Click here to

Session: Breast Imaging [Return to Session]

Image Quality Evaluation of a Compressed-Sensing Algorithm for Image Reconstruction in Dedicated Cone-Beam Breast CT Using Offset-Detector

S Vedantham*, H Tseng, A Karellas, University of Arizona, Tucson, AZ


SU-F-TRACK 3-4 (Sunday, 7/25/2021) 4:30 PM - 5:30 PM [Eastern Time (GMT-4)]

Purpose: We have developed a new version of a dedicated, prone, cone-beam breast computed tomography (CBBCT). This system uses a CMOS detector in offset-detector configuration for imaging large breasts, while maintaining an optimal air-gap between the breast and the detector for scatter rejection. For this geometry, image quality achieved with a compressed-sensing (CS) algorithm was compared to Feldkamp-Davis-Kress algorithm with Wang’s weights (FDK-Wang).

Methods: 30 clinical breast CT datasets with calcifications acquired on a CBBCT system (300 views in 360-degrees) with a large-area (40x30-cm), centered, a flat-panel detector with 1024×768 pixel matrix, were selected. The sinograms were truncated to 768×768 and 640×768 -pixels along the fan-angle to emulate offset-detectors. A modified and GPU-enabled version of a CS-based algorithm, referred to as FIRST, and the FDK algorithm with a weighting window (FDK-Wang), were used to reconstruct these truncated sinograms. FDK reconstructions from complete sinograms served as reference. Signal-to-noise ratio (SDNR), bias, variance, root-mean-squared-error (RMSE), and the full-width at half-maximum (FWHM) of calcified lesions in two orthogonal directions were quantified.

Results: The FWHM of calcifications did not differ (P>0.262) among reconstruction algorithms and detector formats, implying comparable spatial resolution. For a chosen offset-detector, the FIRST algorithm outperformed the FDK-Wang in terms of RMSE, bias, variance, and SDNR. For FIRST reconstructions with 768×768 (640×768) pixels, the median RMSE (cm-2), bias (cm-1), variance (cm-2) and SDNR were 6.45E-6 (7.63E-6), 7.17E-4 (9.26E-4), 3.37E-5 (4.29E-5), and 8.5 (6.54), indicating better performance with the 768x768 pixel matrix. The variance and SDNR from FIRST reconstructions with 768x768 pixels outperformed the reference FDK reconstructions from complete sinograms. The reconstruction time varied with breast size (25–45 minutes).

Conclusion: This study indicates the feasibility of using CS-based, FIRST algorithm to reconstruct sinograms from offset-detectors. Our results indicate that FIRST reconstructions from 768×768 pixel matrix achieved the best performance.



    Breast, Cone-beam CT, Reconstruction


    Not Applicable / None Entered.

    Contact Email