Improved Delineation of Pulmonary Embolism and Venous Thrombosis Through Frequency Selective Nonlinear Blending in Computed Tomography
The aim of this study was to test the hypothesis that a novel frequency selective nonlinear blending (NLB) algorithm increases the delineation of pulmonary embolism and venous thrombosis in portal-venous phase whole-body staging computed tomography (CT).Materials and Methods
A cohort of 67 patients with incidental pulmonary embolism and/or venous thrombosis in contrast-enhanced oncological staging CT were retrospectively selected. Computed tomography data sets were acquired 65 to 90 seconds after intravenous iodine contrast administration using state-of-the-art multi-detector CT scanners. A novel frequency selective NLB postprocessing technique was applied to reconstructed standard CT images. Two readers determined the most suitable settings to increase the delineation of pulmonary embolism and venous thrombosis. Outcome measure included region of interest and contrast-to-noise ratio (CNR) analyses, image noise, overall image quality, subjective delineation, as well as number and size of emboli and thrombi. Statistical testing included quantitative comparisons of Hounsfield units of thrombus and vessel, image noise and related CNR values and subjective image analyses of image noise, image quality and thrombus delineation, number and size in standard, and NLB images.Results
Using frequency selective NLB settings with a center of 100 HU, delta of 40 HU, and a slope of 5, CNR values of pulmonary embolism (StandardCNR, 10 [6, 16]; NLBCNR, 22 [15, 30]; P < 0.001) and venous thrombosis (StandardCNR, 8 [5, 15]; NLBCNR, 12 [7, 19]; P = 0.0007) increased. Mean vascular enhancement using NLB was significantly higher than in standard images for pulmonary arteries (Standard, 138 [118, 191] HU; NLB, 269 [176, 329] HU; P < 0.0001) and veins (Standard, 120 [103, 162] HU; NLB, 169 [132, 217] HU; P < 0.0001), respectively. Image noise was not significantly different between standard and NLB images (P = 0.64-0.88). There was substantial to almost perfect interrater agreement as well as a significant increase of overall image quality (P < 0.004) and subjective delineation of the thrombotic material (P < 0.0001) in both subgroups. Nonlinear blending images revealed 8 additional segmental and 13 subsegmental emboli. Thrombus sizes were not significantly different, but subjective accuracy of the measurement could be significantly increased using NLB (P = 0.03).Conclusions
Postprocessing of standard whole-body staging CT images with frequency selective NLB improves image quality and the delineation of pulmonary embolism and venous thrombosis.