Noise power spectrum analysis for several digital breast tomosynthesis reconstruction algorithms
Digital breast tomosynthesis is a three-dimensional imaging technique that allows the reconstruction of an arbitrary set of planes in the breast from limited-angle series of projection images. Though several tomosynthesis algorithms have been proposed, no complete optimization and comparison of all available methods has been conducted as of yet. This paper presents an analysis of noise power spectrum to examine the noise characteristics of several tomosynthesis algorithms with different imaging acquisition techniques. Flat images were acquired with the following acquisition parameters: 13, 25, 49 projections with ±12.5 and ±25 degrees of angular ranges. Three algorithms, including Shift-And-Add (SAA), Matrix Inversion Tomosynthesis (MITS), and Filtered Back Projection (FBP) were investigated with reconstruction slice spacing of 1mm, 2mm, and 4mm. The noise power spectra of the reconstruction plane at 23.5mm above the detector surface were analyzed. Results showed that MITS has better noise responses with narrower slice spacing for low-to-middle frequencies. No substantial difference was noticed for SAA and FBP with different slice spacings. With the same acquisition technique and slice spacing, MITS performed better than FBP at middle frequencies, but FBP showed better performance at high frequencies because of applied Hamming and Gaussian low-pass filters. For different imaging acquisition techniques, SAA, MITS and FBP performed the best with 49 projections and ±25 degrees. For 25 projections specifically, FBP performed better with wider angular range, while MITS performed better with narrower angular range. For SAA, narrow angular range is slightly better for 25 projections and 13 projections.
