Patrik Rogalla MD, Presenter: Nothing to Disclose

Bernice E. Hoppel PhD, Abstract Co-Author: Employee, Toshiba Corporation

Sonja Kandel MD, Abstract Co-Author: Nothing to Disclose

Hatem Mehrez PhD, Abstract Co-Author: Employee, Toshiba Corporation

Narinder S. Paul MD, Abstract Co-Author: Research funding, Toshiba Corporation

To analyse whether raw-data based rapid iterative denoising algorithm (AIDR3D) improves image quality by reducing image pixel noise, needle artefacts, and by improving CNR in CT fluoroscopy.

Raw-data from 21 patients who underwent diagnostic and therapeutic interventions in the chest and abdomen performed under CT fluoroscopy guidance (Aquilion 64), were reconstructed with filtered back projection (FBP) and AIDR3D either as twin images or every 83 ms to create dynamic images for needle position control. The iterative denoising algorithm (AIDR3D) uses an adaptive method based on raw counts and a statistical model of the system. The image based component was omitted to maintain real-time reconstruction speeds. CT number was determined in aorta (Ao), muscle tissue (Ms). SD of pixel noise (σ) was measured in air. CNR was determined as (Ms-Ao)/σ. The artefacts from the needle were calculated as follows: 1) hyper-enhancement blooming (Bl) from the needle was measured as a line profile near the tip of the needle. The profile was fit to a Gaussian curve, the FWHM and max intensity were determined, 2) the black streak (BS) at the needle tip was measured as mean and SD.

CT attenuation values remained unchanged (p=0.063 for Ao, p=0.19 for Ms), however the standard deviation was significantly higher with FBP than with AIDR3D (Ao = 64 and 32 p<0.001, Ms = 67 and 33, p=0.001). The CNR ratio increased significantly with the use of AIDR3D from 0.68 +/- 0.61 to 0.99 +/- 0.82 (p<0.001). The blooming artefacts showed a slightly wider FWHM in AIDR3D (3.3 +/- 1.14) compared with FBP (3.0 +/- 0.83, p=0.015); the max intensity Bl significantly decreased with AIDR3D (FBP: 2176 HU, AIDR3D: 1639 HU, p<0.001). The dark streak artefact (BS) appeared to slightly decrease in intensity but increase in SD (FBP: -554 +/- 168, AIDR3D: -515 +/- 193), however this was statically not significant (mean, p=0.15 and SD, p=0.06).

Statistical noise reduction modelling improves image characteristics by reducing pixel noise and improving CNR, potentially facilitating radiation dose reduction. Further data processing appears necessary to decrease artefacts and improve conspicuity of the needle tip.

Iterative denoising algorithm may help visualise anatomy and pathology in CT fluoroscopy while reducing radiation dose.

Rogalla, P, Hoppel, B, Kandel, S, Mehrez, H, Paul, N, Statistical Noise-model Dose Reduction for Real-Time Reconstruction in CT Fluoroscopy.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12035415.html