Purpose: Wide-angle digital breast tomosynthesis (DBT) generally allows an increased contrast and a better depth-separation of the overlapping tissues compared to the narrow-angle DBT. However, wide-angle DBT requires a substantially longer scan time, which may not only increase the patient discomfort but also cause patient motion. Fast rotation of the gantry during the scan would therefore desirable in a wide-angle scan, but can cause source motion blur image artifacts. In this work, we address this problem and propose a deblurring method.
Methods: We numerically modeled a DBT scan system and varied the effective size of the source to estimate source motion blur. We used VICTRE breast phantoms and obtained paired data which consist of reconstructed images from the ideal point source and a moving source. To correct for the source motion blur, we implemented an unfolding non-blind deblur network which leverages both learning-based methods for deblurring and model-based methods for denoising.
Results: Our system model showed that the effect of source motion is highly dependent on the axial position of the reconstructed object components; the effect of source blur decreases for the point which is closer to the rotation center, and vice versa. Our physics-inspired deblur network reduces the blurring effect qualitatively. Quantitate analysis on the FWHM of calcification-like structure also showed the improvement.
Conclusion: In this study, we designed a numerical study for simulating the effect of source motion on the wide-angle DBT images. In addition, we implemented a deep unfolding network for reducing the blur artifacts. We believe that our work is potentially valuable in regard to reducing wide-angle DBT scan time.
Breast, Image Artifacts, Image Processing
IM- Breast X-Ray Imaging: Digital Breast Tomosynthesis (DBT)