Timothy Pok Chi Yeung BSC, Presenter: Nothing to Disclose

Nathan De Haan, Abstract Co-Author: Nothing to Disclose

Mark Dekaban, Abstract Co-Author: Nothing to Disclose

Laura Morrison, Abstract Co-Author: Nothing to Disclose

Lisa Hoffman, Abstract Co-Author: Nothing to Disclose

Slav Yartsev, Abstract Co-Author: Nothing to Disclose

Glenn Stuart Bauman MD, Abstract Co-Author: Research Grant, sanofi-aventis Group

Ting-Yim Lee MSc, PhD, Abstract Co-Author: Grant, General Electric Company Royalties, General Electric Company

Many CT perfusion (CTP) studies of small animal tumor models are performed using clinical CT scanner due to its availability, but the tradeoff between spatial resolution and image noise affects the quality of CT perfusion images. This study aimed to evaluate the ability of principal component analysis (PCA) in improving the contrast-to-noise ratio (CNR) of CTP images in a preclinical model of malignant glioma.

Wistar rats (n = 8) implanted with C6 glioma cells were scanned using CTP. Each CTP image set was filtered using 2, 4, 6, 8, and 10 principal components from PCA to result in 40 additional image sets. The noise level and CNR were used to quantify image quality in all 48 unfiltered and filtered image sets. The fractional residual information (FRI) was used to evaluate the amount of information loss after PCA filtering. Blood flow (BF), blood volume (BV), and permeability-surface area product (PS) before and after filtering were calculated. Noise level, CNR, BF, BV, and PS in the normal brain and tumor were expressed as mean ± standard error of the mean. These metrics before and after filtering with different numbers of principal components were compared to evaluate the differences between the filtered and the unfiltered image sets.

PCA filtering significantly decreased noise level and increased CNR (p = 0.01). An average of 26% (range, 11 – 49%) of pixels in the tumor had information loss of ≥ 5% when filtering with only two principal components; this percentage decreased to an average of 1% (range, 0 – 3%) with four or more components. Normal brain BV and PS were significantly different than the values in the tumor (p < 0.01) without or with PCA filtering (using 4 or more principal components). Normal brain and tumor BF values were not significantly different without PCA filtering, but they became significantly different after filtering with 4 principal components (p = 0.03).

PCA filtering improved the CNR in CTP studies. Four or more principal components are required to filter the CTP source images without substantial loss of information leading to higher contrast between tumor and normal brain tissue in BF maps.

Lowering radiation exposure can lead to deterioration of CNR in CTP studies. PCA improves CNR to allow repeated ultralow dose CTP studies for assessing treatment response in the clinical setting.

Yeung, T, De Haan, N, Dekaban, M, Morrison, L, Hoffman, L, Yartsev, S, Bauman, G, Lee, T, Improving CT Perfusion Image Quality Using Principal Component Analysis.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13017426.html