Gaussian frequency blending algorithm with Matrix Inversion Tomosynthesis (MITS) and Filtered Back Projection (FBP) for better digital breast tomosynthesis reconstruction


Journal Article

Breast cancer is a major problem and the most common cancer among women. The nature of conventional mammography makes it very difficult to distinguish a cancer from overlying breast tissues. Digital Tomosynthesis refers to a three-dimensional imaging technique that allows reconstruction of an arbitrary set of planes in the breast from limited-angle series of projection images as the x-ray source moves. Several tomosynthesis algorithms have been proposed, including Matrix Inversion Tomosynthesis (MITS) and Filtered Back Projection (FBP) that have been investigated in our lab. MITS shows better high frequency response in removing out-of-plane blur, while FBP shows better low frequency noise prosperities. This paper presents an effort to combine MITS and FBP for better breast tomosynthesis reconstruction. A high-pass Gaussian filter was designed and applied to three-slice "slabbing" MITS reconstructions. A low-pass Gaussian filter was designed and applied to the FBP reconstructions. A frequency weighting parameter was studied to blend the high-passed MITS with low-passed FBP frequency components. Four different reconstruction methods were investigated and compared with human subject images: 1) MITS blended with Shift-And-Add (SAA), 2) FBP alone, 3) FBP with applied Hamming and Gaussian Filters, and 4) Gaussian Frequency Blending (GFB) of MITS and FBP. Results showed that, compared with FBP, Gaussian Frequency Blending (GFB) has better performance for high frequency content such as better reconstruction of micro-calcifications and removal of high frequency noise. Compared with MITS, GFB showed more low frequency breast tissue content.

Full Text

Duke Authors

Cited Authors

  • Chen, Y; Lo, JY; Baker, JA; Dobbins, JT

Published Date

  • June 30, 2006

Published In

Volume / Issue

  • 6142 I /

International Standard Serial Number (ISSN)

  • 1605-7422

Digital Object Identifier (DOI)

  • 10.1117/12.652264

Citation Source

  • Scopus