Ludger Feyen, Abstract Co-Author: Nothing to Disclose

Harald Seifarth MD, Abstract Co-Author: Research Consultant, Siemens AG

Thomas Niederstadt MD, Abstract Co-Author: Nothing to Disclose

Volker Hesselmann MD, Abstract Co-Author: Nothing to Disclose

Walter Leonhard Heindel MD, Abstract Co-Author: Nothing to Disclose

Andre Kemmling MD, Presenter: Nothing to Disclose

To evaluate contrast enhancement of early infarct by a post processing filter for native CT utilizing non-linear HU lookup in the low frequency band spectrum.

CT images of 35 consecutive patients with confirmed acute middle cerebral arterial stroke (symptom onset to admission imaging <3h) and a representative control group (35 patients) without infarction were selected. Infarct differentiation was evaluated in admission CT images before and after contrast enhancement (Neuro BestContrast, Siemens): (1) Infarct contrast was calculated between regions of interests (ROIs) within infarct and contra-lateral side. (2) Four raters (2 experts, 2 novices) evaluated CT images for presence of parenchymal hypoattenuation (5-point scale). Sensitivity and specificity for infarct detection and interrater kappas were calculated for standard and enhanced CT images. (3) Voxel based sensitivity and specificity for infarct delineation was assessed (Fig): potential infarct lesions in CT (fixed window width; 10 to 80HU) and lesions of infarct core defined by CT perfusion (CBV) maps were segmented slice-by-slice and co-registered (Analyze 10.0, AnalyzeDirect). Voxels in CT infarct lesions were defined as “true positive”, “true negative”, “false positive” or “false-negative” based on overlap with the reference CBV lesions.

(1) The contrast between lesion and non-lesion ROIs was significantly higher in enhanced CT images vs. standard CT images (20.6% vs. 14.7%). (2) The sensitivity for infarct detection by novice raters increased using enhanced CT vs. standard CT (71% vs. 59%, p=0.08). The sensitivity for infarct detection did not significantly increase for expert raters (73% vs. 68%, p=0.19). Interrater kappa between expert and novice raters significantly improved using contrast enhanced CT (0.60 vs. 0.82, p<0.05). There was no difference in specificity for infarct detection using standard or enhanced CT images (>90% regardless of rater). (3) Sensitivity for voxel-wise stroke detection was 67% for standard- and 75% for enhanced CT images (p<0.05). Respectively, specificity for voxel-wise stroke detection remained unchanged (98% vs. 96%, p<0.05).

Detection of early infarct in native CT is facilitated with a contrast enhancement filter using non-linear HU lookup in the low frequency band spectrum.

The presented processing filter facilitates infarct detection particularly for less experienced readers.

Feyen, L, Seifarth, H, Niederstadt, T, Hesselmann, V, Heindel, W, Kemmling, A, Improvement of Native CT in Acute Stroke Imaging: Optimized Contrast Enhancement by Nonlinear HU Lookup in the Low Frequency Domain Facilitates Differentiation of Early Infarct Core.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11014460.html