C Lai

A post-reconstruction method to correct cupping artifacts in cone beam breast computed tomography

In cone beam breast computed tomography (CT), scattered radiation leads to nonuniform biasing of CT numbers known as a cupping artifact. Besides being visual distractions, cupping artifacts appear as background nonuniformities, which impair efficient gray scale windowing and pose a problem in threshold based volume visualization/segmentation. To overcome this problem, we have developed a background nonuniformity correction method specifically designed for cone beam breast CT. With this technique, the cupping artifact is modeled as an additive background signal profile in the reconstructed breast images. Due to the largely circularly symmetric shape of a typical breast, the additive background signal profile was also assumed to be circularly symmetric. The radial variation of the background signals was estimated by measuring the spatial variation of adipose tissue signals in front view breast images. To extract adipose tissue signals in an automated manner, a signal sampling scheme in polar coordinates and a background trend fitting algorithm were implemented. The background fits compared with targeted adipose tissue signal value (constant throughout the breast volume) to get an additive correction value for each tissue voxel. To test the accuracy, we applied the technique to cone beam CT images of mastectomy specimens. After correction, the images demonstrated significantly improved signal uniformity in both front and side view slices. The reduction of both intraslice and interslice variations in adipose tissue CT numbers supported our observations.

Cone-beam CT breast imaging with a flat panel detector - A simulation study

This paper investigates the feasibility of using a flat panel based cone-beam computer tomography (CT) system for 3-D breast imaging with computer simulation and imaging experiments. In our simulation study, 3-D phantoms were analytically modeled to simulate a breast loosely compressed into cylindrical shape with embedded soft tissue masses and calcifications. Attenuation coefficients were estimated to represent various types of breast tissue, soft tissue masses and calcifications to generate realistic image signal and contrast. Projection images were computed to incorporate x-ray attenuation, geometric magnification, x-ray detection, detector blurring, image pixelization and digitization. Based on the two-views mammography comparable dose level on the central axis of the phantom (also the rotation axis), x-ray kVp/filtration, transmittance through the phantom, detected quantum efficiency (DQE), exposure level, and imaging geometry, the photon fluence was estimated and used to estimate the phantom noise level on a pixel-by-pixel basis. This estimated noise level was then used with the random number generator to produce and add a fluctuation component to the noiseless transmitted image signal. The noise carrying projection images were then convolved with a Gaussian-like kernel, computed from measured 1-D line spread function (LSF) to simulated detector blurring. Additional 2-D Gaussian-like kernel is designed to suppress the noise fluctuation that inherently originates from projection images so that the reconstructed image detectability of low contrast masses phantom can be improved. Image reconstruction was performed using the Feldkamp algorithm. All simulations were performed on a 24 PC (2.4 GHz Dual-Xeon CPU) cluster with MPI parallel programming. With 600 mrads mean glandular dose (MGD) at the phantom center, soft tissue masses as small as 1 mm in diameter can be detected in a 10 cm diameter 50% glandular 50% adipose or fatter breast tissue, and 2 mm or larger masses are visible in a 100% glandular 0% adipose breast tissue. We also found that the 0.15 mm calcification can be detected for 100μm detector while only 0.2 μm or above are visible for 200 μm detector. Our simulation study has shown that the cone-beam CT breast imaging can provide reasonable good quality and detectability at a dose level similar to that of two views\mammography. For imaging experiments, a stationary x-ray source and detector, a step motor driven rotating phantom system was constructed to demonstrate cone beam breast CT image. A breast specimen from mastectomy and animal tissue embedded with calcifications were imaged. The resulting images show that 355-425 μm calcifications were visible in images obtained at 77 kVp with a voxel size of 316 μm and a center dose of 600 mrads. 300-315 μm calcifications were visible in images obtained at 60 kVp with a voxel size of 158 μm and a center dose of 3.6 rads.

Dual-energy digital mammography for calcification imaging: theory and implementation

Small microcalcifications essential to the early detection of breast cancer may be obscured by overlapping tissue structures. Dual-energy digital mammography (DEDM), where separate low- and high-energy images are acquired and synthesized to cancel the tissue structures, may improve the ability to detect and visualize microcalcifications. The investigation of DEDM began with a signal-to noise ratio analysis to estimate and relate the noise level in the dual-energy calcification signals to the x-ray spectra, microcalcification size, tissue composition and breast thickness. We investigated various inverse-mapping functions, both linear and non-linear, to estimate the calcification thickness from low- and high-energy measurements. Transmission (calibration) measurements made at two different kVp values for variable aluminum thickness (to simulate calcifications) and variable glandular-tissue ratio for a fixed total tissue thickness were used to determine the coefficients of the inverse-mapping functions by a least-squares analysis. We implemented and evaluated the DEDM technique under narrow-beam geometry. Phantoms, used in the evaluation, were constructed by placing different aluminum strips over breast-tissue-equivalent materials of different compositions. The resulting phantom images consisted of four distinct regions, each with a different combination of aluminum thickness and tissue composition. DEDM with non-linear inverse-mapping functions could successfully cancel the contrast of the tissue-structure background to better visualize the overlapping aluminum strip. We are currently in the process of translating our DEDM techniques into full-field imaging. We have designed special phantoms with variable glandular ratios and variable calcification thicknesses for evaluation of the full-field dual-energy calcification images.

SU‐FF‐I‐41: Accuracy and Computing Time of a Ray‐Driven Projector/back‐Projector for Simulation and Reconstruction in Tomosynthesis and Cone Beam CT Imaging

Accurate and fast projector/back‐projector (PBP) is the key to successful simulation and iterative reconstruction of tomosynthesis and cone beam CT. In this study, the accuracy and computation speed of a ray‐driven PBP operator were investigated. To simulate x‐ray tomographyimaging, a ray‐driven projector was developed and implemented on a 64 computing node PC Cluster. Each x‐ray path is represented by sampling points in the object. Their μ values are summed up to compute the integral attenuation along the path. To evaluate the accuracy of the reprojection algorithm as the function of sampling ratio (Sampling length / voxel length), reprojections obtained with a significantly smaller sampling ratio were used as reference. To minimize the additional loss of accuracy from pixelization, the pixel size of the projection images was selected to be one third of the voxel size projected back to the image plane. Error images were formed by subtracting the re‐projection from the reference re‐projection and used to compute the percentage errors. SART were implemented for image reconstruction in tomosynthesis and CBCTimaging. To evaluate its accuracy, CBCTimagesreconstructed with 300 projection views over 360 degree were compared to the original CBCTimages. Errors were computed from the subtraction images and plotted together the computing time as the function of the sampling ratio. With reduced sampling ratio, the accuracies of both the re‐projection and SART algorithms were improved at the expense of longer computing time. The improvement was accelerated with smaller sampling ratios. Aliasing artifacts were visible when the sampling ratios were greater than 0.5. We have demonstrated that the accuracy of the re‐projection and SART reconstructions improved with reduced the sampling ratio at the expense of longer computing time for a ray‐driven PBP. The tradeoff between accuracy and computing time should be determined by the imaging requirement.

SU‐FF‐I‐15: Effects, Detection and Removal of Zingers From Scattered X‐Rays in CCD Based Cone Beam CT

Purpose: Zingers are tiny spurious white dots that appear randomly in CCDimages. In order to improve the quality of CCD based cone beam CT technique, a new technique for the detection and removal of zingers is described and evaluated. Method and Materials: A bench top CCD based cone beam CT system was used to measure and investigate the presence of zingers. The cause and effects of zingers were studied. A new technique was developed to detect and correct the zingers. With this technique, the statistical behavior of pixel values in a projection image was first analyzed to identify candidates for zingers. Pixel values at the detected zinger locations were then compared in two consecutive projection views to eliminate false detections. To investigate and evaluate this technique, zingers were simulated by increasing the pixel values at randomly selected locations in projection data computed for a modified Shepp‐Logan phantom. The simulated data were then detected and corrected for zingers and used for reconstruction. The resulting reconstructed image was compared with the imagereconstructed from zinger free data and with imagesreconstructed from data corrected using three other zinger removal techniques. Results: Our measurement indicated that zingers may have resulted from scattered x‐rays. They were found to generate visible artifacts and degrade the quality of reconstructed images. It was shown that zingers detection by comparing two identically acquired projections could be highly effective but impractical in CTimaging. Detection by comparing two consecutive projection views was equally effective but may be subject image blurring. Detection by analyzingsignal fluctuations could result in a large number of faulty detections. The proposed new detection technique was found to be practical and effective without resulting in image blurring or faulty detections. This work was supported in part by a research grant CA104759 from NIHNCI.

MO-E-I-609-02: Imaging Properties of Cone Beam Breast CT- Effects of Detector Properties and Imaging Conditions

Purpose: To investigates the effects of detector properties and imaging conditions on the imaging properties of cone‐beam breast CT with both computer simulations and imaging experiments. Method and Materials: Cone beam breast CT was simulated with the breast analytically modeled as cylinder embedded spherical shape soft tissue masses and calcifications. X‐ray spectrum, breast attenuation, geometric magnification, focal spot blurring, x‐ray detection,detector blurring, image pixelization and digitization were all incorporated in computing the projection images.Quantum noise, system noise,detector blurring were also simulated and incorporated in the model. Image filtering and reconstruction were then performed using the Feldkamp algorithm. Simulation was performed for two flat‐panel detectors, one CsI based and the other a‐Se based. Images of phantoms and breast specimens were also obtained to demonstrate the ability of our experimental cone beam breast CT system to image the 3‐D structures of the breast with embedded cancers and calcifications. Results: Our simulation results shows that the a‐Se detector performs slightly better at 30 and 40 kVps while the CsI detector performs better at 50 or higher kVps. ImageSNRs are optimized at 50 and 60 kVp for the s‐Se and CsI detector, respectively. Phantom images obtained with our experimental system show that with higher dose and smaller pixel size, calcifications as small as could be resolved. Images of breast specimens show excellent separation between glandular and adipose tissues. The speculated nature of the tumor masses can be clearly seen in selected projection while ambiguous in other projections or in regular mammograms. It was also found that inclusion of surgical clips (used to indicate tumor location) had caused detrimental reconstruction artifacts. Acknowledgment: This work was supported in part by a research grant EB000117 from the NIBIB and a research grant CA104759 from the NCI.

SU‐E‐I‐02: Effects of Projection View Sampling On CT Numbers and Noise Level in Cone Beam Breast CT

PURPOSE To investigate and optimize the effects of the number of projection views for minimal noise level in cone bream breast CT. METHODS An Anrad flat panel detector was used on a bench top experimental CBCT system with fixed exposure. A wax phantom simulating a breast was used to study the reconstruction artifacts. A Lucite phantom with uniform structure was used to study the variation of the noise level with the number of views. The phantoms were scanned with 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 1200 and 1400 evenly spaced projection views over 360° while keeping the total exposure at a level corresponding to a mean glandular dose (MGD) of 7.2 mGys. Each image set was reconstructed with Feldkamps backprojection algorithm using a ramp filter. Two regions of interest (ROIs) were selected in the reconstructed images, one at the center and the other near the edge. Mean values and standard variations were measured in ROIs and plotted as the function of the number of projection views. RESULTS The mean values for the wax phantom were found to peak at 300 projection views. The standard variation was lowest with 400 projection views for both Lucite and wax phantoms. CONCLUSION For the detector and exposure level used, 400 projection views resulted in lowest noise level, indicating optimal combination of reconstruction artifacts from projection view sampling and image noises, which tend to decrease and increase with the number of projection views, respectively. This work was supported in part by grants CA139830, CA138502 and CA124585 from the NIH-NCI.

SU-FF-I-23: Full-Scan Versus Half-Scan in Cone Beam Breast CT - a Quantitative Comparison

Purpose: To evaluate and compare half‐scan and full scan techniques for cone beam breast CT in terms of CT number, radiation dose distribution and CNR.Method and Materials:CT number comparison: A 11 cm diameter breast phantom, made of a stack of glandular and fat tissue, was analytically modeled to evaluate the difference of CT numbers reconstructed from Feldkamp algorithm based half and full scan reconstruction. Radiation dose distribution: An 11 cm diameter cylindrical breast model was constructed to evaluate the dose distribution with Geant4 based Monte Carlo simulation.CNR evaluation: An 11 cm diameter polycarbonate cylinder was imaged to simulate breast imaging. Five holes of the same size (5 mm in diameter) were drilled at different radial distance from the phantom center. The phantom, filled with iodine solution, was imaged with an aSi/CsI flat panel detector based cone beam CT scanner. Iodine CNRs were measured for different half‐scan coverage selection. Results: It has been found that the CT numbers reconstructed from half‐scan were close to those from full‐scan. The radiation doses in the half‐scan coverage can be twice than those out the coverage as the radial distance increasing. The CNRs for different half‐scan coverage were approximately identical. Conclusion: Our results show that the reconstructed data from half‐scan are comparable to that from full‐scan. The radiation doses are low for the positions out the half‐scan coverage. The CNRs vary little with half‐scan coverage. Acknowledgement: This work was supported in part by grants CA104759 and CA124585 from NIH‐NCI, a grant EB00117 from NIH‐NIBIB, and a subcontract from NIST‐ATP.

SU‐FF‐I‐16: Volume‐Of‐Interest (VOI) Cone Beam CTwith Dual Resolution Image Acquisition

In this study, we investigate the feasibility of using VOI projection data acquired at high resolution in conjunction with full width projection data acquired at low resolution to reconstruct cone beam CTimages for the VOI. To simulate cone beam CT with dual resolution image acquisition, flat panel images of a mastectomy specimen, acquired in the non‐binning mode, were converted into low resolution full width projection data. High resolution VOI projection data were directly extracted from the original data. To prepare for reconstruction, the low resolution projection data were first interpolated, re‐sampled to fill in the truncated space outside the VOI. The dual resolution full width projection data, consisting of true high resolution data in the VOI and interpolated data outside the VOI, were then used to reconstruct the 3‐D image for the VOI. Reconstructed images obtained with dual resolution projection data were compared with those obtained with low resolution data and those obtained with high resolution data for the visibility of small calcifications. We have successfully demonstrated the use of dual resolution projection data for VOI cone beam CTimaging. While the low resolution full width projection data did not allow smaller calcifications to be seen in the reconstructed images, addition of high resolution projection data for the VOI only could make them visible. The use of interpolated low resolution projection data to pad the truncated space outside the VOI did not affect the spatial resolution of reconstructed images inside the VOI. With the dual resolution technique, it would be possible to selectively image a VOI at very high resolution without requiring excessively long acquisition and reconstruction or unnecessarily overexposing the patient outside the VOI. (This work was supported in part by a research grant CA104759 from the NCI and a research grant EB‐00117 from the NIBIB).

MO-F-CAMPUS-I-02: Accuracy in Converting the Average Breast Dose Into the Mean Glandular Dose (MGD) Using the F-Factor in Cone Beam Breast CT- a Monte Carlo Study Using Homogeneous and Quasi-Homogeneous Phantoms

Purpose: To investigate the accuracy in estimating the mean glandular dose (MGD) for homogeneous breast phantoms by converting from the average breast dose using the F-factor in cone beam breast CT. Methods: EGSnrc-based Monte Carlo codes were used to estimate the MGDs. 13-cm in diameter, 10-cm high hemi-ellipsoids were used to simulate pendant-geometry breasts. Two different types of hemi-ellipsoidal models were employed: voxels in quasi-homogeneous phantoms were designed as either adipose or glandular tissue while voxels in homogeneous phantoms were designed as the mixture of adipose and glandular tissues. Breast compositions of 25% and 50% volume glandular fractions (VGFs), defined as the ratio of glandular tissue voxels to entire breast voxels in the quasi-homogeneous phantoms, were studied. These VGFs were converted into glandular fractions by weight and used to construct the corresponding homogeneous phantoms. 80 kVp x-rays with a mean energy of 47 keV was used in the simulation. A total of 109 photons were used to image the phantoms and the energies deposited in the phantom voxels were tallied. Breast doses in homogeneous phantoms were averaged over all voxels and then used to calculate the MGDs using the F-factors evaluated at the mean energy of the x-rays. The MGDs for quasi-homogeneous phantoms were computed directly by averaging the doses over all glandular tissue voxels. The MGDs estimated for the two types of phantoms were normalized to the free-in-air dose at the iso-center and compared. Results: The normalized MGDs were 0.756 and 0.732 mGy/mGy for the 25% and 50% VGF homogeneous breasts and 0.761 and 0.733 mGy/mGy for the corresponding quasi-homogeneous breasts, respectively. The MGDs estimated for the two types of phantoms were similar within 1% in this study. Conclusion: MGDs for homogeneous breast models may be adequately estimated by converting from the average breast dose using the F-factor.