Arkadiusz Sitek PhD, Presenter: Nothing to Disclose

Chun-Shan Yam PhD, Abstract Co-Author: Nothing to Disclose

Robert Glenn Sheiman MD, Abstract Co-Author: Nothing to Disclose

Vassilios D. Raptopoulos MD, Abstract Co-Author: Nothing to Disclose

The quality of the perfusion images created by dynamic CT is poor due to noise and inconsistencies between time frames due to patient motion. To improve the quality of perfusion maps we developed a method that (i) allows for correction of patients motion and, (ii) reduces noise in the dynamic image with no loss in image resolution.

Six patients with known pelvic malignancy underwent dynamic contrast-enhanced helical CT imaging at a single level with a region of interest (ROI) 128x128 pixels applied to tumor for the creation of a perfusion maps. Motion correction was performed by an automatic adjustment of the ROI location carried out using an image matching algorithm. The algorithm minimizes the difference in the ROI placement between the first image at time zero and subsequent images to yield a target time-enhancement curve (TEC). We assumed that there is only a rigid movement in x, y, and z direction. Pixels in the ROI were grouped into clusters using average linkage method and time behaviors of pixels within clusters were averaged. Values of tumor blood perfusion were calculated for each cluster making use of the maximum slope of the TEC and dividing by peak enhancement of an ROI placed on the arterial input. Values of the perfusion calculated for each cluster were then remapped back to original image using non-negative least square fitting.

Correction for abdominal movement removed artifacts from perfusion image that were present especially for the 2/6 patients in the prone position. We found by visual inspection that perfusion images obtained by our method had similar resolution as original CT images, and much better resolution than perfusion images obtained on a commercial GE Advantage workstation. The noise (sample standard deviation σ=7.4 HU) in the perfusion images calculated by our method decreased significantly compared to perfusion maps calculated using pixel-by-pixel approach (σ=34.9, p<0.001), and to pixel-by-pixel approach after data were smoothed (σ=22.9,p<0.001).

The methods presented improve the quality of tumor perfusion maps obtained by dynamic CT studies and may allow for improved assessment of tumor response to treatment.

Sitek, A, Yam, C, Sheiman, R, Raptopoulos, V, Formation of Parametric Perfusion Images in Dynamic CT.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4415737.html