Deep learning-based image restoration algorithm for coronary CT angiography.

The purpose of this study was to investigate whether the novel image-based noise reduction software (NRS) improves image quality, and to assess the feasibility of using this software in combination with hybrid iterative reconstruction (IR) in image quality on thin-slice abdominal CT. In this retrospective study, 54 patients who underwent dynamic liver CT between April and July 2017 and had a body mass index higher than 25 kg/m2 were included. Three image sets of each patient were reconstructed as follows: hybrid IR images with 1-mm slice thickness (group A), hybrid IR images with 5-mm slice thickness (group B), and hybrid IR images with 1-mm slice thickness denoised using NRS (group C). The mean image noise and contrast-to-noise ratio relative to the muscle of the aorta and liver were assessed. Subjective image quality was evaluated by two radiologists for sharpness, noise, contrast, and overall quality using 5-point scales. The mean image noise was significantly lower in group C than in group A (p < 0.01), but no significant difference was observed between groups B and C. The contrast-to-noise ratio was significantly higher in group C than in group A (p < 0.01 and p = 0.01, respectively). Subjective image quality was also significantly higher in group C than in group A (p < 0.01), in terms of noise and overall quality, but not in terms of sharpness and contrast (p = 0.65 and 0.07, respectively). The contrast of images in group C was greater than that in group A, but this difference was not significant. Compared with hybrid IR alone, the novel NRS combined with a hybrid IR could result in significant noise reduction without sacrificing image quality on CT. This combined approach will likely be particularly useful for thin-slice abdominal CT examinations of overweight patients.

The purpose of our study was to compare the diagnostic performance of coronary CT angiography (CTA) subjected to model-based iterative reconstruction (IR) or hybrid IR to rule out coronary in-stent restenosis. We enrolled 16 patients who harboured 22 coronary stents. They underwent coronary CTA on a 320-slice CT scanner. The images were reconstructed with hybrid IR (AIDR 3D) and model-based IR (FIRST) algorithms. We calculated the stent lumen attenuation increase ratio and measured the visible stent lumen diameter. Two blinded observers visually graded the likelihood of in-stent restenosis (lesions ≥ 50%) on hybrid IR and FIRST images. The stent lumen attenuation increase ratio on FIRST- was lower than on AIDR 3D images (0.20 vs 0.32). The ratio of the visible- compared to the true stent lumen diameter was higher on FIRST- than AIDR 3D images (52.5 vs 47.5%). Invasive coronary angiography identified five stents (22.7%) with significant in-stent restenosis. The use of FIRST improved the sensitivity (60 vs 100%), positive (75.0 vs 83.3%) and negative predictive value (88.9 vs 100%) and the accuracy (86.4 vs 95.5%) for the detection of in-stent restenosis. Specificity was 94.1% for both reconstruction methods. The model-based IR algorithm may improve diagnostic performance for the detection of in-stent restenosis. Advances in knowledge: Compared to hybrid IR, the new model-based IR algorithm reduced blooming artefacts and improved the image quality. It can be expected to improve diagnostic performance for the detection of in-stent restenosis on coronary CTA images.

To investigate the feasibility of low-tube-voltage (80 kVp) coronary CT angiography (CCTA) combined with contrast medium (CM) reduction and iterative model reconstruction (IMR) on patients with standard body mass index compared with clinical routine protocol. Retrospectively gated helical CCTA scans were acquired using a 256-slice multi-slice CT (Brilliance iCT; Philips Healthcare, Cleveland, OH) on 94 patients with standard body mass index (20-25 kg m(-2)) who were randomly assigned into 2 groups. The scan protocol for Group 1 was 100 kVp and 600 mAs with 70 ml CM at an injection rate of 4.5-5.5 ml s(-1); images were reconstructed by a hybrid iterative reconstruction technique (iDose(4); Philips Healthcare). Group 2 was scanned at 80 kVp and 600 mAs with 35 ml CM at an injection rate of 3.5-4.5 ml s(-1); images were reconstructed with IMR. Objective measurements such as the mean image noise and contrast-to-noise ratio of the two groups were measured on CT images and compared using the paired t-test. In addition, a subjective image quality evaluation was performed by two radiologists who were blinded to the scan protocol, using a 5-point scale [1 (poor) to 5 (excellent)]. The results of the two groups were compared using Mann-Whitney U test. The iodine delivery rate of Group 2 was 1.0 ± 0.5 gI s(-1) compared with 2.1 ± 0.5 gI s(-1) in Group 1 resulting in a reduction of 52.4%. In addition, an effective radiation dose reduction of 56.4% was achieved in Group 2 (2.4 ± 1.2 mSv) compared with Group 1 (5.5 ± 1.4 mSv). The mean CT attenuation, contrast-to-noise ratio and image quality of all segments in Group 2 were significantly improved compared with those in Group 1 (all, p < 0.01). The use of IMR along with a low tube voltage (80 kVp) combined with a low CM protocol for CCTA can reduce both radiation and CM dose with improved image quality. In this study, we used a novel knowledge-based IMR which remarkably reduced the image noise. We compared the quality of the images obtained when the tube voltage was reduced to 80 kVp and that of those obtained according to the clinical routine protocols to determine whether ultra-low-dose imaging plus IMR is feasible in CCTA scans. We found that a low dose protocol combined with 80 kVp and reduced CM for CCTA can reduce both radiation dose and CM dose with improved image quality by the use of IMR in non-obese patients.

Streak artifacts induced by irregular arm positioning have been an issue in diagnosing the abdomen. To illustrate the risk of misdiagnosis in abdominal computed tomography (CT) of patients with irregular arm positioning through a case-by-case evaluation and to test if it can be solved by the artificial intelligence iterative reconstruction (AIIR) algorithm. By reviewing 5220 cases of chest and thoracoabdominal CT, 64 patients with irregular arm positioning were enrolled, whose image data were reconstructed using AIIR in addition to routine hybrid iterative reconstruction (HIR). Lesion detection for livers, spleens, kidneys, gallbladders, and pancreas on AIIR images, performed by two radiologists, was compared with those on HIR images. Discrepancies arising from AIIR images included both cases with additional abnormalities and those with corrections made on previous detections. For cases with discrepancies, artifact scores for organs where discrepancies were found, and contrast-to-noise ratios (CNRs) of cysts with discrepancies were compared between two image sets. Additional abnormalities were detected for 15 cases: additional liver cirrhosis (n=2); additional gallbladder stone (n=1); additional cholecystitis (n=1), additional spleen nodule (n=1); additional kidney cysts (n=8); additional liver cysts (3); and additional spleen cyst (n=1). A spleen contusion was corrected for one case. All involved artifact scores were improved on AIIR images. CNRs of involved liver, kidney, and spleen cysts were improved by up to 539.7%, 538.5%, and 245.5%, respectively. Irregular arm positioning may induce a variety of misdiagnoses in abdominal CT, which is almost totally avoidable by the AIIR algorithm.

We investigated the impact of model-based iterative reconstruction (MBIR) on 320-detector row computed tomography angiography (CTA) in infants with complex congenital heart disease (CHD). Seventy infants with complex CHD who underwent 320-detector row CTA (40 boys and 30 girls; age range, 0-22 months; median age, 60 days) were retrospectively evaluated. First, the images were reconstructed by filtered back projection (FBP), hybrid iterative reconstruction (HIR), or MBIR in 20 cases, and variables were compared among the three iterative reconstruction methods (IR test). Second, the variables were compared between 25 cases scanned using HIR and 25 cases scanned using MBIR, with a 20 standard deviation noise level for both. Attenuation values and contrast-to-noise ratios (CNRs) of the great vessels and heart chambers were calculated. Total dose-length products were recorded for all patients (radiation dose: RD test). In the IR test, the mean CNR values were 4.8±1.3 for FBP, 6.9±1.4 for HIR, and 8.2±1.7 for MBIR (P < 0.0001). The best subjective image qualities in the great vessels and heart chambers were obtained with MBIR. In RD testing, no significant differences between HIR and MBIR in image quality (CNR: HIR, 8.4±2.4; MBIR, 8.3±2.4) were observed. The effective dose was significantly lower for MBIR than for HIR (0.7±0.2 vs. 1.1±0.3 mSv; P < 0.001). The MBIR algorithm significantly improved image quality and decreased radiation exposure in 320-row CTA of infants with complex CHD, providing an alternative to FBP or HIR that is both safer and produces better results.

Iterative model reconstruction reduces calcified plaque volume in coronary CT angiography

We retrospectively examined image quality (IQ) of thin-slice virtual monochromatic imaging (VMI) of half-iodine-load, abdominopelvic, contrast-enhanced CT (CECT) by dual-energy CT (DECT) with deep learning image reconstruction (DLIR). In 28 oncology patients with moderate-to-severe renal impairment undergoing half-iodine-load (300 mgI/kg) CECT by DECT during the nephrographic phase, we reconstructed VMI at 40-70 keV with a slice thickness of 0.625 mm using filtered back-projection (FBP), hybrid iterative reconstruction (HIR), and DLIR; measured contrast-noise ratio (CNR) of the liver, spleen, aorta, portal vein, and prostate/uterus; and determined the optimal keV to achieve the maximal CNR. At the optimal keV, two independent radiologists compared each organ's CNR and subjective IQ scores among FBP, HIR, and DLIR to subjectively grade image noise, contrast, sharpness, delineation of small structures, and overall IQ. CNR of each organ increased continuously from 70 to 40 keV using FBP, HIR, and DLIR. At 40 keV, CNR of the prostate/uterus was significantly higher with DLIR than with FBP; however, CNR was similar between FBP and HIR and between HIR and DLIR. The CNR of all other organs increased significantly from FBP to HIR to DLIR (P < 0.05). All IQ scores significantly improved from FBP to HIR to DLIR (P < 0.05) and were acceptable in all patients with DLIR only. The combination of 40 keV and DLIR offers the maximal CNR and a subjectively acceptable IQ for thin-slice VMI of half-iodine-load CECT.

Improving head and neck CTA with hybrid and model-based iterative reconstruction techniques

The Influence of Iterative Reconstruction on Coronary Artery Calcium Scoring—Phantom and Clinical Studies

The aim of this study was to conduct a comprehensive assessment of image quality in coronary computed tomography angiography (CCTA) reconstructed using deep learning image reconstruction (DLIR) technique. Objective and subjective assessment methods were utilized to analyze intraindividual image quality and to compare it with conventionally used hybrid iterative reconstruction algorithm (ASiR-V) regarding their dependency to heart rate, the main cause of motion blurring in CCTA. A total of 500 consecutive patients scheduled to undergo clinically indicated CCTA under prospective electrocardiography-triggering on a 256-slice computed tomography scanner were included. The image data were reconstructed using two different reconstruction techniques (ASiR-V and TrueFidelity), enabling a direct comparison. Signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and vessel contrast were used to determine objective image quality. Subjective image quality was assessed using a 4-point Likert scale. DLIR exhibited significantly better SNR in the aortic root, and this difference intensified with increasing heart rate. In contrast, SNR deteriorated with increasing heart rate for ASiR-V. The subjective comparison by two radiologists revealed significantly better ratings for TrueFidelity, both depending on heart rate and independent of it. For the left coronary artery, no significant differences were observed between the two algorithms in terms of SNR and CNR. In addition, vascular contrast did not differ significantly between DLIR and ASiR-V. Artificial intelligence in the form of DLIR offers significant advantages over manually created ASiR-V, objectively and subjectively evaluated, depending on and independent of heart rate. However, this study also identified drawbacks regarding vessel contrast and SNR, suggesting potential for further improvements.

PurposeTo assess whether the joint application of hybrid iterative reconstruction (HIR) and an adaptive filter (AF) could reduce streak artifacts and improve image quality of neck‐and‐shoulder computed tomography (CT).MethodsThis study included 96 patients with suspicious neck lesions who underwent a routine nonenhanced scan on a 64‐slice CT scanner. The raw data were reconstructed using four different settings: filtered back projection (FBP), HIR, FBP + AF, and HIR + AF. Regions of interest were manually drawn in erector spine, axillary fat, latissimus dorsi, and dorsal cervical fat. Mean and standard deviation (SD) of the CT number, signal‐to‐noise ratio (SNR), and contrast‐to‐noise ratio (CNR) were obtained and compared using Wilcoxon signed‐rank tests. The qualitative assessments of five factors were compared by two independent investigators.ResultsCompared to the other three settings, HIR + AF reduced noise in the area where the streak artifact of the lower neck were most serious (SD; all p ≤ 0.001). The SNR and CNR were improved significantly (all p ≤ 0.001). Compared to the other three settings, HIR + AF showed a significant improvement in CT image quality regarding the visibility of suspicious lesions, the extent of streaking artifacts, noise, soft‐tissue contrast, and visualization of small structures (all p ≤ 0.02).ConclusionsThe combination of HIR and AF can significantly reduce streaking artifacts and improve image quality in neck‐and‐shoulder CT imaging.

ObjectiveTo investigate the image quality of ultralow-dose CT (ULDCT) of the chest reconstructed using a cycle-consistent generative adversarial network (CycleGAN)-based deep learning method in the evaluation of pulmonary tuberculosis.Materials and MethodsBetween June 2019 and November 2019, 103 patients (mean age, 40.8 ± 13.6 years; 61 men and 42 women) with pulmonary tuberculosis were prospectively enrolled to undergo standard-dose CT (120 kVp with automated exposure control), followed immediately by ULDCT (80 kVp and 10 mAs). The images of the two successive scans were used to train the CycleGAN framework for image-to-image translation. The denoising efficacy of the CycleGAN algorithm was compared with that of hybrid and model-based iterative reconstruction. Repeated-measures analysis of variance and Wilcoxon signed-rank test were performed to compare the objective measurements and the subjective image quality scores, respectively.ResultsWith the optimized CycleGAN denoising model, using the ULDCT images as input, the peak signal-to-noise ratio and structural similarity index improved by 2.0 dB and 0.21, respectively. The CycleGAN-generated denoised ULDCT images typically provided satisfactory image quality for optimal visibility of anatomic structures and pathological findings, with a lower level of image noise (mean ± standard deviation [SD], 19.5 ± 3.0 Hounsfield unit [HU]) than that of the hybrid (66.3 ± 10.5 HU, p < 0.001) and a similar noise level to model-based iterative reconstruction (19.6 ± 2.6 HU, p > 0.908). The CycleGAN-generated images showed the highest contrast-to-noise ratios for the pulmonary lesions, followed by the model-based and hybrid iterative reconstruction. The mean effective radiation dose of ULDCT was 0.12 mSv with a mean 93.9% reduction compared to standard-dose CT.ConclusionThe optimized CycleGAN technique may allow the synthesis of diagnostically acceptable images from ULDCT of the chest for the evaluation of pulmonary tuberculosis.

We evaluated the feasibility of sub-millisievert (mSv) coronary CT angiography (CCTA) using low tube voltage, prospective ECG gating, and a knowledge-based iterative model reconstruction algorithm. Twenty-four non-obese healthy subjects (M:F 13:11; mean age 50.2±7.8years) were enrolled. Three sets of CT images were reconstructed using three different reconstruction methods: filtered back projection (FBP), iterative reconstruction (IR), and knowledge-based iterative model reconstruction (IMR). The scanning parameters were as follows: step-and-shoot axial scanning, 80kVp, and 200mAs. On the three sets of CT images, the attenuation and image noise values were measured at the aortic root. The signal-to-noise ratio (SNR) and the contrast-to-noise ratio (CNR) were calculated at the proximal right coronary artery and the left main coronary artery. The qualitative image quality of the CCTA with IMR was assessed using a 4-point grading scale (grade 1, poor; grade 4, excellent). The mean radiation dose of the CCTA was 0.89±0.09mSv. The attenuation values with IMR were not different from those of other reconstruction methods. The image noise with IMR was significantly lower than with IR and FBP. Compared to FBP, the noise reduction rate of IMR was 69%. The SNR and CNR of CCTA with IMR were significantly higher than with FBP or IR. On the qualitative analysis with IMR, all included segments were diagnostic (grades 2, 3, and 4), and the mean image quality score was 3.6±0.6. In conclusion, CCTA with low tube voltage, prospective ECG gating, and an IMR algorithm might be a feasible method that allows for sub-millisievert radiation doses and good image quality when used with non-obese subjects.

The increasing number of coronary computed tomography angiography (CCTA) requests raised concerns about dose exposure. New dose reduction strategies based on artificial intelligence have been proposed to overcome limitations of iterative reconstruction (IR) algorithms. Our prospective study sought to explore the added value of deep-learning image reconstruction (DLIR) in comparison with a hybrid IR algorithm (adaptive statistical iterative reconstruction-veo [ASiR-V]) in CCTA, even in clinical challenging scenarios, as obesity, heavily calcified vessels and coronary stents. We prospectively included 103 consecutive patients who underwent CCTA. Data sets were reconstructed with ASiR-V and DLIR. For each reconstruction signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) was calculated, and qualitative assessment was made with a four-point Likert scale by two independent and blinded radiologists with different expertise. Both SNR and CNR were significantly higher in DLIR (SNR-DLIR median value [interquartile range] of 13.89 [11.06-16.35] and SNR-ASiR-V 25.42 [22.46-32.22], P < 0.001; CNR-DLIR 16.84 [9.83-27.08] vs CNR-ASiR-V 10.09 [5.69-13.5], P < 0.001).Median qualitative score was 4 for DLIR images versus 3 for ASiR-V ( P < 0.001), with a good interreader reliability [intraclass correlation coefficient(2,1)e intraclass correlation coefficient(3,1) 0.60 for DLIR and 0.62 and 0.73 for ASiR-V].In the obese and in the "calcifications and stents" groups, DLIR showed significantly higher values of SNR (24.23 vs 11.11, P < 0.001 and 24.55 vs 14.09, P < 0.001, respectively) and CNR (16.08 vs 8.04, P = 0.008 and 17.31 vs 10.14, P = 0.003) and image quality. Deep-learning image reconstruction in CCTA allows better SNR, CNR, and qualitative assessment than ASiR-V, with an added value in the most challenging clinical scenarios.

Non-contrast cerebral computed tomography (CT) is frequently performed as a first-line diagnostic approach in patients with suspected ischemic stroke. The purpose of this study was to evaluate the performance of hybrid and model-based iterative image reconstruction for standard-dose (SD) and low-dose (LD) non-contrast cerebral imaging by multi-detector CT (MDCT). We retrospectively analyzed 131 patients with suspected ischemic stroke (mean age: 74.2 ± 14.3 years, 67 females) who underwent initial MDCT with a SD protocol (300 mAs) as well as follow-up MDCT after a maximum of 10 days with a LD protocol (200 mAs). Ischemic demarcation was detected in 26 patients for initial and in 64 patients for follow-up imaging, with diffusion-weighted magnetic resonance imaging (MRI) confirming ischemia in all of those patients. The non-contrast cerebral MDCT images were reconstructed using hybrid (Philips “iDose4”) and model-based iterative (Philips “IMR3”) reconstruction algorithms. Two readers assessed overall image quality, anatomic detail, differentiation of gray matter (GM)/white matter (WM), and conspicuity of ischemic demarcation, if any. Quantitative assessment included signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) calculations for WM, GM, and demarcated areas. Ischemic demarcation was detected in all MDCT images of affected patients by both readers, irrespective of the reconstruction method used. For LD imaging, anatomic detail and GM/WM differentiation was significantly better when using the model-based iterative compared to the hybrid reconstruction method. Furthermore, CNR of GM/WM as well as the SNR of WM and GM of healthy brain tissue were significantly higher for LD images with model-based iterative reconstruction when compared to SD or LD images reconstructed with the hybrid algorithm. For patients with ischemic demarcation, there was a significant difference between images using hybrid versus model-based iterative reconstruction for CNR of ischemic/contralateral unaffected areas (mean ± standard deviation: SD_IMR: 4.4 ± 3.1, SD_iDose: 3.5 ± 2.3, P < 0.0001; LD_IMR: 4.6 ± 2.9, LD_iDose: 3.2 ± 2.1, P < 0.0001). In conclusion, model-based iterative reconstruction provides higher CNR and SNR without significant loss of image quality for non-enhanced cerebral MDCT.