J Winslow

Automated Technique to Measure Noise in Clinical CT Examinations

OBJECTIVE The purpose of this study was to develop and validate an automated method to measure noise in clinical CT examinations. MATERIALS AND METHODS An automated algorithm was developed to measure noise in CT images. To assess its validity, the global noise level was compared with image noise measured using an image subtraction technique in an anthropomorphic phantom. The global noise level was further compared with image noise values from clinical patient CT images obtained by an observer study. Finally, the clinical utility of the global noise level was shown by assessing variability of image noise across scanner models for abdominopelvic CT examinations performed in 2358 patients. RESULTS The global noise level agreed well with the phantom-based and clinical image-based noise measurements, with an average difference of 3.4% and 4.7% from each of these measures, respectively. No significant difference was detected between the global noise level and the validation dataset in either case. It further indicated differences across scanners, with the median global noise level varying significantly between different scanner models (15-35%). CONCLUSION The global noise level provides an accurate, robust, and automated method to measure CT noise in clinical examinations for quality assurance programs. The significant difference in noise across scanner models indicates the unexploited potential to efficiently assess and subsequently improve protocol consistency. Combined with other automated characterization of imaging performance (e.g., dose monitoring), the global noise level may offer a promising platform for the standardization and optimization of CT protocols.

A method for characterizing and matching CT image quality across CT scanners from different manufacturers

Purpose: The purpose of this study was to quantitatively characterize the fundamental aspects of image quality (IQ) associated with different computed tomography (CT) reconstruction algorithms, the resolution, noise texture, noise magnitude per dose, and use those data to devise a methodology to match IQ between different CT systems. Methods and materials: This study entailed a 3‐step methodology involving (a) characterizing the noise magnitude, texture, and resolution for a CT system‐reconstruction using the relationship between noise magnitude and Computed Tomography Dose Index (CTDI), noise power spectrum (NPS), and modulation transfer function (MTF), (b) developing clinically relevant strategies of weighting the differences among system‐reconstructions as a means to determine the best match (c) identifying for each target system‐reconstruction, system‐reconstructions with matched in terms of that minimum IQ differences. Images of the ACR CT phantom were acquired at two dose levels on each of two CT scanners. Images were reconstructed using all available reconstruction kernels and multiple iterative reconstruction (IR) settings. Each reconstruction was characterized as described above. Percent changes for each IQ metric were calculated for every possible pair of system‐reconstructions. Weighting functions, reflecting the human visual systems limit to discriminate between spatial frequencies with differences below 5%, were applied to the differences and the product of the weighted values was used to indicate the best match for each system‐reconstruction. Results: Noise texture and resolution are governed by choice of reconstruction kernel and IR strength, while noise magnitude is additionally dependent on dose. Harder kernels have better resolution, finer noise texture, and increase the required dose for a given noise magnitude, and vice versa. Increasing IR strength generally improves resolution, coarsens noise texture, and lowers the required dose. Seventy‐one percent of Siemens matches for GE target reconstructions had percent changes in noise texture/resolution under 5%. Seventy‐three percent of GE matches for Siemens target reconstructions had percent changes in noise texture/resolution under 5%. ACR phantom images for each matched reconstruction pair appeared similar in both noise magnitude and noise texture. Conclusion: Matching image appearance in terms of resolution, noise magnitude, and noise texture provides a quantitative and reproducible strategy to improve consistency in image quality among different CT scanners and reconstructions.

WE-D-18A-02: Performance Evaluation of Automatic Exposure Control (AEC) Across 12 Clinical CT Systems

PURPOSE Automatic exposure control (AEC) is not typically evaluated or monitored in CT quality assurance programs. The purpose of this study was to develop/evaluate a new AEC testing platform for the clinical physics program at our institution, and characterize AEC performance across different CT systems. METHODS The Mercury Phantom comprises three tapered and four uniform regions of polyethylene(16, 23, 30, and 37 cm in diameter); each region includes four inserts: air, Polystyrene, Acrylic, and Teflon. The phantom was imaged using AEC and a fixed tube current technique across 12 clinical CT scanners. Those included five Siemens Somatom Definition Flash, four GE Discovery CT750HD, and three GE Lightspeed VCT systems. A custom MATLAB software package provided MTF, NPS, and detectability indices for each diameter section of the phantom. Detectability indices were used to evaluate the relationship between AEC setting, patient size, and image quality. The magnitude of the power of a best fit exponential curve to the detectability indices and phantom diameter was used as a measure of AEC strength. Results were compared within/across scanner models, and as baseline values for comparison with future system performance testing. RESULTS For each scanner model, the percent difference in expected image quality and AEC setting was under 3%(+/-2%). The average decrease in detectability between the small and large diameter phantom sections for the Siemens Flash, GE CT750, and GE VCT was 99%(+/-10%), 42%(+/-25%), and 33%(+/-41%), respectively. The value signifying AEC strength was 0.051(+/-13%), 0.019(+/-18%), and 0.018(+/-26%), for the Siemens Flash, GE CT750, and GE VCT models, respectively. CONCLUSION This study demonstrated a practical approach to test the AEC performance of clinical CT systems at a large academic medical center. The quantification and evaluation of AEC performance should be included in acceptance testing and in annual physics testing of clinical CT systems.

TU-H-207A-05: Automated Early Identification of An Excessive Air-In-Oil X-Ray Tube Artifact That Mimics Acute Cerebral Infarct

PURPOSE There is an infrequent but serious CT artifact that occurs when there is too much air in the cooling oil of an x-ray tube. This artifact manifests as patchy hypodensities and mimics acute cerebral infarct. Routine quality control testing is unlikely to detect this artifact before it is observed in patient images. The purpose of this project was to develop an automated, quantitative method that increased the likelihood of identifying and preventing such artifacts. METHODS Using QC phantom images with a known air-in-oil artifact, a 1D radial representation of the 2D noise power spectrum(NPS) was calculated and compared against that for artifact-free images. The QC program software used at our institution to analyze daily phantom images was modified to include measuring the average frequency of NPS within the water section of daily phantom scans. The threshold values developed for each CT system were incorporated into our daily QC program and email notification system. RESULTS The NPS for the known air-in-oil artifact images included a large low frequency peak compared with artifact-free images; the average NPS frequency for these images were 0.197 and 0.319 (1/mm), respectively. The average NPS frequency values (mean+/- standard deviation) for the GE CT750, GE VCT, GE Lightspeed Xtra, and Siemens SOMATOM Definition Flash scanners were 0.322+/- 0.0058, 0.324+/-0.0024, 0.320+/-0.0020, and 0.303+/-0.0039 (1/mm), respectively. Threshold values were chosen to be the average plus or minus twice the standard deviation. Automated QC successfully identified an air-in-oil artifact in the Lightspeed Xtra before any detrimental clinical effect occurred; the average NPS frequency value that triggered service was 0.307, which is six standard deviations smaller than average. CONCLUSION Clinically serious problems associated with the air-in-oil artifact can be detected earlier and mitigated/avoided by incorporating the average frequency of NPS measurements of daily phantom images into an automated QC program.

Dependency of prescribed CT dose on table height, patient size, and localizer acquisition for one clinical MDCT

PURPOSE The purpose of this study was to quantify the effect that table height, patient size, and localizer acquisition order may have on AEC prescribed dose. METHOD AND MATERIALS Three phantoms were used for this study: the Mercury Phantom, acrylic sheets, and an anthropomorphic phantom. A lateral (LAT) and a posterior-anterior (PA) localizer was acquired for each phantom at different table heights on a MDCT scanner (GE Discovery CT750 HD). AEC scan acquisitions were prescribed for each combination of phantom, localizer orientation, and table height ±4 cm with the center position; the displayed CTDIvol was recorded. Based on the institutional dose monitoring program, the relationship between change in CTDIvol and change in table height were studied for LAT and AP localizers for clinical exams. RESULTS For all phantom scans based on the PA localizer, the percent change in ranged between -18% and 42% for table heights 4 cm below and above proper centering; while for the LAT localizer, the percent change in CTDIvol from ideal were no greater than 12% different for ±4 cm differences in table height. Change in CTDIvol and change in table height displayed a strong linear relationship for AP localizer exams (P = 0.002), and weak correlation for LAT localizer exams (P = 0.12). CONCLUSIONS Since uncertainty in vertical patient positioning is inherently greater than lateral positioning, the LAT localizer should be utilized to precisely and reproducibly deliver the intended amount of radiation prescribed by CT protocols.

SU‐F‐18C‐07: Automated CT QC Program with Analytics, Archival, and Notification Capabilities

PURPOSE Tracking metrics over time is a well-established means of establishing a quality control program. The number of metrics followed and testing frequency is limited by available resources. Automating the image analysis and data archival of a QC program enables objective and efficient tracking of performance metrics. The purpose of this study was to develop such a QC method and to assess its utility at a large clinical facility. METHODS The QC program at our institution is based on the acquisition of daily water-phantom scans, and biweekly ACR-phantom scans for each CT system. We developed a QC program to analyze these data. The QC software operates on the images sent directly to our server. The relevant information from DICOM headers was extracted, data analyzed, and a database was populated. The measurements performed on the waterphantom included water CT-number, uniformity, noise, and artifact. The measurements performed on the ACR-phantom included the MTF, NPS, detectability, artifact, uniformity, CNR, and the CT-numbers for water, polyethylene, bone, air, and acrylic. Email notifications and criteria limits were directly based upon ACR accreditation requirements and developing threshold values. RESULTS Across ten clinical CT scanners, the daily water CT-number was -0.2+/-1.4 HU(mean+/-standard deviation). The corresponding numbers for 10%MTF, uniformity, CNR squared normalized to CTDI, and detectability squared normalized to CTDI were 0.69+/-0.01 (1/mm), 0.93+/-0.29, 0.06+/-0.02 (1/mGy), and 3.3+/-0.7 (1/mGy), respectively. For all ACR-phantom inserts, the largest standard deviation for any individual scanner was 1.9 HU. Artifact analysis triggers successfully identified problematic images. CONCLUSION Automating image analysis allows one to frequently track meaningful metrics that would be impractical to follow otherwise. System inconsistencies are more likely to be identified and corrected earlier. Much tighter system specific criteria limits are possible.

TH-C-18A-06: Combined CT Image Quality and Radiation Dose Monitoring Program Based On Patient Data to Assess Consistency of Clinical Imaging Across Scanner Models

PURPOSE One of the principal challenges of clinical imaging is to achieve an ideal balance between image quality and radiation dose across multiple CT models. The number of scanners and protocols at large medical centers necessitates an automated quality assurance program to facilitate this objective. Therefore, the goal of this work was to implement an automated CT image quality and radiation dose monitoring program based on actual patient data and to use this program to assess consistency of protocols across CT scanner models. METHODS Patient CT scans are routed to a HIPPA compliant quality assurance server. CTDI, extracted using optical character recognition, and patient size, measured from the localizers, are used to calculate SSDE. A previously validated noise measurement algorithm determines the noise in uniform areas of the image across the scanned anatomy to generate a global noise level (GNL). Using this program, 2358 abdominopelvic scans acquired on three commercial CT scanners were analyzed. Median SSDE and GNL were compared across scanner models and trends in SSDE and GNL with patient size were used to determine the impact of differing automatic exposure control (AEC) algorithms. RESULTS There was a significant difference in both SSDE and GNL across scanner models (9-33% and 15-35% for SSDE and GNL, respectively). Adjusting all protocols to achieve the same image noise would reduce patient dose by 27-45% depending on scanner model. Additionally, differences in AEC methodologies across vendors resulted in disparate relationships of SSDE and GNL with patient size. CONCLUSION The difference in noise across scanner models indicates that protocols are not optimally matched to achieve consistent image quality. Our results indicated substantial possibility for dose reduction while achieving more consistent image appearance. Finally, the difference in AEC methodologies suggests the need for size-specific CT protocols to minimize variability in image quality across CT vendors.

SU-E-I-91: Reproducibility in Prescribed Dose in AEC CT Scans Due to Table Height, Patient Size, and Localizer Acquisition Order

PURPOSE In CT scanners, the automatic exposure control (AEC) tube current prescription depends on the acquired prescan localizer image(s). The purpose of this study was to quantify the effect that table height, patient size, and localizer acquisition order may have on the reproducibility in prescribed dose. METHODS Three phantoms were used for this study: the Mercury Phantom (comprises three tapered and four uniform regions of polyethylene 16, 23, 30, and 37 cm in diameter), acrylic sheets, and an adult anthropomorphic phantom. Phantoms were positioned per clinical protocol by our chief CT technologist or broader symmetry. Using a GE Discovery CT750HD scanner, a lateral (LAT) and posterior-anterior (PA) localizer was acquired for each phantom at different table heights. AEC scan acquisitions were prescribed for each combination of phantom, localizer orientation, and table height; the displayed volume CTDI was recorded for each. Results were analyzed versus table height. RESULTS For the two largest Mercury Phantom section scans based on the PA localizer, the percent change in volume CTDI from ideal were at least 20% lower and 35% greater for table heights 4 cm above and 4 cm below proper centering, respectively. For scans based on the LAT localizer, the percent change in volume CTDI from ideal were no greater than 12% different for 4 cm differences in table height. The properly centered PA and LAT localizer-based volume CTDI values were within 13% of each other. CONCLUSION Since uncertainty in vertical patient positioning is inherently greater than lateral positioning and because the variability in dose exceeds any dose penalties incurred, the LAT localizer should be used to precisely and reproducibly deliver the intended amount of radiation prescribed by CT protocols. CT protocols can be adjusted to minimize the expected change in average patient dose.

TU‐C‐103‐10: An Automated Technique to Measure CT Noise in Patient Images

Purpose: Phantoms are routinely used for image quality assessment of CT scanners. The image quality in phantom images, however, may not be representative of the image quality in patient images due to differences in physical composition and scan parameters. Access to image quality for every scan conducted would facilitate standardization of image quality across the broad range of CT scanner models present at many medical centers. Similarly, this data would aid in matching image quality between institutions. Therefore, the goal of this work was to develop an automated technique capable of measuring the noise for every CT scan conducted at a major medical center. Methods: The standard deviation was calculated by convolving the patient image with a 6mm square filter. Uniform regions of the patient anatomy were identified through histogram analysis and the standard deviations of the uniform regions were averaged to calculate the global noise of the image. The automated CT noise detection algorithm was applied to six clinical CT images. The results were compared to the noise measured by four observers placing ROIs in what they considered to be uniform sections of patient anatomy. Results: There was a one‐to‐one relationship between the noise measured by the computer algorithm and the mean observer noise with an R2 of 0.98. The average difference between the computer and observer measurements (4.7%) was less than the interobserver variability (4.8%). Running on a standard Lenovo Thinkpad T420 laptop, the computing time was less than 0.05 seconds per image. Conclusion: The noise measured using this automated technique agrees well with human observer measurements. The speed of the calculation indicates that it may be applied to every patient scan to monitor image quality on a per scan basis facilitating quality control on a level previously unachievable.

WE‐C‐103‐06: An Automated CT Quality Control Program

PURPOSE Daily image quality assessment is an essential part of a rigorous quality control program. Time constraints; however, make it difficult to fully analyze image quality daily. Therefore, the goal of this work was to develop a robust automated CT quality control program to extract meaningful image quality metrics including artifact analysis, noise texture measurement, and the calculation of a detectability index. METHODS The ACR CT phantom was scanned on five different CT scanners. The automated algorithm was used to calculate the standard metrics including HU accuracy, CNR, noise, and uniformity as well as more advanced metrics including the MTF, NPS, detectability index, and artifact detection. To validate the automated program, the HU accuracy, CNR, and noise metrics were compared to measurements conducted by a human observer. Additionally, the 10th percentile of the MTF (MTF 10) was compared the high-contrast resolution and the detectability index was compared to the low-contrast detectability recorded by the human observer using the Spearman rank sum correlation. RESULTS The HU, CNR, and noise determined by the automated algorithm agreed well with the human observer measurements (0.12%, 5.1%, and 7.1% difference for the HU, CNR, and noise respectively). Further, there was a strong correlation between MTF10 and the observer high-contrast resolution as well as the detectability index and the low-contrast detectability (rho=1 in both cases). CONCLUSION There was a strong agreement between the results of the automated quality control program and the human observer measurements. Further, the Fourier based MTF and detectability index were found to correlate strongly with observer image quality assessment. Therefore, this automated quality control program offers a viable alternative for routine image quality assessment of the ACR phantom.