Manual Thrombus Density Measurement Depends on the Method of Thrombus Delineation.

Dear Sir:

We would like to comment on the review by Heo et al. [1] on computed tomography (CT) thrombus imaging in acute large intracranial vessel occlusion, with focus on manual thrombus delineation. As noted, better recanalization is achieved when hyperdense thrombi are subjected to intravenous thrombolysis or endovascular thrombectomy [2-5], but this finding was inconsistent [6-9].

To demonstrate the influence of manual thrombus delineation, we retrospectively assembled data from 20 patients with large intracranial vessel occlusion who underwent pretreatment, non-enhanced cranial CT (Brilliance 64, Philips, Amsterdam, the Netherlands; 120 kV, 320 mA, reconstructed slice thickness 2.5 mm). Regions of interest (ROIs) were placed by one reader using different methods (Fig. 1A). One (method 1) [4,6] or three (method 2) [2,9] circular ROIs were placed in the most hyperdense-appearing part of the occluded artery. Using method 3, the hyperdense artery was delineated with irregular ROIs for every CT slice [3,5], yielding 124 ROIs. Hyperdensities with >100 Hounsfield units (HU) were considered calcifications and excluded. For methods 2 and 3, the mean density was calculated. Since interrater reliability was reportedly high [10], 50 randomly selected ROIs were reanalyzed by a second, blinded reader. Mean density differences between the raters and the intraclass correlation coefficient (ICC) were calculated. Density values were subjected to the Kolmogorov-Smirnov test to verify normal distribution. One-way analysis of variance (ANOVA) was performed to detect between-group differences. The Tukey-Kramer post hoc test was applied to determine differences between groups, and P<0.01 was considered significant.

Figure 1.

(A) Three methods of manual thrombus delineation. One (method 1, large circular region of interest [ROI]) or three (method 2, circular ROIs) were placed into the most hyperdense-appearing part of the vessel, or the whole hyperdense vessel was encompassed within an ROI on every slice on which the vessel was visualized (method 3, white outline). The mean density was determined (method 2 and 3). (B) Box and whisker plots of density values for the three methods. HU, Hounsfield unit.

Density values were normally distributed (P>0.999). Differences between the two readers were small (2.8±1.2 HU) and ICC between raters was 0.93. Henceforth, only ROIs defined by the first observer were used for analysis. One-way ANOVA was significant (P<0.0001). The mean thrombus density was 65.04±6.59 (method 1), 60.92±6.65 (method 2), or 54.93±5.7 HU (method 3) (all P<0.01) (Fig. 1B). Mean differences in thrombus density were 5.98±2.34 (method 1 vs. 2), 4.12±2.48 (method 2 vs. 3), and 10.11±3.06 HU (method 1 vs. 3).

Thus, manual thrombus density measurements depend on the measurement method. Despite small sample size, measured densities were significantly different, with whole thrombus delineation yielding lowest density values. Further factors (slice thickness, treatment modality, heterogeneous thrombi, and beam hardening artefacts) must be considered. While automated algorithms [11] might be useful, we agree with Angermaier and Langner [7] that technical standards are needed for manual thrombus density measurements.