Oleg S. Pianykh, Presenter: Nothing to Disclose

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

Sofia Gourtsoyianni MD, Abstract Co-Author: Nothing to Disclose

Reducing CT radiation benefits the patients, but downgrades image quality. Although proprietary filters from CT vendors can denoise raw scanner data, they are hard to implement consistently in multivendor hospital environments. The purpose of our work was to develop a vendor-independent DICOM image filter, restoring high-dose diagnostic image quality from low-dose CT.

Low-dose image quality restoration was considered as a mathematical regularization problem, affected by several important filter parameters: - Filter support (pixel neighborhood) size - Fuzzy filter intensity thresholding (intensity similarity measure) - Fuzzy filter distance thresholding (spatial similarity measure) - Filter regularization coefficient (strength of filtering). - Edge preservation weighting - Number of iterations To determine the optimal filtering parameters, we developed new image quality metric, matching noise histograms from filtered images to those of the high-dose. Low-dose, high-dose, and filtered low-dose images were obtained from the 1.25 mm scans of 4kg ham sample, taken at 12 radiation levels, gradually decreasing by 10% (thus corresponding to 3X dose reduction at the lowest level). The resulting images were processed to find the optimal filter settings, and the optimized filter was then applied to denoise low-dose patient scans from two different vendors, available at our facility. The results of this filtering were assessed in a blind test, performed by 3 trained radiologists.  

Filters with optimal parameters significantly improved visual image quality. In thick slices, the quality of filtered lowest-dose images (3X dose reduction) was approaching to that of the high-dose images. The blind image comparison test scored our filters to perform at least as good (or better) as proprietary raw-data scanner-side filtering (p<0.001). Optimized filter implementation achieved 5 images/second processing speed on a plain off-the-shelf laptop, making it practical for real-time applications.

Nonlinear DICOM image filtering can be efficiently applied to reduce CT radiation dose by up to 50%, in real-time, on an average computer.

Low-dose CT denoising can be efficiently implemented in the image domain, to be used for vendor-independent CT dose reduction.

Pianykh, O, Raptopoulos, V, Gourtsoyianni, S, Nonlinear Edge-preserving DICOM Image Filters for CT Dose Reduction.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11034490.html