Share Email Print

Proceedings Paper

High quality high spatial resolution functional classification in low dose dynamic CT perfusion using singular value decomposition (SVD) and k-means clustering
Author(s): Francesco Pisana; Thomas Henzler; Stefan Schönberg; Ernst Klotz; Bernhard Schmidt; Marc Kachelrieß
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

Dynamic CT perfusion acquisitions are intrinsically high-dose examinations, due to repeated scanning. To keep radiation dose under control, relatively noisy images are acquired. Noise is then further enhanced during the extraction of functional parameters from the post-processing of the time attenuation curves of the voxels (TACs) and normally some smoothing filter needs to be employed to better visualize any perfusion abnormality, but sacrificing spatial resolution. In this study we propose a new method to detect perfusion abnormalities keeping both high spatial resolution and high CNR. To do this we first perform the singular value decomposition (SVD) of the original noisy spatial temporal data matrix to extract basis functions of the TACs. Then we iteratively cluster the voxels based on a smoothed version of the three most significant singular vectors. Finally, we create high spatial resolution 3D volumes where to each voxel is assigned a distance from the centroid of each cluster, showing how functionally similar each voxel is compared to the others. The method was tested on three noisy clinical datasets: one brain perfusion case with an occlusion in the left internal carotid, one healthy brain perfusion case, and one liver case with an enhancing lesion. Our method successfully detected all perfusion abnormalities with higher spatial precision when compared to the functional maps obtained with a commercially available software. We conclude this method might be employed to have a rapid qualitative indication of functional abnormalities in low dose dynamic CT perfusion datasets. The method seems to be very robust with respect to both spatial and temporal noise and does not require any special a priori assumption. While being more robust respect to noise and with higher spatial resolution and CNR when compared to the functional maps, our method is not quantitative and a potential usage in clinical routine could be as a second reader to assist in the maps evaluation, or to guide a dataset smoothing before the modeling part.

Paper Details

Date Published: 9 March 2017
PDF: 8 pages
Proc. SPIE 10132, Medical Imaging 2017: Physics of Medical Imaging, 101320M (9 March 2017); doi: 10.1117/12.2251161
Show Author Affiliations
Francesco Pisana, Deutsches Krebsforschungszentrum (Germany)
Siemens Healthineers GmbH (Germany)
Thomas Henzler, Institute of Clinical Radiology and Nuclear Medicine, Ruprecht-Karls-Univ. Heidelberg (Germany)
Stefan Schönberg, Institute of Clinical Radiology and Nuclear Medicine, Ruprecht-Karls-Univ. Heidelberg (Germany)
Ernst Klotz, Siemens Healthineers GmbH (Germany)
Bernhard Schmidt, Siemens Healthineers GmbH (Germany)
Marc Kachelrieß, Deutsches Krebsforschungszentrum (Germany)

Published in SPIE Proceedings Vol. 10132:
Medical Imaging 2017: Physics of Medical Imaging
Thomas G. Flohr; Joseph Y. Lo; Taly Gilat Schmidt, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?