Abstract TMP14: Masked Smoothing for Computed Tomography Perfusion Imaging

• Published in: Stroke (1970) 2013-02, Vol.44 (suppl_1)
• Main Authors: Wack, David, Snyder, Kenneth, Seals, Kevin, Natarajan, Sabareesh, Fisher, Jason, Creighton, Terrance, Eller, Jorge, Siddiqui, Adnan
• Format: Article
• Language: English
• ISSN: 0039-2499; 1524-4628
• DOI: 10.1161/str.44.suppl_1.ATMP14

Abstract only Objective: CT perfusion imaging is characterized by a high contrast ratio between different structure types. This facilitates the segmentation of an image into different classes, such as tissue and vasculature. However, grey and white matter tissue regions have relatively low values and can suffer from poor signal to noise ratios. While smoothing can improve the image quality of brain parenchyma, the inclusion of high contrast vascular voxels skews the dataset. It is thus desirable to smooth the tissue voxels independently from other voxel types, as has been implemented previously using mean filters with small kernel sizes. Bilateral filters have also been proposed, allowing a varying influence of neighboring voxels based on the similarity of the voxel’s intensity. These filters are orders of magnitude slower, however, as the smoothing kernel is not directly separable. A separable kernel can be applied to an image volume as 3 implementations of a 1D filter. A kernel of size 50x50x50, for example, is applied to a volume with 3x50 (150) multiplications per voxel if it is separable and 50^3 (125,000) if it is not. Our approach, which applies smoothing to tissue voxels without including neighboring vessel voxels in the kernel, can be calculated by dividing the results of just two applications of a separable kernel. Methods: We compare our Masked Smoothing method to alternative Gaussian smoothing approaches using an unaltered image and an image where vascular voxels have been set to zero. The base image size is 512x512x320x19. Results: Using simulations we demonstrate that Masked Smoothing does not bias the underlying tissue value, whereas the other smoothing methods cause significant bias. Furthermore, using actual CT perfusion data, we demonstrate a significant difference in the calculated cerebral blood flow and cerebral blood volume based on the smoothing method used. Conclusions: Given the enormous datasets inherent to CT perfusion and the need for rapid clinical evaluation, it is crucial to develop a smoothing method that prevents the bias of tissue values from neighboring vessels and executes rapidly. Our novel smoothing method achieves both goals.