Reducing the risk of hallucinations with interpretable deep learning models for low-dose CT denoising: comparative performance analysis.

Objective.Reducing CT radiation dose is an often proposed measure to enhance patient safety, which, however results in increased image noise, translating into degradation of clinical image quality. Several deep learning methods have been proposed for low-dose CT (LDCT) denoising. The high risks posed by possible hallucinations in clinical images necessitate methods which aid the interpretation of deep learning networks. In this study, we aim to use qualitative reader studies and quantitative radiomics studies to assess the perceived quality, signal preservation and statistical feature preservation of LDCT volumes denoised by deep learning. We aim to compare interpretable deep learning methods with classical deep neural networks in clinical denoising performance.Approach.We conducted an image quality analysis study to assess the image quality of the denoised volumes based on four criteria to assess the perceived image quality. We subsequently conduct a lesion detection/segmentation study to assess the impact of denoising on signal detectability. Finally, a radiomic analysis study was performed to observe the quantitative and statistical similarity of the denoised images to standard dose CT (SDCT) images.Main results.The use of specific deep learning based algorithms generate denoised volumes which are qualitatively inferior to SDCT volumes(p< 0.05). Contrary to previous literature, denoising the volumes did not reduce the accuracy of the segmentation (p> 0.05). The denoised volumes, in most cases, generated radiomics features which were statistically similar to those generated from SDCT volumes (p> 0.05).Significance.Our results show that the denoised volumes have a lower perceived quality than SDCT volumes. Noise and denoising do not significantly affect detectability of the abdominal lesions. Denoised volumes also contain statistically identical features to SDCT volumes.

Limited parameter denoising for low-dose X-ray computed tomography using deep reinforcement learning.

BACKGROUND: The use of deep learning has successfully solved several problems in the field of medical imaging. Deep learning has been applied to the CT denoising problem successfully. However, the use of deep learning requires large amounts of data to train deep convolutional networks (CNNs). Moreover, due to the large parameter count, such deep CNNs may cause unexpected results. PURPOSE: In this study, we introduce a novel CT denoising framework, which has interpretable behavior and provides useful results with limited data. METHODS: We employ bilateral filtering in both the projection and volume domains to remove noise. To account for nonstationary noise, we tune the σ parameters of the volume for every projection view and every volume pixel. The tuning is carried out by two deep CNNs. Due to the impracticality of labeling, the two-deep CNNs are trained via a Deep-Q reinforcement learning task. The reward for the task is generated by using a custom reward function represented by a neural network. Our experiments were carried out on abdominal scans for the Mayo Clinic the cancer imaging archive (TCIA) dataset and the American association of physicists in medicine (AAPM) Low Dose CT Grand Challenge. RESULTS: Our denoising framework has excellent denoising performance increasing the peak signal to noise ratio (PSNR) from 28.53 to 28.93 and increasing the structural similarity index (SSIM) from 0.8952 to 0.9204. We outperform several state-of-the-art deep CNNs, which have several orders of magnitude higher number of parameters (p-value [PSNR] = 0.000, p-value [SSIM] = 0.000). Our method does not introduce any blurring, which is introduced by mean squared error (MSE) loss-based methods, or any deep learning artifacts, which are introduced by wasserstein generative adversarial network (WGAN)-based models. Our ablation studies show that parameter tuning and using our reward network results in the best possible results. CONCLUSIONS: We present a novel CT denoising framework, which focuses on interpretability to deliver good denoising performance, especially with limited data. Our method outperforms state-of-the-art deep neural networks. Future work will be focused on accelerating our method and generalizing it to different geometries and body parts.

Quantum Iterative Reconstruction for Abdominal Photon-counting Detector CT Improves Image Quality.

Background An iterative reconstruction (IR) algorithm was introduced for clinical photon-counting detector (PCD) CT. Purpose To investigate the image quality and the optimal strength level of a quantum IR algorithm (QIR; Siemens Healthcare) for virtual monoenergetic images and polychromatic images (T3D) in a phantom and in patients undergoing portal venous abdominal PCD CT. Materials and Methods In this retrospective study, noise power spectrum (NPS) was measured in a water-filled phantom. Consecutive oncologic patients who underwent portal venous abdominal PCD CT between March and April 2021 were included. Virtual monoenergetic images at 60 keV and T3D were reconstructed without QIR (QIR-off; reference standard) and with QIR at four levels (QIR 1-4; index tests). Global noise index, contrast-to-noise ratio (CNR), and voxel-wise CT attenuation differences were measured. Noise and texture, artifacts, diagnostic confidence, and overall quality were assessed qualitatively. Conspicuity of hypodense liver lesions was rated by four readers. Parametric (analyses of variance, paired t tests) and nonparametric tests (Friedman, post hoc Wilcoxon signed-rank tests) were used to compare quantitative and qualitative image quality among reconstructions. Results In the phantom, NPS showed unchanged noise texture across reconstructions with maximum spatial frequency differences of 0.01 per millimeter. Fifty patients (mean age, 59 years ± 16 [standard deviation]; 31 women) were included. Global noise index was reduced from QIR-off to QIR-4 by 45% for 60 keV and by 44% for T3D (both, P < .001). CNR of the liver improved from QIR-off to QIR-4 by 74% for 60 keV and by 69% for T3D (both, P < .001). No evidence of difference was found in mean attenuation of fat and liver (P = .79-.84) and on a voxel-wise basis among reconstructions. Qualitatively, QIR-4 outperformed all reconstructions in every category for 60 keV and T3D (P value range, <.001 to .01). All four readers rated QIR-4 superior to other strengths for lesion conspicuity (P value range, <.001 to .04). Conclusion In portal venous abdominal photon-counting detector CT, an iterative reconstruction algorithm (QIR; Siemens Healthcare) at high strength levels improved image quality by reducing noise and improving contrast-to-noise ratio and lesion conspicuity without compromising image texture or CT attenuation values. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Sinitsyn in this issue.

Noise Texture Deviation: A Measure for Quantifying Artifacts in Computed Tomography Images With Iterative Reconstructions.

OBJECTIVES: The aims of this study were to introduce the measure noise texture deviation as quantitative parameter for evaluating iterative reconstruction (IR)-specific artifacts in computed tomography (CT) images and to test whether IR-specific artifacts, quantified through this measure, are reduced in advanced modeled IR (ADMIRE) as compared with sinogram-affirmed IR (SAFIRE) images of the liver ex vivo and in patients with hypodense liver lesions. MATERIALS AND METHODS: In the ex vivo study part, an abdominal phantom was used. In the institutional review board-approved in vivo study part, 40 consecutive patients (mean age, 63 years) with hypodense liver lesions undergoing abdominal CT in the portal-venous phase were included. Images were reconstructed with filtered back projection, with the second-generation IR algorithm SAFIRE and with the third-generation IR algorithm ADMIRE. Noise power spectra and noise texture deviation were calculated in the phantom; image noise was measured in the phantom and in patients. Two blinded readers evaluated all image data regarding IR-specific artifacts (plastic-like, blotchy appearance); patient data were evaluated regarding conspicuity and confidence for detecting hypodense liver lesions. RESULTS: Image noise was significantly reduced at increasing IR levels (P < 0.001) with both algorithms, with no significant differences between corresponding strength levels of SAFIRE and ADMIRE (all, P > 0.05). Noise power spectra were similar at corresponding strength levels of SAFIRE and ADMIRE (all, P > 0.05). Noise texture deviation in ADMIRE was reduced compared with corresponding strength levels of SAFIRE (all, P < 0.001) and strongly correlated with subjective IR-specific artifacts (r = 0.88, P < 0.001). Iterative reconstruction-specific artifacts were significantly reduced in ADMIRE compared with that in SAFIRE images at strength levels 3 or greater, both ex vivo and in vivo (all, P < 0.001). There were no significant differences in the readers' ratings of lesion conspicuity and lesion confidence in detecting hypodense liver lesions between SAFIRE and ADMIRE (P > 0.05). Only lesion conspicuity was superior with SAFIRE and ADMIRE compared with filtered back projection (all, P < 0.001). CONCLUSIONS: Noise texture deviation is a quantitative measure reflecting IR-specific artifacts and is reduced in CT images with ADMIRE compared with SAFIRE.

Impact of iterative metal artifact reduction on diagnostic image quality in patients with dental hardware.

Background Metal artifacts often impair diagnostic accuracy in computed tomography (CT) imaging. Therefore, effective and workflow implemented metal artifact reduction algorithms are crucial to gain higher diagnostic image quality in patients with metallic hardware. Purpose To assess the clinical performance of a novel iterative metal artifact reduction (iMAR) algorithm for CT in patients with dental fillings. Material and Methods Thirty consecutive patients scheduled for CT imaging and dental fillings were included in the analysis. All patients underwent CT imaging using a second generation dual-source CT scanner (120 kV single-energy; 100/Sn140 kV in dual-energy, 219 mAs, gantry rotation time 0.28-1/s, collimation 0.6 mm) as part of their clinical work-up. Post-processing included standard kernel (B49) and an iterative MAR algorithm. Image quality and diagnostic value were assessed qualitatively (Likert scale) and quantitatively (HU ± SD) by two reviewers independently. Results All 30 patients were included in the analysis, with equal reconstruction times for iMAR and standard reconstruction (17 s ± 0.5 vs. 19 s ± 0.5; P > 0.05). Visual image quality was significantly higher for iMAR as compared with standard reconstruction (3.8 ± 0.5 vs. 2.6 ± 0.5; P < 0.0001, respectively) and showed improved evaluation of adjacent anatomical structures. Similarly, HU-based measurements of degree of artifacts were significantly lower in the iMAR reconstructions as compared with the standard reconstruction (0.9 ± 1.6 vs. -20 ± 47; P < 0.05, respectively). Conclusion The tested iterative, raw-data based reconstruction MAR algorithm allows for a significant reduction of metal artifacts and improved evaluation of adjacent anatomical structures in the head and neck area in patients with dental hardware.

Noise Reduction in Abdominal Computed Tomography Applying Iterative Reconstruction (ADMIRE).

RATIONALE AND OBJECTIVES: The study aimed to compare image quality of filtered back projection (FBP) and iterative reconstruction (advanced modeled iterative reconstruction, ADMIRE) in contrast-enhanced computed tomography (CT) of the abdomen, and to assess the differences of reconstructions according to these methods. It also aimed to investigate the potential for noise reduction of ADMIRE for different reconstructed slice thicknesses. MATERIALS AND METHODS: CT data of the abdomen and pelvis were acquired using a 128-slice single-source CT system using automated kV selection and tube current adaption based on patients' anatomy. Raw data sets from patients scanned at 100 kV were selected, and images were reconstructed with slice thicknesses of 1 mm, 3 mm, and 5 mm, both with FBP and ADMIRE. Filter strength F1, F3, and F5 of the ADMIRE algorithm and the corresponding reconstruction kernels were used. In total, 58 raw data sets from 17 patients were used to reconstruct from the same raw data FBP and ADMIRE images, representing identical body regions. Identical regions of interest were placed at the same position of up to four images and image noise was measured. Differences of reconstructed images and detail preservation were tested using an image subtraction technique, and subjective image quality was assessed using a 5-point Likert scale. RESULTS: On average, for 1-mm slice thickness, noise reduction was 9.15% ± 2.4% with filter strength level F1, 30.2% ± 3.4% with F3, and 54.4% ± 7.0% with F5 as compared to FBP. For a slice thickness of 3 mm, noise reduction was 8.5% ± 3.7% with F1, 28.6% ± 3.9% with F3, and 52.2% ± 9.1% with F5. For 5 mm, the corresponding values are 8.9% ± 2.7%, 31.4% ± 2.8%, and 52.7% ± 7.7%. On subtraction images, edge information of tissue classes with a high attenuation gradient was found, but structures with small differences in attenuation were not detectable on subtraction images, confirming that no relevant details were lost in the iterative reconstruction process. CONCLUSIONS: ADMIRE is able to reduce image noise considerably (up to 50%) without any obvious negative impact on lesion depiction as assessed visually. Noise reduction of ADMIRE seems to be independent of slice thickness.

Comparison and Combination of Dual-Energy- and Iterative-Based Metal Artefact Reduction on Hip Prosthesis and Dental Implants.

PURPOSE: To compare and combine dual-energy based and iterative metal artefact reduction on hip prosthesis and dental implants in CT. MATERIAL AND METHODS: A total of 46 patients (women:50%,mean age:63±15years) with dental implants or hip prostheses (n = 30/20) were included and examined with a second-generation Dual Source Scanner. 120kV equivalent mixed-images were derived from reconstructions of the 100/Sn140kV source images using no metal artefact reduction (NOMAR) and iterative metal artefact reduction (IMAR). We then generated monoenergetic extrapolations at 130keV from source images without IMAR (DEMAR) or from source images with IMAR, (IMAR+DEMAR). The degree of metal artefact was quantified for NOMAR, IMAR, DEMAR and IMAR+DEMAR using a Fourier-based method and subjectively rated on a five point Likert scale by two independent readers. RESULTS: In subjects with hip prosthesis, DEMAR and IMAR resulted in significantly reduced artefacts compared to standard reconstructions (33% vs. 56%; for DEMAR and IMAR; respectively, p<0.005), but the degree of artefact reduction was significantly higher for IMAR (all p<0.005). In contrast, in subjects with dental implants only IMAR showed a significant reduction of artefacts whereas DEMAR did not (71%, vs. 8% p<0.01 and p = 0.1; respectively). Furthermore, the combination of IMAR with DEMAR resulted in additionally reduced artefacts (Hip prosthesis: 47%, dental implants 18%; both p<0.0001). CONCLUSION: IMAR allows for significantly higher reduction of metal artefacts caused by hip prostheses and dental implants, compared to a dual energy based method. The combination of DE-source images with IMAR and subsequent monoenergetic extrapolation provides an incremental benefit compared to both single methods.

Coronary stent imaging with dual-source CT: assessment of lumen visibility using different convolution kernels and postprocessing filters.

BACKGROUND: Assesment of the coronary arteries after stent placement using coronary computed tomography angiography (CCTA) currently requires reconstruction of images with soft kernels for the assessment of atherosclerotic plaques and dedicated edge enhancing kernels for the evaluation of the stent lumen. PURPOSE: To evaluate a two-dimensional filter tool that provides instant postprocessing of images reconstructed with soft kernels into edge-enhanced images and vice versa and thus may eliminate the need for two separate reconstrcutions for the assessment of coronary artery stents using CCTA. MATERIAL AND METHODS: Twenty stents with a diameter of 3.0 mm placed in a vascular phantom were scanned with a dual-source CT using standard parameters. Images were reconstructed with a soft B30f and an edge-enhancing B46f kernel and postprocessed with the corresponding filter algorithm (F30 for B30f images; F46 for B46f images). The resulting four data-sets were evaluated for lumen visibility, intraluminal attenuation, and image noise by two independent readers. Results were validated in vivo against invasive coronary angiography in data-sets from patients with coronary artery stents. RESULTS: Average intraluminal attenuation was 552.6 HU, 527.3 HU, 207.9 HU, and 267.5 HU for B30f, F30, B46f, and F46 images, respectively (P < 0.0001). Average image noise was 11.3, 10.6, 19.2, and 15.0 HU, respectively (P < 0.0001). The visible stent diameter was significantly higher in the B46f (59.6%) and F46 images (54%) compared to the B30f (48.3%) and F30 (51.5%) images (P < 0.0001). In the patient study, lumen assessability was significantly better in B46f images than in F46 images. Sensitivity for stenosis detection was best in the original B46f images with a sensitivity of 67% and a specificity of 94%. CONCLUSION: The postprocessing filter reduces image noise, however currently it does not offer an alternative to image reconstruction using the edge-enhancing kernels for the evaluation of the stent lumen.

Initial results of a new generation dual source CT system using only an in-plane comb filter for ultra-high resolution temporal bone imaging.

OBJECTIVES: To prospectively evaluate radiation dose and image quality of a third generation dual-source CT (DSCT) without z-axis filter behind the patient for temporal bone CT. METHODS: Forty-five patients were either examined on a first, second, or third generation DSCT in an ultra-high-resolution (UHR) temporal bone-imaging mode. On the third generation DSCT system, the tighter focal spot of 0.2 mm(2) removes the necessity for an additional z-axis-filter, leading to an improved z-axis radiation dose efficiency. Images of 0.4 mm were reconstructed using standard filtered-back-projection or iterative reconstruction (IR) technique for previous generations of DSCT and a novel IR algorithm for the third generation DSCT. Radiation dose and image quality were compared between the three DSCT systems. RESULTS: The statistically significantly highest subjective and objective image quality was evaluated for the third generation DSCT when compared to the first or second generation DSCT systems (all p < 0.05). Total effective dose was 63%/39% lower for the third generation examination as compared to the first and second generation DSCT. CONCLUSIONS: Temporal bone imaging without z-axis-UHR-filter and a novel third generation IR algorithm allows for significantly higher image quality while lowering effective dose when compared to the first two generations of DSCTs. KEY POINTS: • Omitting the z-axis-filter allows a reduction in radiation dose of 50% • A smaller focal spot of 0.2 mm (2) significantly improves spatial resolution • Ultra-high-resolution temporal-bone-CT helps to gain diagnostic information of the middle/inner ear.

Ultralow-dose chest computed tomography for pulmonary nodule detection: first performance evaluation of single energy scanning with spectral shaping.

PURPOSE: The purpose of this study was to evaluate the image quality and sensitivity of ultralow radiation dose single-energy computed tomography (CT) with tin filtration for spectral shaping and iterative reconstructions for the detection of pulmonary nodules in a phantom setting. METHODS: Single-energy CT was performed using third-generation dual-source CT (SOMATOM Force; 2 × 192 slices) at 70 kVp, 100 kVp with tin filtration (100Sn kVp), and 150Sn kV with tube current-time product adjustments resulting in standard dose (CT volume dose index, 3.1 mGy/effective dose, 1.3 mSv at a scan length of 30 cm), 1/10th dose level (0.3 mGy/0.13 mSv), and 1/20th dose level (0.15 mGy/0.06 mSv). An anthropomorphic chest phantom simulating an intermediate-sized adult with randomly distributed solid pulmonary nodules of various sizes (2-10 mm; attenuation, 75 HU at 120 kVp) was used. Images were reconstructed with advanced model-based iterative reconstruction (ADMIRE; strength levels 3 and 5) and were compared with those acquired with second-generation dual-source CT at 120 kVp (reconstructed with filtered back projection) and sinogram-affirmed iterative reconstruction (strength level 3) at the lowest possible dose at 120 kVp (CT volume dose index, 0.28 mGy). One blinded reader measured image noise, and 2 blinded, independent readers determined overall image quality on a 5-grade scale (1 = nondiagnostic to 5 = excellent) and marked nodule localization with confidence rates on a 5-grade scale (1 = unsure to 5 = high confidence). The constructional drawing of the phantom served as reference standard for calculation of sensitivity. Two patients were included, for proof of concept, who were scanned with the 100Sn kVp protocol at the 1/10th and 1/20th dose level. RESULTS: Image noise was highest in the images acquired with second-generation dual-source CT and reconstructed with filtered back projection. At both the 1/10th and 1/20th dose levels, image noise at a tube voltage of 100Sn kVp was significantly lower than in the 70 kVp and 150Sn kV data sets (ADMIRE 3, P < 0.01; ADMIRE 5, P < 0.05). Sensitivity of nodule detection was lowest in images acquired with second-generation dual-source CT at 120 kVp and the lowest possible dose. Protocols at 100Sn kVp and ADMIRE 5 showed highest sensitivity at the 1/10th and 1/20th dose levels. Highest numbers of false-positives occurred in second-generation dual-source CT images (range, 12-15), whereas lowest numbers occurred in the 1/10th and 1/20th dose data sets acquired with third-generation dual-source CT at 100Sn kVp and reconstructed with ADMIRE strength levels 3 and 5 (total of 1 and 0 false-positives, respectively). Diagnostic confidence at 100Sn kVp was significantly higher than at 70 kVp or 150Sn kV (ADMIRE 3, P < 0.05; ADMIRE 5, P < 0.01) at both the 1/10th and 1/20th dose levels. Images of the 2 patients scanned with 100Sn kVp at the 1/10th and 1/20th dose levels were of diagnostic quality. CONCLUSIONS: Our study suggests that chest CT for the detection of pulmonary nodules can be performed with third-generation dual-source CT producing high image quality, sensitivity, and diagnostic confidence at a very low effective radiation dose of 0.06 mSv when using a single-energy protocol at 100 kVp with spectral shaping and when using advanced iterative reconstruction techniques.

Iterative image reconstruction techniques for CT coronary artery calcium quantification: comparison with traditional filtered back projection in vitro and in vivo.

PURPOSE: To investigate in vitro and in vivo the use of image-based and raw data-based iterative reconstruction algorithms for quantification of coronary artery calcium by using the Agatston score and subsequent cardiac risk stratification. MATERIALS AND METHODS: In vitro data were obtained by using a moving anthropomorphic cardiac phantom containing calcium inserts of different concentrations and sizes. With institutional review board approval and HIPAA compliance, coronary calcium imaging data of 110 consecutive patients (mean age ± standard deviation, 58.2 years ± 9.8; 48 men) were reconstructed with filtered back projection (FBP), iterative reconstruction in image space (IRIS), and sinogram-affirmed iterative reconstruction (SAFIRE). Image noise was measured and the Agatston score was obtained for all reconstructions. Assignment to Agatston scores and percentile-based cardiac risk categories was compared. Statistical analysis included the Cohen κ coefficient and Friedman and Wilcoxon testing. RESULTS: In vitro, mean Agatston scores ± standard deviation for FBP (638.9 ± 9.6), IRIS (622.7 ± 15.2), and SAFIRE (631.4 ± 17.6) were comparable (P = .30). The smallest phantom calcifications were more frequently detected when iterative reconstruction techniques were used. The Agatston scores in the patient cohort were not significantly different among FBP, IRIS, and SAFIRE in paired comparisons (median Agatston score [25th and 75th percentiles]: 76.0 [20.6, 243.9], 76.4 [22, 249.3], and 75.7 [21.5, 49.1], respectively; P = .20 each). Comparison of categorization based on Agatston score percentiles showed excellent agreement for both IRIS and SAFIRE with FBP (κ = 0.975 [0.942-1.00] and κ = 0.963 [0.922-1.00], respectively). The mean effective dose was 1.02 mSv ± 0.51. Mean image noise was significantly (P < .001) higher with FBP than that with iterative reconstructions. CONCLUSION: In comparison with FBP, iterative reconstruction techniques do not have a profound effect on the reproducible quantification of coronary calcium according to Agatston score and subsequent cardiac risk classification, although risk reclassification may occur in a small subset of subjects.

Spatial resolution improvement and dose reduction potential for inner ear CT imaging using a z-axis deconvolution technique.

PURPOSE: To assess the z-axis resolution improvement and dose reduction potential achieved using a z-axis deconvolution technique with iterative reconstruction (IR) relative to filtered backprojection (FBP) images created with the use of a z-axis comb filter. METHODS: Each of three phantoms were scanned with two different acquisition modes: (1) an ultrahigh resolution (UHR) scan mode that uses a comb filter in the fan angle direction to increase in-plane spatial resolution and (2) a z-axis ultrahigh spatial resolution (zUHR) scan mode that uses comb filters in both the fan and cone angle directions to improve both in-plane and z-axis spatial resolution. All other scanning parameters were identical. First, the ACR CT Accreditation phantom, rotated by 90° so that the high-contrast spatial resolution targets were parallel to the coronal plane, was scanned to assess limiting spatial resolution and image noise. Second, section sensitivity profiles (SSPs) were measured using a copper foil embedded in an acrylic cylinder and the full-width-at-half-maximum (FWHM) and full-width-at-tenth-maximum (FWTM) of the SSPs were calculated. Third, an anthropomorphic head phantom containing a human skull was scanned to assess clinical acceptability for imaging of the temporal bone. For each scan, FBP images were reconstructed for the zUHR scan using the narrowest image thickness available. For the CT accreditation phantom, zUHR images were also reconstructed using an IR algorithm (SAFIRE, Siemens Healthcare, Forchheim, Germany) to assess the influence of the IR algorithm on image noise. A z-axis deconvolution technique combined with the IR algorithm was used to reconstruct images at the narrowest image thickness possible from the UHR scan data. Images of the ACR and head phantoms were reformatted into the coronal plane. The head phantom images were evaluated by a neuroradiologist to assess acceptability for use in patients undergoing clinically indicated CT imaging of the temporal bone. RESULTS: The limiting spatial resolution was 12 lp/cm for the FBP-zUHR images and the IR-UHR images, although visual assessment indicated a slight improvement for the IR-UHR images. Image noise was 213.0, 181.8, and 153.5 for the FBP-zUHR, IR-zUHR, and IR-UHR images, respectively. While the FWHM was essentially the same for the FBP-zUHR and IR-UHR images, the FWTM of the IR-UHR images was almost 50% smaller compared to the FBP-zUHR images (0.83 vs 1.25 mm, respectively). Images of the anthropomorphic head phantom were judged to be of higher quality for the IR-UHR images compared to the FBP-zUHR images. CONCLUSIONS: With use of a z-axis deconvolution technique, z-axis spatial resolution was improved for scans acquired using a comb filter only in the fan angle direction relative to FBP images acquired with a comb filter in both the fan and cone angle directions. By avoiding use of the comb filter in the cone angle direction and use of an IR algorithm, image noise was substantially reduced for the same scanner output (CTDIvol). Thus, overall image quality (spatial resolution and image noise) can be maintained relative to the FBP-zUHR technique at a lower radiation dose.

Lowering kilovoltage to reduce radiation dose in contrast-enhanced abdominal CT: initial assessment of a prototype automated kilovoltage selection tool.

OBJECTIVE: The purpose of this study was to determine whether the use of an automated CT kilovoltage (kV) selection tool (Auto kV) can result in lower radiation dose without sacrificing image quality in contrast-enhanced abdominopelvic CT. MATERIALS AND METHODS: Tube potential, radiation dose, and iodine contrast-to-noise ratio (CNR) were retrospectively evaluated in 36 patients who underwent abdominopelvic CT with Auto kV, and compared with results from size-matched control patients using identical protocols. Two radiologists evaluated image quality (sharpness, noise, and diagnostic confidence) blinded to kV. Volume CT dose index (CTDI(vol)) was also compared with what each patient would have received from scanning at 120 kV. RESULTS: Mean (SD) CTDI(vol) was 16.0 (4.4) mGy after Auto kV versus 19.5 (4.0) mGy using standard 120-kV prescription and was 19.3 (6.0) mGy in control subjects (yielding dose reductions of 18.0% and 17.2%, respectively; p < 0.001 for both). Thirty of 36 patients were scanned at 100 kV (median dose reduction, 25%). Auto kV images were rated as very sharp in 33 (92%) and 36 (100%) cases versus 36 (100%) and 35 (97%) of the control cases, with all cases scored as having optimal noise. Readers had full diagnostic confidence in 34 (94%) and 36 (100%) of Auto kV cases; one reader scored "probably confident" in two cases (6%). Iodine CNRs for the aorta, liver, and portal vein were similar between Auto kV cases and control cases (p > 0.50, all comparisons). CONCLUSION: The use of an automated kV selection tool results in significant dose savings while maintaining diagnostic image quality and iodine CNR.

Performance evaluation of x-ray differential phase contrast computed tomography (PCT) with respect to medical imaging.

PURPOSE: To evaluate the maximum performance gain that can theoretically be achieved with differential phase contrast computed tomography (PCT) compared with absorption-based CT, applied to in vivo medical imaging. METHODS: We develop a mathematical framework for analyzing the performance of PCT relative to CT under ideal conditions. We validate our model by interpreting the results of published PCT experiments. Finally, we utilize our framework to evaluate the relative performance of PCT versus CT for in vivo medical imaging of the human body, investigating several clinically relevant material contrasts. RESULTS: We show that the performance of PCT relative to CT depends on the ratio of phase contrast and absorption contrast of the examined materials and increases with increasing effective coherence length and increasing spatial resolution. The introduced effective coherence length characterizes an experimental PCT setup; it comprises coherence of the beam as well as properties of the x-ray interferometer. Whole body medical CT will not benefit from phase-contrast imaging, because the higher phase contrast is overcompensated by the low coherence lengths of PCT setups with low-brilliance sources, and by limited spatial resolution. The relative performance of PCT, which is inferior to CT for all examined material contrasts at the resolution level of today's medical CT, can be improved by increasing spatial resolution at the expense of increased patient dose. At the break-even point of equal performance for PCT and CT, a radiation dose at least 1 order of magnitude higher than today is required. Mammographic CT already operates at higher spatial resolution and may benefit from PCT for some applications in terms of reduced patient dose at equal image quality. CONCLUSIONS: Phase-contrast imaging utilizing low-brilliance x-ray sources has limited potential for an application in routine whole body CT. Breast CT, however, may benefit from phase-contrast imaging. These conclusions are due to fundamental arguments and independent of whether technical issues (quality of gratings, etc.) can be solved. PCT will only be suitable for in vivo medical imaging if x-ray sources with much better spatial coherence are routinely available.

Normalized metal artifact reduction in head and neck computed tomography.

OBJECTIVE: Artifacts from dental hardware affect image quality and the visualization of lesions in the oral cavity and oropharynx in computed tomography (CT). Therefore, magnetic resonance imaging is considered the imaging modality of choice in this region. Standard methods for metal artifact reduction (MAR) in CT replace the metal-affected raw data by interpolation, which is prone to new artifacts. We developed a generalized normalization technique for MAR (NMAR) that aims to suppress algorithm-induced artifacts and validated the performance of this algorithm in a clinical trial. MATERIAL AND METHODS: A 3-dimensional forward projection identifies the metal-affected raw data in the original projections after metal is segmented in the image domain by thresholding. A prior image is used to normalize the projections before interpolation. The original raw data are divided pixel-wise by the projection data of the prior image and, after interpolation, are denormalized again. Data from 19 consecutive patients with metal artifacts from dental hardware were reconstructed with standard filtered backprojection (FBP), linear interpolation MAR (LIMAR), and NMAR. The image quality of slices containing metal was analyzed for the severity of artifacts and diagnostic value; magnetic resonance imaging performed the same day on a 3-T system served as a reference standard in all cases. RESULTS: A total of 260 slices containing metal dental hardware were analyzed. A total of 164 slices were nondiagnostic with FBP, 157 slices with LIMAR, and 87 slices with NMAR. The mean (SD) number of slices per patient with severe artifacts was 10.1 (3.7), 9.6 (4.6), and 5.4 (3.6) and the mean (SD) number of slices with artifacts affecting diagnostic confidence was 3.3 (1.7), 4.9 (2.9), and 3.7 (1.9) for FBP, LIMAR, and NMAR, respectively (P < 0.001). Pairwise comparison did not show significant differences between FBP and LIMAR (P = 0.40), but there were significant differences between FBP and NMAR as well as LIMAR and NMAR (both P < 0.001). Interobserver agreement was excellent (κ = 0.974). Two malignant lesions were unmasked with NMAR image reconstructions. No algorithm-related artifacts were detected in regions that did not contain metal in NMAR images. CONCLUSION: Normalized MAR has the potential to improve image quality in patients with artifacts from dental hardware and to improve the diagnostic accuracy of CT of the oral cavity and oropharynx.

Frequency split metal artifact reduction (FSMAR) in computed tomography.

PURPOSE: The problem of metal artifact reduction (MAR) is almost as old as the clinical use of computed tomography itself. When metal implants are present in the field of measurement, severe artifacts degrade the image quality and the diagnostic value of CT images. Up to now, no generally accepted solution to this issue has been found. In this work, a method based on a new MAR concept is presented: frequency split metal artifact reduction (FSMAR). It ensures efficient reduction of metal artifacts at high image quality with enhanced preservation of details close to metal implants. METHODS: FSMAR combines a raw data inpainting-based MAR method with an image-based frequency split approach. Many typical methods for metal artifact reduction are inpainting-based MAR methods and simply replace unreliable parts of the projection data, for example, by linear interpolation. Frequency split approaches were used in CT, for example, by combining two reconstruction methods in order to reduce cone-beam artifacts. FSMAR combines the high frequencies of an uncorrected image, where all available data were used for the reconstruction with the more reliable low frequencies of an image which was corrected with an inpainting-based MAR method. The algorithm is tested in combination with normalized metal artifact reduction (NMAR) and with a standard inpainting-based MAR approach. NMAR is a more sophisticated inpainting-based MAR method, which introduces less new artifacts which may result from interpolation errors. A quantitative evaluation was performed using the examples of a simulation of the XCAT phantom and a scan of a spine phantom. Further evaluation includes patients with different types of metal implants: hip prostheses, dental fillings, neurocoil, and spine fixation, which were scanned with a modern clinical dual source CT scanner. RESULTS: FSMAR ensures sharp edges and a preservation of anatomical details which is in many cases better than after applying an inpainting-based MAR method only. In contrast to other MAR methods, FSMAR yields images without the usual blurring close to implants. CONCLUSIONS: FSMAR should be used together with NMAR, a combination which ensures an accurate correction of both high and low frequencies. The algorithm is computationally inexpensive compared to iterative methods and methods with complex inpainting schemes. No parameters were chosen manually; it is ready for an application in clinical routine.

Automated attenuation-based tube potential selection for thoracoabdominal computed tomography angiography: improved dose effectiveness.

PURPOSE: To introduce a novel algorithm of automated attenuation-based tube potential selection and to assess its impact on image quality and radiation dose of body computed tomography angiography (CTA). MATERIALS AND METHODS: In all, 40 patients (mean age 71±11.8 years, body mass index (BMI) 25.7±3.8 kg/m², range 18.8-33.8 kg/m²) underwent 64-slice thoracoabdominal CTA (contrast material: 80 mL, 5 mL/s) using an automated tube potential selection algorithm (CAREkV), which optimizes tube-potential (70-140 kV) and tube-current (138.8±18.6 effective mAs, range 106-177 mAs) based on the attenuation profile of the topogram and on the diagnostic task. Image quality was semiquantitatively assessed by 2 blinded and independent readers (scores 1: excellent to 5: nondiagnostic). Attenuation and noise were measured by another 2 blinded and independent readers. Contrast-to-noise ratio was calculated. The CT dose index (CTDIvol) was recorded and compared with the estimated CTDIvol of a standard 120 kV protocol without using the algorithm in each patient. Selected tube potentials were correlated with BMI and attenuation of the topogram. RESULTS: Diagnostic image quality was obtained in all patients (excellent: 14; good: 21; moderate: 5; interreader agreement: κ=0.78). Mean attenuation, noise, and contrast-to-noise ratio were 260.8±63.5 Hounsfield units, 15.5±3.3 Hounsfield units, and 14±4.2, respectively, with good to excellent agreement between readers (r=0.50-0.99, P<0.01 each). Automated attenuation-based tube potential selection resulted in a kV-reduction from 120 to 100 kV in 23 patients and to 80 kV in 1 patient, whereas tube potential increased to 140 kV in 1 patient. Automatically selected tube potential showed a significant correlation with both BMI (r=0.427, P<0.05) and attenuation of the topogram (r=0.831, P<0.001). CTDIvol (7.95±2.6 mGy) was significantly lower when using the algorithm compared with the standard 120 kV protocol (10.59±1.8 mGy, P<0.001), corresponding to an overall dose reduction of 25.1%. CONCLUSION: Automated attenuation-based tube potential selection based on the attenuation profile of the topogram is feasible, provides a diagnostic image quality of body CTA, and reduces overall radiation dose by 25% as compared with a standard protocol with 120 kV.

Iterative image reconstruction techniques: Applications for cardiac CT.

BACKGROUND: Traditional limitations of cardiac CT are related to image noise, blooming artifacts from calcifications and stents, and radiation exposure. We evaluated whether these limitations can be ameliorated by the use of iterative reconstruction in image space (IRIS) instead of traditional filtered back projection (FBP) image reconstruction techniques. METHODS: We compared image reconstruction with the use of IRIS with traditional FBP for their effect on image quality, noise, volume of heavy coronary artery calcifications, and stents as a measure of "blooming" artifacts, and radiation dose at cardiac CT. The radiation dose comparison was performed as a matched pair analysis, whereas all other comparisons were performed within the same group of patients. RESULTS: The subjective image quality of IRIS reconstructions was rated higher than FBP reconstructions. Image noise was lower with IRIS than with FBP. The volume of stents and heavy coronary artery calcifications measured lower in IRIS reconstructed series compared with FBP. Similar levels of image noise were achieved with 80/100 kVp of tube voltage with IRIS compared with 120 kVp and FBP, resulting in a 62% reduction in effective dose. CONCLUSION: Our preliminary experiences suggest that IRIS incrementally improves the CT evaluation of coronary arteries, especially in challenging scenarios. Substantial radiation reduction seems feasible without associated increases in image noise.