Ronny Hendrych, Presenter: Employee, Artemis Imaging GmbH

Marcel Beister, Abstract Co-Author: Employee, Artemis Imaging GmbH

Willi A. Kalender PhD, Abstract Co-Author: Consultant, Siemens AG Consultant, Bayer AG Founder, CT Imaging GmbH Scientific Advisor, CT Imaging GmbH Shareholder, CT Imaging GmbH Founder, Artemis Imaging GmbH CEO, Artemis Imaging GmbH Shareholder, Artemis Imaging GmbH

In breast cancer diagnosis it is important to both identify high-contrast micro-calcifications and to detect small low-contrast lesions. In low-dose scans, calcifications can be identified due to their high signal-to-noise ratio, whereas detectability of soft-tissue lesions is a more challenging task. In order to eliminate multiple reconstructions for different tasks and for improvement of low-contrast visibility, four 3D filtering techniques were evaluated.

We simulated breast phantoms at different resolution with dose levels from 1.5 up to 6 mGy. The raw data was reconstructed using filtered back projection (FBP) in a high-resolution mode for micro-calcifications (µCa) with (75 µm)3 and a standard resolution mode using a soft reconstruction kernel for soft-tissue lesions (STL) with (150 µm)3 voxel size. All volumes were filtered using a standard boxed filter with different kernel sizes as reference. In addition we used the following filter methods for the reconstructed volumes: 3D Gaussian, 3D Median and an iterative 3D Impulse Detector using weighted Median. The linear filter methods were interactively applied in real-time, which was not possible for the Median and the Impulse Detector due to the higher computational demands. Furthermore, we evaluated the possibility to apply the Gaussian filter to the µCa volumes in order to achieve comparable results as with the filtered STL data sets.

Low contrast detectability was significantly enhanced with all filter methods and allowed for the detection of lesions of 2 mm diameter and larger in the STL volumes. The noise levels were reduced by 76% up to 87%. The iterative method and the median produced comparable results than the linear filters, but at the cost of much higher computational effort. It was possible to transform µCa volumes into their STL comparable counter parts via Gaussian filtering.

Low-contrast detectability can be enhanced substantially with common filter techniques. Interactive Gaussian filtering allows adjusting the trade-off between image noise and spatial resolution on large µCa volumes (20483) in real-time and may make the separate reconstruction of the STL volumes obsolete in the future.

Using standard filter techniques on reconstructed image volumes of a dedicated breast CT scanner at dose levels accepted for breast screening can significantly improve soft-tissue detectability.

Hendrych, R, Beister, M, Kalender, W, Comparison of Different Filtering Techniques for Enhancement of Low-Contrast Detectability in Low-Dose Isotropic Data of a Dedicated Breast CT Scanner.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11034612.html