SUMMARY

Recently, dedicated cone beam breast CT (CBBCT) has been approved by FDA to provide three-dimensional images for diagnostic breast imaging. Nevertheless, high scatter contamination stemming from large irradiation volume results in severe contrast lost and shading artifacts, impeding its quantitative uses in certain clinical tasks. Existing scatter correction methods demonstrate different drawbacks including low efficacy, dose or scan time increase, etc. In this work, we propose two scatter correction methods, library based and forward projection model based, to overcome the deficiencies while achieving high correction efficacy. In the library based method, a scatter library is precomputed via Monte Carlo simulation based on a simple breast model. Due to the relatively simple shape and composition, we find that a small library size with one input parameter of breast size is sufficient for effective scatter correction on general population. In the forward projection model based method, we first estimate primary signals of CBBCT projections via forward projection of the segmented first-pass reconstruction. By subtracting the simulated primary projection from the raw projection, we obtain a raw scatter estimate containing both low-frequency scatter and errors. After discarding untrusted errors from the resultant raw scatter map, the final scatter is obtained via a novel Fourier-Transform based local filtration algorithm. Both methods have demonstrated high correction efficacy on patient data, the library-based method is superior in computational efficiency while the forward projection model based method is more flexible.
Image-domain scatter correction methods are highly efficient and require no access to projection data. A second goal of this work is to improve the current image-domain methods for CBBCT scatter correction on both efficacy and computational efficiency. The proposed algorithm is inspired by an image-domain method for improving cone-beam CT in radiotherapy using high quality planning CT(pCT) as prior information. In diagnostic CBBCT imaging where the pCT is not available, we propose to first assign each voxel on CBBCT image three tissue probability indexes using a Fuzzy-C-Mean algorithm, the scatter errors induced by each tissue type are estimated by a weighted lowpass filtration. Preliminary results with two CBBCT patient data show that the method is effective on improving the uniformity of CBBCT image.