Supplementary Material for: Thin Slices and Maximum Intensity Projection Reconstructions Increase Sensitivity to Hyperdense Middle Cerebral Artery Sign in Acute Ischemic Stroke
datasetposted on 28.07.2020, 11:23 by Rosskopf J., Kloth C., Dreyhaupt J., Braun M., Schmitz B.L., Graeter T.
Introduction: Cranial nonenhanced CT (NECT) imaging in hyperacute ischemic stroke is rarely used for assessing arterial obstruction of middle cerebral artery by identifying hyperdense artery sign (HAS). Considering, however, its growing importance due to its impact on the decision-making process of thrombolysis with or without mechanical thrombectomy improved sensitivity to HAS is necessary, particularly in the group of less experienced clinicians being frequently the first one assessing the presence of HAS on NECT. Objective: The present study aimed to investigate the effect of different NECT image reconstructions on the correct detection of hyperdense middle cerebral artery sign in a cohort of observers with lower experience level on NECT. Particularly, MIP image reconstructions were expected to be useful for less experienced observers due to both strengthening of the hyperdensity of HAS and streamlining to less image slices. Methods: Twenty-five of 100 patients’ NECT image data presented with HAS. Sixteen observers with lower practice level on NECT (10 radiologists and 6 neurologists) evaluated independently the 3 image reconstructions of each data set with thin slice 1.5 mm, thick slab 5 mm, and 6-mm maximum intensity projection (MIP) and rated the presence of HAS in middle cerebral artery. A GEE model with random observer effect was used to examine the influence of the 3 image reconstructions on sensitivity to HAS. A linear mixed effects regression model was used to investigate the ranking of detectability of HAS. Interrater reliability was determined by Fleiss’ kappa coefficient (κ). Results: Recognition of HAS and sensitivity to HAS significantly differed between the 3 image reconstructions (p = 0.0106). MIP and thin slice reconstructions yielded each on average the highest sensitivities with 73% compared to thick slab reconstruction with 45% sensitivity. The interobserver reliability was fair (κ, 0.3–0.4). Detectability of HAS was significantly easier and better visible ranked on MIP and thin slice reconstructions compared to thick slab (p < 0.05). Conclusion: MIP and thin slice reconstructions increased the sensitivity to HAS (73%), whereas thick slab reconstructions seemed to be less appropriate (45%).