Purpose: X‐ray differential phase contrast (DPC) projection imaging has shown promise for superior image contrast. In the field of medical imaging, however, its differential nature makes the projection image difficult to use for direct diagnosis. In this study, a phase retrieval method is introduced to generate phase images, which are compared with absorption images to demonstrate the benefits of phase contrast projection imaging. Methods: The method involves the minimization of a cost function that includes two terms: a quadratic norm that enforces consistency between the phase image and the DPC image, and a nonconvex norm of the phase imageˈs gradient in the direction perpendicular to the stripes. A weighting parameter α is applied to the second term to tailor the content of stripes. This cost function is minimized to solve for the phase image. A grating‐based DPC system was used to acquire the DPC and absorption images of a cylindrical phantom. The DPC, absorption, directly‐integrated phase and regularized phase images were compared. To evaluate the image quality, simulated spheres of different contrasts and diameters were added to the experimental data. CNR of the beads in the integrated image was used as the figure of merit to optimize the regularization. Results: The retrieved phase image provides complementary information of the phantom that is otherwise lost in the direct‐integrated images because of the presence of stripe noise. Regularized integration improved image quality in terms of CNR and the content of stripe noise. The optimal regularization parameter can be determined from the peaks of the CNR‐α curves and is dependent on the objectˈs contrast. Conclusions: Regularized integration allows for artifact‐free phase retrieval from DPC images and a fair comparison between phase contrast and absorption images. DPC imagingˈs potential application in medical imaging is highlighted by the complementary information provided by the phase image.

Purpose: To develop a high quality 4D cone beam CT (4DCBCT) method that is immune to patient/couch truncations and to investigate its application in adaptive replanning of lung XRT. Methods: In this study, IRB‐approved human subject CBCT data was acquired using a Varian on‐board imager with 1 minute rotation time. The acquired projection data was retrospectively sorted into 20 respiratory phase bins, from which 4DCBCT images with high SNR and high temporal resolution were generated using Prior Image Constrained Compressed Sensing (PICCS). Couch and patient truncations generate strong data inconsistency in the projection data and artifacts in the 4DCBCT image. They were addressed using an adaptive PICCS method. The artifact‐free PICCS‐4DCBCT images were used to generate adaptive treatment plans for the same patient at the 10th (day 21) and 30th (day 47) fractions. Dosimetric impacts with and without PICCS‐ 4DCBCT were evaluated by isodose distributions, DVHs, and other dosimetric factors. Results: The adaptive PICCS‐4DCBCT method improves image quality by removing residue truncation artifacts; measured universal image quality increased 37%. The isodose lines and DVHs with PICCS‐4DCBCT‐based adaptive replanning were significantly more conformal to PTV than without replanning due to changes in patient anatomy caused by progress of the treatment. The mean dose to PTV at the 10th fraction was 63.1Gy with replanning and 64.2Gy without replanning, where the prescribed dose was 60Gy, in 2Gy × 30 fractions. The mean dose to PTV at the 30th fraction was 61.6Gy with replanning and 64.9Gy without replanning. Lung V20 was 37.1%, 41.9% and 43.3% for original plan, 10th fraction plan and 30th fraction plan; with re‐planning, Lung V20 was 37.1%, 32%, 27.8%. Conclusion: 4DCBCT imaging using adaptive PICCS is able to generate high quality, artifact‐free images that potentially can be used to create replanning for improving radiotherapy of the lung. K Niu, K Li, J Smilowitz: Nothing to Disclose. G Chen: General Electric Company Research funded, Siemens AG Research funded, Varian Medical Systems Research funded, Hologic Research funded.

Purpose: Digital breast tomosynthesis (DBT) is an emerging breast imaging modality. It uses x‐ray absorption projection images acquired over a narrow angular span to provide high spatial resolution pseudo‐three dimensional images for breast cancer screening and diagnosis. Unfortunately, the narrow range used to acquire data (typically <30 degrees) results in significant out of plane artifacts for high contrast objects when conventional reconstruction techniques are used. In this work, the denoised ordered‐subset statistically penalized algebraic reconstruction technique (DOS‐SPART) algorithm was adapted to breast tomosynthesis imaging to reduce out‐of‐plane blurring artifacts. Methods: The DOS‐SPART algorithm was implemented for use with DBT datasets. The ACR mammography accreditation phantom was imaged for quantitative measurement and subjective image analysis. The artifact spread function (ASF) was measured for images generated by the commercial reconstruction method as well as DOS‐SPART with different total number of iterations (5, 10, 15, and 20). The FWHM of the ASF was used to quantify out‐of‐plane blurring. A cluster of calcifications in the ACR phantom was also inspected subjectively, and images above and below the feature were examined for artifacts. Results: The DOS‐SPART algorithm was able to efficiently reconstruct sharp image at the in‐focus plane, while significantly reducing out‐of‐plane blurring artifacts (total reconstruction time <90 seconds for a 1996×2457×70 voxel volume with 5 iterations). The FWHM of the ASF was reduced by as much as 35% using the algorithm. Subjectively, the out‐of‐plane artifacts from the high contrast calcification are still clearly visible in the commercial reconstruction at 1 cm above or below the in‐focus plane, whereas no artifacts from the calcification are detectable in the out‐of‐plane images generated with DOS‐SPART. Conclusion: The application of DOS‐SPART to DBT can help limit out‐of‐plane artifacts in DBT imaging, potentially improving localization of microcalcifications and image contrast for adjacent structures. Funding support: This work was support in part by NIH R01 EB020521. Disclosures: J Garrett: None. Y. Li: None. K. Li: None. GH Chen: Research funded, GE Healthcare and Siemens AX

Purpose: The anatomical noise power spectra (NPS) for differential phase contrast (DPC) and dark field (DF) imaging have recently been characterized using a power‐law model with two parameters, alpha and beta, an innovative extension to the methodology used in x‐ray attenuation based breast imaging such as mammography, DBT, or cone‐beam CT. Beta values of 3.6, 2.6, and 1.3 have been measured for absorption, DPC, and DF respectively for cadaver breasts imaged in the coronal plane; these dramatic differences should be reflected in their detection performance. The purpose of this study was to determine the impact of anatomical noise on breast calcification detection and compare the detection performance of the three contrast mechanisms of a multi‐contrast x‐ray imaging system. Methods: In our studies, a calcification image object was segmented out of the multi‐contrast images of a cadaver breast specimen. 50 measured total NPS were measured from breast cadavers directly. The ideal model observer detectability was calculated for a range of doses (5–100%) and a range of calcification sizes (diameter = 0.25–2.5 mm). Results: Overall we found the highest average detectability corresponded to DPC imaging (7.4 for 1 mm calc.), with DF the next highest (3.8 for 1 mm calc.), and absorption the lowest (3.2 for 1 mm calc.). However, absorption imaging also showed the slowest dependence on dose of the three modalities due to the significant anatomical noise. DPC showed a peak detectability for calcifications ∼1.25 mm in diameter, DF showed a peak for calcifications around 0.75 mm in diameter, and absorption imaging had no such peak in the range explored. Conclusion: Understanding imaging performance for DPC and DF is critical to transition these modalities to the clinic. The results presented here offer new insight into how these modalities complement absorption imaging to maximize the likelihood of detecting early breast cancers. J. Garrett, Y. Ge, K. Li: Nothing to disclose. G.‐H. Chen: Research funded, GE Healthcare; Research funded, Siemens AX.

positive correlation between TCBF and CVO (r=0.81 and P<0.001) was seen. The TCBF (20.21 ±4.58 ml/s versus 11.78±2.03 ml/s; P<0.001) and CVO (12.80 ± 3.82 ml/s versus 9.03 ±2.31 ml/s; P=0.010) were significantly higher in children compared to adult volunteers. The CVO/TCBF ratio was significantly lower in children versus adult volunteers (0.63 ± 0.01 versus 0.76 ± 0.02, P=0.025). In adults, the correlation of TCBF with age remains strong (rho =-0.69, t-stat =-4.5, P=0.00018). However, CVO (rho =-0.29, t-stat =-1.42, P=0.171) and CVO/TCBF ratio (r=0.16, P=0.446) were not significantly associated with age in the adult cohort. The ratio of cerebral arterial inflow to systemic aortic outflow was significantly higher in children compared to adults (0.45±0.08 versus 0.15 ±0.02, P<0.001). Conclusions Both TCBF and CVO decrease with age, however unlike TCBF, there is no correlation between the decrease in CVO through the Transverse sinuses and age, which could suggest the early development of alternative venous drainage pathways through the emissary and extracranial veins. This could also explain the differential ratio of CVO to TCBF, which suggests that more than 20% of cerebral venous outflow in adults and more than 35% of outflow in Children are not through the Transverse sinuses in the supine position. Understanding the quantitative differences between TCBF and CVO in healthy volunteers could help identify and manage changes related to venous outflow abnormalities.