Comparison among iterative reconstruction techniques for phase-sensitive breast tomography

Introduction Iterative tomographic reconstruction techniques are particularly interesting when the number of acquired projections is not enough to match the Nyquist criteria. These techniques differ in computational time and image quality. Purpose SIRT, SART and CGLS algorithms have been studied in order to evaluate the application of iterative reconstruction methods to low-dose phase-contrast breast tomography, in the framework of the SYRMA-CT project. The number of iterations has been optimized while reducing the number of projections used during the reconstruction. Materials and methods The tomographic data set of two large breast tissue samples (9 cm) has been acquired using a monochromatic X-ray beam at 38 keV. Images were acquired in propagation-based phase-contrast mode and a single-material single-distance phase-retrieval algorithm was applied to the projections before reconstruction. The codes applied are part of the ASTRA Tomography Toolbox. The quality of the images was evaluated using the CNR and few full-reference quality indexes. The reconstruction performed with the filter back-projection including all the acquired projections was considered as gold standard. Results According to full-reference indexes the best quality is achieved when the CNR is similar to the one of the gold standard. The number of iterations can be used to optimize the noise preserving the visibility of the details. Conclusion Full reference indexes require a gold standard acquired at higher statistics, due to their strong dependence on the noise. The evaluation of a team of radiologists is required to compare full-references quality indexes and diagnostic quality of the reconstructed images. Disclosure Nothing to declare.