Lifeng Yu PhD, Presenter: Nothing to Disclose

Shuai Leng PhD, Abstract Co-Author: Nothing to Disclose

Armando Manduca PhD, Abstract Co-Author: Nothing to Disclose

Joel Garland Fletcher MD, Abstract Co-Author: Research grant, Siemens AG

Cynthia H. McCollough PhD, Abstract Co-Author: Research grant, Siemens AG

To characterize the noise properties of two noise reduction methods and compare them with a filtered back-projection (FBP)-based reconstruction on a clinical CT scanner using local 3D noise power spectrum (NPS) analysis.

Noise reduction methods, including various projection- and image-space denoising, have been increasingly used in clinical CT due to the growing need in reducing radiation dose. These methods can reduce the noise while maintaining the high-contrast resolution. However, a different noise texture (blocky and pixelized) is usually introduced by these methods, which may have a negative impact on the low-contrast performance of the images. NPS is a more complete descriptor of noise properties than noise magnitude alone. The use of NPS in CT has to be in a small local region because CT noise is generally non-stationary. We used a local NPS analysis to evaluate an image-space denoising, a projection-space denoising, and an FBP-based reconstruction. The projection-space denoising incorporated a CT noise model in the projection data, which was recently developed by us. The image-space denoising was a type of 3D adaptive filter. A 30 cm water cylinder was scanned repeatedly on a 128-slice scanner to obtain an ensemble of noisy images. Both image- and projection-space denoising methods were applied to generate denoised datasets. NPS were calculated at a small ROI in five different locations within both axial and sagittal planes.

NPS varied across the FOV. All the NPS in axial plane from the FBP reconstruction had the same ramp shape at low frequencies and different peak frequencies (B10: 1.6 lp/cm; B20:1.9 lp/cm; B30:2.2 lp/cm, B40:2.7 lp/cm). When the noise level was matched to that of B20, the image-space denoising generated a flat spectrum, while the projection-space denoising maintained the shape of the spectrum in FBP reconstruction with a peak frequency similar to that of a B40 kernel.

Local 3D NPS was used to characterize the noise properties in three image reconstruction and denoising methods. Image-space denoising (a 3D adaptive filter) changed the noise texture dramatically, while sinogram-space denoising maintained the noise texture.

Local 3D noise power spectrum (NPS) is a more complete descriptor of noise properties of a CT image. NPS can be used to characterize the noise texture changes introduced by various denoising methods.

Yu, L, Leng, S, Manduca, A, Fletcher, J, McCollough, C, Local 3D Noise Power Spectrum Analysis of Filtered-backprojection Reconstruction: Projection- and Image-Space Denoising in Multislice CT.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9008678.html