Index of content:
Volume 33, Issue 6, June 2006
- Imaging Scientific Session: Room 330 A
- Computed Tomography
MO‐D‐330A‐01: Retrospective Sorting of 4D CT Into Breathing Phases Based On Imaging Analysis of a Fixed‐Geometry Fiducial33(2006); http://dx.doi.org/10.1118/1.2241403View Description Hide Description
Purpose: To evaluate a novel fiducial method for retrospective sorting of four‐dimensional computed tomography (4D‐CT). Breathing‐induced motion and deformation of internal anatomy confounds planning and delivery of radiotherapy. Patient‐specific assessment of respiratory motion using 4D‐CT is becoming more clinically accepted as it depicts discrete sampling of the internal anatomy throughout the respiratory cycle. Various strategies for sorting the images exist. External strategies rely on integration of a motion sensor with the CT device but may lack robustness with respect to sensor location. Internal strategies based on image analysis alone may not be robust enough. This work proposes a hybrid retrospective sorting method based on an easily positioned fiducial device that does not require additional hardware and is usable with any multi‐slice or helical CT scanner. An image‐processing algorithm automatically extracts the breathing phase of each slice by fiducial position and sorts images into phases accordingly. Method and Materials: A 50 cm rod‐like device covering the entire field‐of‐view is placed along the patient, providing a well distinguishable fiducial position in each CT slice without impacting image quality. Image analysis determines the fiducial centroid position in each slice with sub‐mm accuracy, allowing phase‐based binning of the image slices according to breathing phase. To validate the method, a motion phantom with the rod affixed was subject to a cine 8‐slice CT scan with 2.5 mm slice thickness. Images were sorted with the fiducial method and compared with images sorted by a commercial 4D‐CT system. Results: Phase‐sorted images of the phantom were reconstructed using the fiducial method. Image quality was comparable to those reconstructed by the commercial 4D‐CT system. Conclusions:Image analysis of a rod fiducial allows retrospective sorting of 4D CT according to breathing phase. This method does not require additional hardware, interfacing with the CT scanner, or manual interaction with the images.
33(2006); http://dx.doi.org/10.1118/1.2241404View Description Hide Description
Purpose: During diagnostic x‐ray CTimaging procedures or image guided radiotherapy,image quality will be degraded if target organs move during the data acquisition. This can be caused by patients' occasional motion or by intrinsic motion like cardiac and respiratory motion. The inconsistency in the projection data is the major reason for the image quality degradation. We present and validate a method to improve the consistency of the projections using a novel Fan‐beam Data Consistency Condition (FDCC) such that the image quality can be improved. Method and Materials: Computer‐simulated dynamic phantoms are generated and projection data are acquired from these dynamic phantoms. Using the FDCC, individual projection data from one view of fan‐beam projections can be estimated from filtering all the other projection data acquired from different view angles. Then those projections contaminated by motion are re‐estimated using the FDCC, resulting in a corrected sinogram. A standard Fan‐beam Filtered Back Projection (FBP) reconstruction algorithm is then used to reconstructimages from the corrected sinograms. Motion artifacts can be alleviated using this procedure. Results:Images are reconstructed from both the original sinogram where projections are contaminated by motion and the corrected sinogram after applying the FDCC. Strong motion artifacts are observed in the imagesreconstructed from the contaminated sinogram while improvement can be found in the reconstructed images using the corrected sinogram.Conclusions: A novel method using the new FDCC is proposed to combat the motion artifacts due to the temporal inconsistency in the projection data. Numerical simulations were conducted to demonstrate the potential of this correction scheme to mitigate motion artifacts. Thus, the preliminary numerical results indicate that the FDCC has potential use in combating both cardiac and respiratory motion in CTimaging.
MO‐D‐330A‐03: Correction of Streaking Artifacts in CT Images and Its Influence On Monte Carlo Dose Calculations33(2006); http://dx.doi.org/10.1118/1.2241405View Description Hide Description
Purpose: To quantify the impact of streaking artifacts in CTimages due to metal implants in patients on Monte Carlo dose calculation and to determine the impact of their correction. Method and Materials: For CT artifact correction a method of interpolation of missing data in sinograms was developed. Three contrast phantoms were constructed containing two steel cylinders that produced streaking artifacts. CT scans of the phantoms were obtained and the images were corrected for the artifacts. Three sets of Monte Carlo dose calculations (MCDC) using EGSnrc/DOSXYZnrc code were performed. Dose was calculated on: (1) the original CTimage, (2) the CT artifact corrected image, and (3), the exact phantom geometry. The dose distributions of the original CTimages and the CT artifact corrected images were then compared to the dose calculated on the exact geometry. Results: A calibration point for metal had to be added to the default EGSnrc CTcalibration curve to improve dose calculation results. Additional improvement in dose calculation results and in image quality was noted after the artifact correction was done. MCDC without adding the extra calibration point and without correction for streaking artifacts was found to lead to large dose errors. The error in dose calculations performed with the default calibration was found to be 25% in the original CTimages. The error improved greatly when the CTimages were corrected for artifacts and when the extended calibration was used; the error decreased then to less than 2 %. Conclusion: This work proves that the correction of streaking artifacts is important for MCDC; it significantly decreases dose calculation error and it improves image quality. The work also suggests that for MCTP an additional calibration point for a metallic material should be added to the default CTcalibration curve.
33(2006); http://dx.doi.org/10.1118/1.2241406View Description Hide Description
Purpose: We aim to devise new weighting schemes to improve the signalto‐noise ratio (SNR) in imagesreconstructed from data acquired in reduced fan‐beam scans by fully utilizing the redundant information. Method and Materials: Recently, we have developed a backprojection‐filtration (BPF) algorithm, which can reconstruct ROI‐images from the transversely truncated data in a reduced‐scan with a scanning angular range less than that in a short‐scan. However, the measured data in a practical reduced scan contain some redundant information. In this work, we devised two weighting schemes to appropriately incorporate the redundant data. Applying the weighting schemes to the BPF algorithm, we derived two new algorithms, derivative of weighted BPF (WD‐BPF) and weighted‐derivative BPF (DW‐BPF) algorithms. Both of these two algorithms can be used to improve the SNR in reconstructed images.Results: The ROI‐images are reconstructed from the truncated projection data in a short‐scan and a reduced‐scan, respectively. For the reduced‐scan, some shading artifacts appear in the imagesreconstructed by use of the WD‐BPF algorithm. This is caused by the numerical error in the derivative of the discontinuous weighting function. In contrast, artifact‐free images can be obtained by use of the DW‐BPF algorithm. For the short‐scan, both algorithms can obtain artifact‐free images, because the weighting function is smooth. The results in the noise study show that imagenoise obtained by use of WD‐BPF and DW‐BPF algorithms are similar within the ROI. Conclusion: We propose two weighting schemes for handling the data redundancy in reduced‐scan fan‐beam CT. Our results demonstrate that the proposed two algorithms can utilize the redundant data to improve the signal‐noise ratio. Moreover, the noise properties of these two algorithms are similar to each other.
33(2006); http://dx.doi.org/10.1118/1.2241407View Description Hide Description
We evaluated Feldkamp artifacts, which are specific to cone‐beam CT, in phantom and clinical studies using the 256 multi‐detector‐row CT (256MDCT), and compared the reconstruction accuracy of axial and helical scans.
Imagenoise, slice sensitivity profile (SSP) and artifacts with the 256MDCT were evaluated using a phantom and the results were compared to those with a 64MDCT. We also examined chest and abdomen scans produced with the 256MDCT in volunteers.
For the axial scan, Feldkamp artifacts were visualized as high‐frequency streak‐like artifacts that are oriented horizontally at the edge of the scan region in the phantom study. Similar results were obtained with the volunteers in soft‐tissue regions near either bony structures or air pockets. Feldkamp artifacts with the 256MDCT can lead to misdiagnosis if not correctly identified and minimized via helical scanning. Imagenoise was less for axial than helical scans, while SSP was better with helical than axial scans.
Feldkamp artifacts observed in the 256MDCT images, however, did not generally affect the interpretation of images. The 256MDCT promises more accurate diagnosis, and will provide volumetric cine images of wider cranio‐caudal coverage, enabling new applications of CT.
MO‐D‐330A‐06: Development of the X‐Ray Detector with Sequential Readout Circuits for Multidetector‐Row Computed Tomography33(2006); http://dx.doi.org/10.1118/1.2241408View Description Hide Description
Purpose: To develop a low‐cost X‐ray detector with sequential readout circuits, to realize enough low noise for multidetector‐row computed tomography(MDCT), and to evaluate image quality. Method and Materials: We have developed an X‐ray detector that has a MOS‐switch for each pixel, connects many pixels of a common column with the electric readout circuit, and outputs the signals of these pixels from one circuit by turning on lines of switches in order. It has fewer readout circuits than a conventional MDCT detector, but new design is necessary to realize enough low noise for MDCT. First, to make the required noise specific, we simulated the relation of the detectornoise and imagenoise (simulation(A)). Second, to consider how to realize it, we simulated the detectornoise with the circuit noise model (simulation(B)). Third, we constructed the detector in order to evaluate its noise. Last, we developed a test CTsystem with these detectors to evaluate imagenoise with phantoms. Results: The result of the simulation(A) indicated that detectornoise had to be less than about 10‐k rms electrons, and we found to be able to achieve it by optimizing the circuit parameters of the low pass filter and the data line as a result of the simulation(B). We constructed the detectors with these parameters to evaluate these noise, and it turned out that it was about 10.5‐k rms electrons and the required noise was achieved. Moreover, the result to evaluate the noise from images with phantoms indicated that the main was X‐ray quantum noise and the detectornoise was low enough to be ignored when the object was a cylindrical water‐filled phantom less than about 30 cm in diameter and the slice thickness of the images was 0.625 mm. Conclusion: We developed a low‐noise X‐ray detector with sequential readout circuits for MDCT.
33(2006); http://dx.doi.org/10.1118/1.2241409View Description Hide Description
Purpose: Most tomographic imaging systems available today use a single x‐ray source and multiple projection images are obtained by rotating the x‐ray source around the object. Therefore the data acquisition rate is limited by the gantry rotation speed, which is approaching the physical limit. We proposed to develop a novel stationary scanning x‐ray imaging system based on carbon nanotubefield emission x‐ray (FEX) technology. Instead of a single x‐ray source the proposed system is based on a multi‐pixel FEX source. The new scanner promises a dramatically faster data acquisition rate by reducing or totally eliminating the mechanical motion. Method and Materials: We have constructed a prototype stationary scanning x‐ray imaging system with an array of 9 individually addressable x‐ray source pixels, each of which can produce a different projection image of the object. The core of this novel x‐ray imagingtechnology is a gated carbon nanotubefield emissioncathode array. By programming the gate voltage of the cathode array, the multi‐pixel x‐ray source can generate an electronically triggered scanning x‐ray beam and produce multiple projection images from different viewing angles without mechanical motion. A Hamamatsu C7921 flat panel x‐ray sensor was used to collect all 9 projection images.Results:Tomosynthesisimages of a mouse and a standard breast‐imaging phantom (Stereotactic Needle‐biopsy Tissue Equivalent Phantom, Nuclear Associates, NY) using the prototype stationary scanning x‐ray imaging system are acquired. Tomosynthesisreconstructions were applied to the breast phantom. The slice imagesreconstructed using an iterative reconstruction algorithm clearly show the internal structures of the breast‐imaging phantom at different depths. Conclusion: We have developed a stationary scanning x‐ray imaging system using a carbon nanotube based multi‐pixel FEX source. The mechanical motion free approach can lead to a faster and simplified tomographic imaging system.
Conflict of Interest: Research partially supported by Xintek Inc.
33(2006); http://dx.doi.org/10.1118/1.2241410View Description Hide Description
Purpose: Implementing Quantitative Computed Tomography (QCT) on Multi‐Slice Computed Tomography (MSCT) scanners requires investigating the effects of axial vs. helical scan modes and protocol parameter variations on quantitative data. While previous work in this area focused on single‐slice axial techniques, technological developments in Computed Tomography(CT) justify more complex assessment. Method and Materials: All scans were obtained using two phantoms designed for bone mineral (BM) densitometry: a reference phantom (three different density cores) and a QA torso phantom. Both phantoms have known properties and are required for long‐term quantitative BM density assessment. The scan acquisition parameters that were varied included kV, mA, rotation speed, pitch, image thickness, detector configuration, reconstruction algorithm, table height, and tube temperature. To assess long‐term scanner drift, the QA Torso phantom was scanned multiple times over three months on each of seven MSCT scanners (five GE Lightspeed‐ 16s, one GE Lightspeed Qx/i, and one GE Lightspeed‐Plus). The daily variability of the individual MSCT scanners and scanner‐to‐scanner variability was determined by coefficient of variation (mean/variance) from the QA Torso phantom data sets over time. All data were collected and analyzed in Hounsfield Units (HU) to provide insight about variations upstream of the actual BM density analysis through commercial software.Results: This study found no significant difference (p > 0.05) in mean HU between phantom images obtained using axial and helical scan mode, or when varying most of the other scan acquisition parameters. However, varying kV and reconstruction algorithm did result in significant (p<0.0001) quantitative shifts. Preliminary data indicated daily variability of 0.8% – 1.9% and scanner‐to‐scanner variability of 1.4%.Conclusion: MSCT systems can be optimized for use in determining the BM density of a vertebral body, provided very careful control of scan acquisition protocol is observed.
33(2006); http://dx.doi.org/10.1118/1.2241411View Description Hide Description
Purpose: Different fanbeam reconstruction algorithms are being evaluated and the noise performances of these algorithms are compared at equivalent MTF.Method and Materials: The fanbeam reconstruction algorithms under comparison are FBP with Parkers smooth weighting (PFBP), LCFBP, DFBP, reconstruction algorithm by Noo et.al and exact reconstruction algorithm by Kudo et.al . The MTF from the five different algorithms were plotted and compared. In order to establish the basis for an unbiased comparison of the noise variance between different algorithms, we established the condition of equivalent spatial resolution. To achieve this, a window function was applied to the ramp filtering kernel before back‐projection for PFBP, LCFBP and Kudo's algorithm. A homogenous phantom was numerically simulated and Poisson noise with N0= 2e5 was added to the projection data. The images were reconstructed from projection data with and without the Poisson noise added. These images were then subtracted from each other to result in a subtracted or pure noise image. The variance in these noise images over five different ROIs was subsequently compared. FBP with Parkers smooth weighting was chosen as the gold‐standard and percentage decrease in variance in the imagesreconstructed using other four algorithms with respect to that of PFBP was tabulated. Results: The results showed that the new reconstruction algorithms had better noise performance than state‐of‐the‐art reconstruction algorithm (PFBP) after establishing the condition of equivalent spatial resolution. DFBP and exact reconstruction algorithm by Noo performed much better than the other three algorithms. Equal weighting scheme utilized definitely improved the noise performance over smooth weighting. DFBP showed a decrease of variance by about 23 % compared to PFBP. Conclusion: The reduction in noise variance theoretically leads to a radiation dose reduction by about 23 %. This will be of significant importance especially in pediatric imaging.
- Cone‐Beam CT
33(2006); http://dx.doi.org/10.1118/1.2241933View Description Hide Description
Purpose: Estimation of breast skin thickness and breast shape using a radial‐geometry segmentation algorithm on breast CTimages.Method and Materials: Forty‐two breast imagedata sets were obtained from a prototype breast CT scanner, and were used to evaluate breast skin thickness and effective diameter. Patients were imaged at 80‐kVp using x‐ray tube currents (0.7 – 7.6 mA) depending on the patient's breast size. For each coronal breast image, the breast silhouette was segmented using a threshold value computed by a histogram‐based iterative algorithm. Breast area was also computed from the thresholded coronal images. A 360‐degree radial scan, originating at the center of mass of each breast CTimage and continuing to the image edge, produced a radial profile of breast tissue intensity as a function of angle. A derivative filter was used to identify the inner and outer breast skin layers. In order to accurately estimate breast skin thickness, a tangent‐finding algorithm was developed to correct the thickness measurement in non‐circular breast geometries. A standard‐deviation‐based iterative algorithm was also implemented to reduce noise in the skin thickness estimation. Results: Among 42 patients, breast skin thickness was determined to be between 1.50 – 1.55 mm. Plots of effective breast diameter as a function of posterior‐anterior position serve as a concise method for characterizing idealized 3D breast shape, and these parameterized curves are reported for breasts of different size classes based on the cup size metric. Conclusion: Breast CT acquisition techniques, combined with algorithms designed for determining specific breast metrics, were useful for classifying breast shape and skin thickness. Most breast dosimetry coefficients (DgN) are based on the assumption of a 4 mm skin thickness, and the thinner skin dimensions found in this study will likely have a small but significant influence (increase) on breast dosimetry in mammography.
33(2006); http://dx.doi.org/10.1118/1.2241934View Description Hide Description
Purpose: Circular cone‐beam computed tomography is challenged by a lack of plane sums resulting in incomplete inversion of the radon transform. These missing plane sums have been identified by theory, and may be represented as a shift‐variant cone of missing frequencies in the Fourier domain. The aim was to verify the presence of this cone in real data, and to show the dependency of resulting image artifacts on the frequency distribution of the imaged object. Method and Materials: A mini disk phantom (mylar/foam) was constructed to probe the local frequency response at various locations in the reconstruction space. Projections obtained using an experimental CBCT benchtop were reconstructed with 120 μm3 voxel size using a modified Feldkamp filtered backprojection routine. Local Fourier transforms of the mini disks were analyzed for missing frequency data and compared with theory. Large disk phantoms of acrylic and cellular polyurethane were also imaged for further demonstration of the effect of varying the frequency content of the imaged object. Results: The cone of missing frequency was successfully identified in the mini disk phantom and agreed well with theory. Image artifact was found to have dependency on the local distribution of the object's frequency power spectrum relative to the cone of missing frequency information. Decreased resolution of the disks occurred when their dominant spatial frequency components were directionally aligned to coincide with the predicted null cone, as expected. Image reconstructions of large disk phantoms showed good detail in cellular disks even at locations that showed strong artifacts in the acrylic disk. Conclusion: The predicted null cone is observable in Fourier transforms of localized objects. Resolution in reconstruction is dependent on the relative frequency distribution of the imaged object; features that are most poorly resolved will be those with strong frequency components directed in the expected null space.
TH‐E‐330A‐03: Reduction of Ring Artifacts in Cone Beam CT: Artifact Detection and Correction for Flat Panel Imagers33(2006); http://dx.doi.org/10.1118/1.2241935View Description Hide Description
Purpose: Defective pixels in a flat panel detector may be characterized by a nonlinear signal response drastically different from those in the neighboring pixels. They could lead to ring artifacts in cone beam CT. In this presentation, we will describe and demonstrate the use of a filter based calibration technique to detect these pixels. In addition, we will employ this technique as a flat field correction method to correct nonlinear signal response. Methods and Materials: To force the signal responses of all pixels to vary smoothly without sudden changes, we acquired images at fixed mAs but with various filtrations. Each filtered image is fitted to a smooth surface whose values are close to those of normal pixels but vary smoothly in the image. The ratios of the surface fit values to the original values were then computed on a pixel‐by‐pixel basis and used to map pixel values during subsequent image acquisition. This mapping would compensate for the nonlinear signal response associated with the defective pixels thus eliminating the ring artifacts. Using the surface fits as the reference, defective pixels were detected by automatic thresholding. Results: Using filter based calibration, defective (∼0.035 %) pixels were successfully detected and corrected for. Therefore, ring artifacts were largely eliminated in cone beam CTimages. A reduction in patterned noise artifacts in the projection images was also observed. The automated surface fitting procedure was found to be robust. Conclusions: Although the conventional flat field correction addresses non‐uniform response across the detector, artifacts may still form as the pixel response varies with the beam quality and signal intensity. The filter based calibration procedure was successfully used to detect and correct for these artifacts.
This work was supported in part by a research grant EB00117 from the NIH‐NIBIB and a research grant CA104759 from the NIH‐NCI.
TH‐E‐330A‐04: Investigation Into the Cause of a New Artifact in Cone Beam CT Reconstructions On a Flat Panel Imager33(2006); http://dx.doi.org/10.1118/1.2241936View Description Hide Description
Purpose: To investigate the source of and possible corrections for a new artifact seen in cone‐beam CT(CBCT)images acquired using an a‐Si flat panel imager(FPI) (Varian 4030CB). The new artifact is a bright circular region, tangent to the phantom edge, in images of elliptical and off‐center phantoms. Methods and Materials: Temporal response of the FPI was measured using a step‐wedge phantom, as a function of dose and irradiation history (10 cycles of 80 s exposure, with 9 off‐cycles varying in time between 2 and 30 minutes, total time 1hour 20 min). A linear time invariant (LTI) model was developed by fitting multi‐exponentials to the lag response from the step‐wedge phantom. Anthropomorphic phantoms — pelvis placed centrally, and head placed off‐center — were scanned and reconstructed with and without the developed correction. Results: Detector lag and continuous and significant monotonic gain increase (up to 10% for long irradiation periods) were observed during constant irradiation. Even after long periods of no exposure, with the detector being continuously read out, the gain did not return to the original, start‐of‐day value. Unlike Overdick et al., we did not see a saturation effect in the gain change. In CBCTreconstructions, differences up to 35 HU existed close to the edges of the artifact. After applying our correction model, differences were reduced to less than 10 HU. Our anthropomorphic phantoms did not generate streaks or comet tails, which other investigators have shown to be due to lag. Conclusion: We have determined that the source of the circular artifact observed is a non‐ideal temporal response. This artifact can be mostly eliminated by applying a correction based on a LTI model. Future work will focus on more accurate modeling to completely eliminate the artifact. Conflict of Interest: Funding was provided by Varian Medical Systems.
TH‐E‐330A‐05: Reducing Metal Artifacts in Cone‐Beam CT by Tracking and Eliminating Metal Shadows in Raw Projection Data33(2006); http://dx.doi.org/10.1118/1.2241937View Description Hide Description
Purpose:CT streaky artifacts, caused by metallic implants such as fiducial markers or dental fillings, remain a challenge for automatic processing of image data. The effect of these metal artifacts is magnified in cone‐beam CT(CBCT)images due to the fact that the soft tissuecontrast is usually lower in these images and therefore is more sensitive to the artifacts. The goal of this study is to develop an effective offline processing technique to minimize the effect. Method and Materials: The geometry calibration cue of the CBCT system was used for tracking the position of the metal implant in the raw projection data. The 3D representation of the metallic object can be established from only two user‐selected viewing angles. The position of the shadowed region in any view can be accurately tracked by re‐projecting the 3D coordinates of the metal object. Then automatic image segmentation was performed to obtain a binary mask of the shadow at each projection angle. Finally, a Laplacian diffusion filter was used to replace the pixels in the masked region with the boundary pixels. The modified projection data were then sent back to the CBCTreconstruction engine to create a new CBCTimage. Varian's Trilogy system was used in this study. The procedure was tested phantoms and patient cases. Results: It was demonstrated that this procedure can significantly minimize the metal artifacts and at the same time restore soft tissuecontrast near the metallic object, even for the more difficult head and neck case with irregularly shaped dental fillings. Soft tissue visibility was improved drastically. Although not designed for on‐line applications, the processing time is approximately 1–2 second per projection on an Intel Pentium processor at 2.6GHz. Conclusion: We have implemented an effective metal artifact suppressing algorithm to improve the quality of CBCTimages.
33(2006); http://dx.doi.org/10.1118/1.2241938View Description Hide Description
Purpose: X‐ray scatter significantly degrades the quality of cone‐beam CT(CBCT)reconstructions by introducing cupping and streaking artifacts. Simple correction techniques, based on subtracting a constant value across a projection, fail when the flat‐panel detector is transaxially offset as is required to increase the reconstruction field‐of‐views for body scans. The purpose of this study was to measure x‐ray scatter profiles for transaxially offset detectors and to characterize the resulting artifacts. Method and Materials: Data were collected on a table‐top CBCT system. A pelvic phantom was imaged with a Varian 4030CB imager offset by 16 cm. Scatter was estimated by subtracting a nearly “scatter‐free” projection data set, obtained by narrowing the axial collimator blades, from the full CBCT data set obtained with the blades in their fully open position. The resulting scatter estimate, valid in the narrow region of overlap, was extrapolated to generate a scatter estimate across the entire axial extent of the detector. This scatter estimate was then subtracted from the original CBCT data to generate scatter‐corrected images.Results: The scatter profile in full‐fan projections is relatively flat and symmetric. In contrast, in the half‐fan configuration the measured scatter profile was asymmetric decreasing monotonically from the phantom‐air boundary through the phantom center to the imager edge. The slope of this profile varied smoothly from the AP to lateral views resulting in reconstructions with abnormally bright and dark regions. Scatter correction using the measured profile proved effective. Cupping and doming amplitudes were reduced by 2/3. The average reconstruction error in the prostate region was reduced from over 120 HU to less than 40 HU. Conclusions: The asymmetries introduced by an offset detector result in non‐uniform scatter profiles that generate unusual cupping artifacts. Our technique provides a means of characterizing these profiles. Conflict of Interest: Funding provided by Varian Medical Systems.
33(2006); http://dx.doi.org/10.1118/1.2241939View Description Hide Description
Purpose: Linac‐mounted cone‐beam CT(CBCT)imaging is a promising approach for improving the soft‐tissue targeting accuracy of external‐beam radiotherapy. However, the large proportion of signal due to scattered radiation results in large cupping artifacts and reduced contrast‐to‐noise ratio (CNR), hampering delineation of soft‐tissue structures. The goal of this investigation is evaluate the impact of three scatter mitigation strategies: scatter removal (SR), subtractive scatter corrections (SSC), and anti‐scatter grids (ASG) on CBCTimageCNR.
Method and Materials: A one‐dimensional model was developed that to predict imagenoise, intensity, and contrast from the photon flux incident upon the flat‐panel detector. Using previously published scatter‐to‐primary ratio (SPR) data, the impact of Poisson signal statistics, Gaussian readout noise, reconstruction filter, and total air‐kerma on image uniformity, contrast, and CNR was evaluated. Using previously measured CBCT scatter and primary grid transmission; the impact of ASGs, SSR, and SR (e.g., via bowtie filters) was evaluated. In addition, synthetic polyenergetic CBCT projections of cylindrical low‐contrast threshold phantoms, including compound Poisson process noise, were used to evaluate the impact of scatter mitigation on filtered backprojection (FBP) images.Results: Relative to complete SR, typical CBCT scatter reduces CNR by 30%–70% depending on patient diameter. Both 1D analysis and simulated images demonstrate that SSC effectively reduces cupping artifacts but does not improve CNR. ASGs with high scatter selectivity (>5) or large primary transmission (>0.7) modestly increase (20–30%) CNR for thick subjects. However, ASGs diminish CNR for smaller body sections or for exposure levels where additive noise dominates sinogram signal statistics. Conclusion: Correcting CBCTsinograms for scatter radiation is effective in reducing structural image artifacts but does not improve CNR. Antiscatter grids effectively suppress scatter artifacts but improve CNR only in selected clinical settings. This project was supported in part by a grant from Varian Medical Systems.
- Image Segmentation and CAD
33(2006); http://dx.doi.org/10.1118/1.2241546View Description Hide Description
Purpose: To develop a computerized lesion detection method for DCE‐MRI breast images using the fuzzy c‐means clustering algorithm. Method: Contrast‐enhanced MR imaging is increasingly being incorporated into procedures for the screening of women at high risk of developing breast cancer. Such screening programs may potentially benefit from computer prompts that indicate potential lesion sites. In addition, analysis of other enhancing regions in the breast may reduce the number of false detections. Thus, we are developing an automated computerized lesion detection method based on the fuzzy c‐means clustering algorithm. The proposed method consists of four stages: (1) Breast volume segmentation based on a volume growing method; (2) Fuzzy c‐means clusteringanalysis on voxel‐based kinetics within the 4D breast image data (3D over time); (3) Voxel‐by‐voxel membership assignment to the most‐enhancing categories; and (4) Connectivity & size criteria for eliminating some false‐positive detections. Methods were evaluated by calculating detection sensitivity for malignant lesions, detection sensitivity for all lesions, and number of false‐positive detections per breast volume for output from the most‐enhancing kinetic categories. Results: Our preliminary studies are based on 20 MRI cases including 21 lesions (9 biopsy‐proven malignant cases, 5 biopsy‐proven benign cases; 6 cases without pathological proof). Based on computer‐identified regions from the most enhancing membership category, the proposed method correctly detected 16 lesions, including all nine malignant ones. In addition, most of the benign cases fell into either the most‐enhancing or second‐most‐enhancing categories. Preliminary results yielded, on average, 9 false‐positive detections per breast volume, which will subsequently be input to the classifier stage that exams morphological and kinetic characteristics for false positive reduction. Conclusion: The preliminary results with our FCM‐based computerized MRI lesion detection method are promising for potential use in breast cancer screening. Conflict of Interest: M.L.G. is a shareholder in R2 Technology, Inc.
33(2006); http://dx.doi.org/10.1118/1.2241547View Description Hide Description
Search involves locating lesions in images under conditions of uncertainty regarding the number and locations of lesions that may be present. A mathematical model of search is presented that applies to situations, as in the free‐response paradigm, where on each image the number of normal regions that could be mistaken for lesions is unknown and the number of observer generated localizations of suspected regions (marks) is unpredictable. The search model is based on a two‐stage descriptive model proposed in the literature according to which at the first stage the observer uses mainly peripheral vision to identify likely lesion candidates, and at the second stage the observer decides whether or not to report the candidates. The mathematical search model regards the unpredictable numbers of lesion and non‐lesion localizations as random variables and models them via appropriate statistical distributions. The model has three parameters quantifying the perceived lesion signal‐to‐noise ratio, the observer's expertise at rejecting non‐lesion locations and the observer' expertise at finding lesions. A figure‐of‐merit quantifying the observer's search performance is described, as well as ROC and FROC curves predicted by the search model. Finally, we describe a preliminary method for estimating the parameters of the search model from free‐response data.
TU‐D‐330A‐03: Comparison of Image Segmentation Algorithms On Digitized Mammograms and FFDM Images for CAD33(2006); http://dx.doi.org/10.1118/1.2241548View Description Hide Description
Purpose: To investigate lesion tumor segmentation methods for both digitized screen‐film mammograms (DSF) and full‐field digital mammography (FFDM). Method and Materials: Breast lesion segmentation methods are important in the overall image analysis for computer‐aided diagnosis. Our initial development was performed on DSF, and we are currently evaluating our methods for use with FFDM. Three of our lesion segmentation methods were investigated using a database of 84 DFM and 287 FFDM cases including malignant and benign lesions. A region growing method utilizes the size and shape of the evolving lesion contour to determine the lesion margin. A radial gradient index (RGI) segmentation method uses a Gaussian constraint function to suppress the influence of distant pixels. Then for a series of contours returned by grey level thresholding, the contour with maximum RGI is chosen as the one that best delineates the lesion. A two‐stage, region‐based active contour method minimizes an energy function based on the homogeneities inside and outside of the evolving contour. The minimization algorithm solves the Euler‐Lagrange equation describing the contour evolution. Prior to the application of the active contour model, RGI segmentation is applied to delineate an initial contour closer to the lesion margin and estimate the effective background. The methods were compared to radiologist‐delineated margins on both DSF and FFDM images using an area similarity metric. Results: At an overlap threshold of 0.3, the region growing, RGI, and two‐stage methods correctly segmented 84%, 86% and 94% of the digitized screen‐film lesions, and 81%, 83% and 88% of lesions on FFDM, respectively. Conclusion: Our results indicate that the two‐stage method yields improved segmentation for both DSF and FFDM, and also that methods developed with DSF can be efficiently converted for use with FFDM. Conflict of Interest: MLG is a shareholder in R2 Technology, Inc.
TU‐D‐330A‐04: Lung Cancer Screening Performance of Chest CT Images Using Frequency of Radiologists' Location Identifications33(2006); http://dx.doi.org/10.1118/1.2241549View Description Hide Description
The radiology exam reports and the computer‐aided detection (CAD) findings on 26 low‐dose CT, lung‐cancer screening exams were compared to the findings of a reference database. The reference database was developed using 6 radiologists localizations of image features of concern. The radiologists were instructed to use a lax criterion for the identification of compact features that could be further evaluated for lungcancer. An unsupervised computer procedure using the distances between report locations determined which localizations were associated with the same image feature.
The reference database consisted of 609 findings of locations and their identification frequency. Forty two percent (257) of the reference database findings had a frequency of 2 or greater; these were considered positive finding using a lenient standard. The radiologist exam report identified 22% of the lenient standard findings, while the CAD system identified 16%. The database contained 50 findings (8%) using a stringent standard that the finding had to be reported on 5 or 6 occasions. The radiologist exam report identified 50% of the stringent standard findings, while the CAD system identified 40%. The radiology exam report had 6 findings not in the reference database.
Low doseCT,lungcancer screening exams contain numerous features of concern that could be further evaluated for lungcancer. The radiologist exam report missed 50% to 80% of these image features. The CAD system is unlikely improve the radiologists detection performance because it identified even fewer features of concern.
This study was made possible in part by equipment provided by R2 Technology and Siemens Medical.