mean ), and pulsatility index (PI) of the included perforators.

Statistical tests: For within-method comparisons, inter- and intrarater reliability were assessed for the manual method, and test-retest reliability was assessed for both methods from repeated 2D PC scans (without repositioning). Intraclass correlation coefficients (ICCs) and their 95% confidence intervals (CIs) were determined for N_included , V_mean , and PI for all within-method comparisons. The ICC to compare between the two methods was determined with the use of both (test-retest) scans using a multilevel nonlinear mixed model.

Results: The automated censoring method showed a moderate to good ICC (95% CI) vs. manual censoring for N_included (0.73 [0.58-0.87]) and V_mean (0.90 [0.84-0.96]), and a moderate ICC for PI (0.57 [0.37-0.76]). The test-retest reliability of the manual censoring method was considerably lower than the interrater and intrarater reliability, indicating that scanner noise dominates the uncertainty of the analysis.

Data conclusion: The proposed automated censoring method can reliably exclude small perforators affected by ghosting artifacts.

Level of evidence: 3.

Technical efficacy stage: 1.

Keywords: 7 Tesla MRI; blood-flow pulsatility; blood-flow velocity; cerebral arterial perforators; ghosting artifacts.

FIGURE 1

Magnitude, phase, and phase contrast (PC)‐modulus images. Ghosting artifacts (between the arrowheads) are apparent in the phase‐encoding direction, mainly in the phase and PC‐modulus images.

FIGURE 2

Planning of the 2D PC slice in the centrum semi‐ovale projected on the T₁‐weighted image. The dashed line is aligned with the bottom of the corpus callosum, indicating the angle of the slice. The slice is then located 15 mm superior to the top of the corpus callosum (solid line).

FIGURE 3

Example of the manual censoring method. Detected perforators located inside ghosting artifacts are circled in red, perforators located outside ghosting artifacts are circled in green, projected on the magnitude, phase, and PC‐modulus image.

FIGURE 4

Overview of the automated censoring method. Steps 1–4, as described in the Materials and Methods under the heading Automated Censoring. In step 1, spatial inhomogeneities are corrected. In step 2, the automatically identified large arteries of more than two voxels are shown in red on the detrended and filtered image. In step 3, the vertical stripes created are shown in blue, first for one cluster, then for all clusters. The white matter mask contour is shown in white on the magnitude image. In step 4, the included perforators are circled in green (located outside the stripes) and the excluded perforators are circled in red (located on the stripes) on the magnitude, phase, and PC‐modulus image.