Deep learning reconstruction vs standard reconstruction for abdominal CT: the influence of BMI.

OBJECTIVE: This study aimed to evaluate the image quality and lesion conspicuity of the deep learning image reconstruction (DLIR) algorithm compared with standard image reconstruction algorithms on abdominal enhanced computed tomography (CT) scanning with a wide range of body mass indexes (BMIs). METHODS: A total of 112 participants who underwent contrast-enhanced abdominal CT scans were divided into three groups according to BMIs: the 80-kVp group (BMI ≤ 23.9 kg/m2), 100-kVp group (BMI 24-28.9 kg/m2), and 120-kVp group (BMI ≥ 29 kg/m2). All images were reconstructed using filtered back projection (FBP), adaptive statistical iterative reconstruction-V of 50% level (IR), and DLIR at low, medium, and high levels (DL, DM, and DH, respectively). Subjective noise, artifact, overall image quality, and low- and high-contrast hepatic lesion conspicuity were all graded on a 5-point scale. The CT attenuation value (in HU), image noise, and contrast-to-noise ratio (CNR) were quantified and compared. RESULTS: DM and DH improved the qualitative and quantitative parameters compared with FBP and IR for all three BMI groups. DH had the lowest image noise and highest CNR value, while DM had the highest subjective overall image quality and low- and high-contrast lesion conspicuity scores for the three BMI groups. Based on the FBP, the improvement in image quality and lesion conspicuity of DM and DH images was greater in the 80-kVp group than in the 100-kVp and 120-kVp groups. CONCLUSION: For all BMIs, DLIR improves both image quality and hepatic lesion conspicuity, of which DM would be the best choice to balance both. CLINICAL RELEVANCE STATEMENT: The study suggests that utilizing DLIR, particularly at the medium level, can significantly enhance image quality and lesion visibility on abdominal CT scans across a wide range of BMIs. KEY POINTS: • DLIR improved the image quality and lesion conspicuity across a wide range of BMIs. • DLIR at medium level had the highest subjective parameters and lesion conspicuity scores among all reconstruction levels. • On the basis of the FBP, the 80-kVp group had improved image quality and lesion conspicuity more than the 100-kVp and 120-kVp groups.

Automated inversion time selection for late gadolinium-enhanced cardiac magnetic resonance imaging.

OBJECTIVES: To develop and share a deep learning method that can accurately identify optimal inversion time (TI) from multi-vendor, multi-institutional and multi-field strength inversion scout (TI scout) sequences for late gadolinium enhancement cardiac MRI. MATERIALS AND METHODS: Retrospective multicentre study conducted on 1136 1.5-T and 3-T cardiac MRI examinations from four centres and three scanner vendors. Deep learning models, comprising a convolutional neural network (CNN) that provides input to a long short-term memory (LSTM) network, were trained on TI scout pixel data from centres 1 to 3 to identify optimal TI, using ground truth annotations by two readers. Accuracy within 50 ms, mean absolute error (MAE), Lin's concordance coefficient (LCCC) and reduced major axis regression (RMAR) were used to select the best model from validation results, and applied to holdout test data. Robustness of the best-performing model was also tested on imaging data from centre 4. RESULTS: The best model (SE-ResNet18-LSTM) produced accuracy of 96.1%, MAE 22.9 ms and LCCC 0.47 compared to ground truth on the holdout test set and accuracy of 97.3%, MAE 15.2 ms and LCCC 0.64 when tested on unseen external (centre 4) data. Differences in vendor performance were observed, with greatest accuracy for the most commonly represented vendor in the training data. CONCLUSION: A deep learning model was developed that can identify optimal inversion time from TI scout images on multi-vendor data with high accuracy, including on previously unseen external data. We make this model available to the scientific community for further assessment or development. CLINICAL RELEVANCE STATEMENT: A robust automated inversion time selection tool for late gadolinium-enhanced imaging allows for reproducible and efficient cross-vendor inversion time selection. KEY POINTS: • A model comprising convolutional and recurrent neural networks was developed to extract optimal TI from TI scout images. • Model accuracy within 50 ms of ground truth on multi-vendor holdout and external data of 96.1% and 97.3% respectively was achieved. • This model could improve workflow efficiency and standardise optimal TI selection for consistent LGE imaging.

Normal cohorts in automated brain atrophy estimation: how many healthy subjects to include?

OBJECTIVES: This study investigates the influence of normal cohort (NC) size and the impact of different NCs on automated MRI-based brain atrophy estimation. METHODS: A pooled NC of 3945 subjects (NCpool) was retrospectively created from five publicly available cohorts. Voxel-wise gray matter volume atrophy maps were calculated for 48 Alzheimer's disease (AD) patients (55-82 years) using veganbagel and dynamic normal templates with an increasing number of healthy subjects randomly drawn from NCpool (initially three, and finally 100 subjects). Over 100 repeats of the process, the mean over a voxel-wise standard deviation of gray matter z-scores was established and plotted against the number of subjects in the templates. The knee point of these curves was defined as the minimum number of subjects required for consistent brain atrophy estimation. Atrophy maps were calculated using each NC for AD patients and matched healthy controls (HC). Two readers rated the extent of mesiotemporal atrophy to discriminate AD/HC. RESULTS: The maximum knee point was at 15 subjects. For 21 AD/21 HC, a sufficient number of subjects were available in each NC for validation. Readers agreed on the AD diagnosis in all cases (Kappa for the extent of atrophy, 0.98). No differences in diagnoses between NCs were observed (intraclass correlation coefficient, 0.91; Cochran's Q, p = 0.19). CONCLUSION: At least 15 subjects should be included in age- and sex-specific normal templates for consistent brain atrophy estimation. In the study's context, qualitative interpretation of regional atrophy allows reliable AD diagnosis with a high inter-reader agreement, irrespective of the NC used. CLINICAL RELEVANCE STATEMENT: The influence of normal cohorts (NCs) on automated brain atrophy estimation, typically comparing individual scans to NCs, remains largely unexplored. Our study establishes the minimum number of NC-subjects needed and demonstrates minimal impact of different NCs on regional atrophy estimation. KEY POINTS: • Software-based brain atrophy estimation often relies on normal cohorts for comparisons. • At least 15 subjects must be included in an age- and sex-specific normal cohort. • Using different normal cohorts does not influence regional atrophy estimation.

Joint radiomics and spatial distribution model for MRI-based discrimination of multiple sclerosis, neuromyelitis optica spectrum disorder, and myelin-oligodendrocyte-glycoprotein-IgG-associated disorder.

OBJECTIVE: To develop a discrimination pipeline concerning both radiomics and spatial distribution features of brain lesions for discrimination of multiple sclerosis (MS), aquaporin-4-IgG-seropositive neuromyelitis optica spectrum disorder (NMOSD), and myelin-oligodendrocyte-glycoprotein-IgG-associated disorder (MOGAD). METHODS: Hyperintensity T2 lesions were delineated in 212 brain MRI scans of MS (n = 63), NMOSD (n = 87), and MOGAD (n = 45) patients. To avoid the effect of fixed training/test dataset sampling when developing machine learning models, patients were allocated into 4 sub-groups for cross-validation. For each scan, 351 radiomics and 27 spatial distribution features were extracted. Three models, i.e., multi-lesion radiomics, spatial distribution, and joint models, were constructed using random forest and logistic regression algorithms for differentiating: MS from the others (MS models) and MOGAD from NMOSD (MOG-NMO models), respectively. Then, the joint models were combined with demographic characteristics (i.e., age and sex) to create MS and MOG-NMO discriminators, respectively, based on which a three-disease discrimination pipeline was generated and compared with radiologists. RESULTS: For classification of both MS-others and MOG-NMO, the joint models performed better than radiomics or spatial distribution model solely. The MS discriminator achieved AUC = 0.909 ± 0.027 and bias-corrected C-index = 0.909 ± 0.027, and the MOG-NMO discriminator achieved AUC = 0.880 ± 0.064 and bias-corrected C-index = 0.883 ± 0.068. The three-disease discrimination pipeline differentiated MS, NMOSD, and MOGAD patients with 75.0% accuracy, prominently outperforming the three radiologists (47.6%, 56.6%, and 66.0%). CONCLUSIONS: The proposed pipeline integrating multi-lesion radiomics and spatial distribution features could effectively differentiate MS, NMOSD, and MOGAD. CLINICAL RELEVANCE STATEMENT: The discrimination pipeline merging both radiomics and spatial distribution features of brain lesions may facilitate the differential diagnoses of multiple sclerosis, neuromyelitis optica spectrum disorder, and myelin-oligodendrocyte-glycoprotein-IgG-associated disorder. KEY POINTS: • Our study introduces an approach by combining radiomics and spatial distribution models. • The joint model exhibited superior performance in distinguishing multiple sclerosis from aquaporin-4-IgG-seropositive neuromyelitis optica spectrum disorder and myelin-oligodendrocyte-glycoprotein-IgG-associated disorder as well as discriminating the latter two diseases. • The three-disease discrimination pipeline showcased remarkable accuracy, surpassing the performance of experienced radiologists, highlighting its potential as a valuable diagnostic tool.

Deep learning reconstruction CT for liver metastases: low-dose dual-energy vs standard-dose single-energy.

OBJECTIVES: To assess image quality and liver metastasis detection of reduced-dose dual-energy CT (DECT) with deep learning image reconstruction (DLIR) compared to standard-dose single-energy CT (SECT) with DLIR or iterative reconstruction (IR). METHODS: In this prospective study, two groups of 40 participants each underwent abdominal contrast-enhanced scans with full-dose SECT (120-kVp images, DLIR and IR algorithms) or reduced-dose DECT (40- to 60-keV virtual monochromatic images [VMIs], DLIR algorithm), with 122 and 106 metastases, respectively. Groups were matched by age, sex ratio, body mass index, and cross-sectional area. Noise power spectrum of liver images and task-based transfer function of metastases were calculated to assess the noise texture and low-contrast resolution. The image noise, signal-to-noise ratios (SNR) of liver and portal vein, liver-to-lesion contrast-to-noise ratio (LLR), lesion conspicuity, lesion detection rate, and the subjective image quality metrics were compared between groups on 1.25-mm reconstructed images. RESULTS: Compared to 120-kVp images with IR, 40- and 50-keV VMIs with DLIR showed similar noise texture and LLR, similar or higher image noise and low-contrast resolution, improved SNR and lesion conspicuity, and similar or better perceptual image quality. When compared to 120-kVp images with DLIR, 50-keV VMIs with DLIR had similar low-contrast resolution, SNR, LLR, lesion conspicuity, and perceptual image quality but lower frequency noise texture and higher image noise. For the detection of hepatic metastases, reduced-dose DECT by 34% maintained observer lesion detection rates. CONCLUSION: DECT assisted with DLIR enables a 34% dose reduction for detecting hepatic metastases while maintaining comparable perceptual image quality to full-dose SECT. CLINICAL RELEVANCE STATEMENT: Reduced-dose dual-energy CT with deep learning image reconstruction is as accurate as standard-dose single-energy CT for the detection of liver metastases and saves more than 30% of the radiation dose. KEY POINTS: • The 40- and 50-keV virtual monochromatic images (VMIs) with deep learning image reconstruction (DLIR) improved lesion conspicuity compared with 120-kVp images with iterative reconstruction while providing similar or better perceptual image quality. • The 50-keV VMIs with DLIR provided comparable perceptual image quality and lesion conspicuity to 120-kVp images with DLIR. • The reduction of radiation by 34% by DLIR in low-keV VMIs is clinically sufficient for detecting low-contrast hepatic metastases.

Automated artificial intelligence-based three-dimensional comparison of orthodontic treatment outcomes with and without piezocision surgery.

OBJECTIVE(S): This study aims to evaluate the influence of the piezocision surgery in the orthodontic biomechanics, as well as in the magnitude and direction of tooth movement in the mandibular arch using novel artificial intelligence (AI)-automated tools. MATERIALS AND METHODS: Nineteen patients, who had piezocision performed in the lower arch at the beginning of treatment with the goal of accelerating tooth movement, were compared to 19 patients who did not receive piezocision. Cone beam computed tomography (CBCT) and intraoral scans (IOS) were acquired before and after orthodontic treatment. AI-automated dental tools were used to segment and locate landmarks in dental crowns from IOS and root canals from CBCT scans to quantify 3D tooth movement. Differences in mesial-distal, buccolingual, intrusion and extrusion linear movements, as well as tooth long axis angulation and rotation were compared. RESULTS: The treatment time for the control and experimental groups were 13.2 ± 5.06 and 13 ± 5.52 months respectively (P = .176). Overall, anterior and posterior tooth movement presented similar 3D linear and angular changes in the groups. The piezocision group demonstrated greater (P = .01) mesial long axis angulation of lower right first premolar (4.4 ± 6°) compared with control group (0.02 ± 4.9°), while the mesial rotation was significantly smaller (P = .008) in the experimental group (0.5 ± 7.8°) than in the control (8.5 ± 9.8°) considering the same tooth. CONCLUSION: The open source-automated dental tools facilitated the clinicians' assessment of piezocision treatment outcomes. The piezocision surgery prior to the orthodontic treatment did not decrease the treatment time and did not influence in the orthodontic biomechanics, leading to similar tooth movements compared to conventional treatment.

Using automated software evaluation to improve the performance of breast radiographers in tomosynthesis screening.

OBJECTIVE: To improve breast radiographers' individual performance by using automated software to assess the correctness of breast positioning and compression in tomosynthesis screening. MATERIALS AND METHODS: In this retrospective longitudinal analysis of prospective cohorts, six breast radiographers with varying experience in the field were asked to use automated software to improve their performance in breast compression and positioning. The software tool automatically analyzes craniocaudal (CC) and mediolateral oblique (MLO) views for their positioning quality by scoring them according to PGMI classifications (perfect, good, moderate, inadequate) and checking whether the compression pressure is within the target range. The positioning and compression data from the studies acquired before the start of the project were used as individual baselines, while the data obtained after the training were used to test whether conscious use of the software could help the radiographers improve their performance. The percentage of views rated perfect or good and the percentage of views in target compression were used as overall metrics to assess changes in performance. RESULTS: Following the use of the software, all radiographers significantly increased the percentage of images rated as perfect or good in both CCs and MLOs. Individual improvements ranged from 7 to 14% for CC and 10 to 16% for MLO views. Moreover, most radiographers exhibited improved compression performance in CCs, with improvements up to 16%. CONCLUSION: Active use of a software tool to automatically assess the correctness of breast compression and positioning in breast cancer screening can improve the performance of radiographers. CLINICAL RELEVANCE STATEMENT: This study suggests that the use of a software tool for automatically evaluating correctness of breast compression and positioning in breast cancer screening can improve the performance of radiographers on these metrics, which may ultimately lead to improved screening outcomes. KEY POINTS: • Proper breast positioning and compression are critical in breast cancer screening to ensure accurate diagnosis. • Active use of the software increased the quality of craniocaudal and mediolateral oblique views acquired by all radiographers. • Improved performance of radiographers is expected to improve screening outcomes.

Coronary calcium scoring on virtual non-contrast and virtual non-iodine reconstructions compared to true non-contrast images using photon-counting computed tomography.

OBJECTIVES: To compare coronary artery calcification (CAC) scores measured on virtual non-contrast (VNC) and virtual non-iodine (VNI) reconstructions computed from coronary computed tomography angiography (CCTA) using photon-counting computed tomography (PCCT) to true non-contrast (TNC) images. METHODS: We included 88 patients (mean age = 59 years ± 13.5, 69% male) who underwent a TNC coronary calcium scan followed by CCTA on PCCT. VNC images were reconstructed in 87 patients and VNI in 88 patients by virtually removing iodine from the CCTA images. For all reconstructions, CAC scores were determined, and patients were classified into risk categories. The overall agreement of the reconstructions was analyzed by Bland-Altman plots and the level of matching classifications. RESULTS: The median CAC score on TNC was 27.8 [0-360.4] compared to 8.5 [0.2-101.6] (p < 0.001) on VNC and 72.2 [1.3-398.8] (p < 0.001) on VNI. Bland-Altman plots depicted a bias of 148.8 (ICC = 0.82, p < 0.001) and - 57.7 (ICC = 0.95, p < 0.001) for VNC and VNI, respectively. Of all patients with CACTNC = 0, VNC reconstructions scored 63% of the patients correctly, while VNI scored 54% correctly. Of the patients with CACTNC > 0, VNC and VNI reconstructions detected the presence of coronary calcium in 90% and 92% of the patients. CACVNC tended to underestimate CAC score, whereas CACVNI overestimated, especially in the lower risk categories. According to the risk categories, VNC misclassified 55% of the patients, while VNI misclassified only 32%. CONCLUSION: Compared to TNC images, VNC underestimated and VNI overestimated the actual CAC scores. VNI reconstructions quantify and classify coronary calcification scores more accurately than VNC reconstructions. CLINICAL RELEVANCE STATEMENT: Photon-counting CT enables spectral imaging, which might obviate the need for non-contrast enhanced coronary calcium scoring, but optimization is necessary for the clinical implementation of the algorithms. KEY POINTS: • Photon-counting computed tomography uses spectral information to virtually remove the signal of contrast agents from contrast-enhanced scans. • Virtual non-contrast reconstructions tend to underestimate coronary artery calcium scores compared to true non-contrast images, while virtual non-iodine reconstructions tend to overestimate the calcium scores. • Virtual non-iodine reconstructions might obviate the need for non-contrast enhanced calcium scoring, but optimization is necessary for the clinical implementation of the algorithms.

Is it possible to use low-dose deep learning reconstruction for the detection of liver metastases on CT routinely?

OBJECTIVES: To compare the image quality and hepatic metastasis detection of low-dose deep learning image reconstruction (DLIR) with full-dose filtered back projection (FBP)/iterative reconstruction (IR). METHODS: A contrast-detail phantom consisting of low-contrast objects was scanned at five CT dose index levels (10, 6, 3, 2, and 1 mGy). A total of 154 participants with 305 hepatic lesions who underwent abdominal CT were enrolled in a prospective non-inferiority trial with a three-arm design based on phantom results. Data sets with full dosage (13.6 mGy) and low dosages (9.5, 6.8, or 4.1 mGy) were acquired from two consecutive portal venous acquisitions, respectively. All images were reconstructed with FBP (reference), IR (control), and DLIR (test). Eleven readers evaluated phantom data sets for object detectability using a two-alternative forced-choice approach. Non-inferiority analyses were performed to interpret the differences in image quality and metastasis detection of low-dose DLIR relative to full-dose FBP/IR. RESULTS: The phantom experiment showed the dose reduction potential from DLIR was up to 57% based on the reference FBP dose index. Radiation decreases of 30% and 50% resulted in non-inferior image quality and hepatic metastasis detection with DLIR compared to full-dose FBP/IR. Radiation reduction of 70% by DLIR performed inferiorly in detecting small metastases (< 1 cm) compared to full-dose FBP (difference: -0.112; 95% confidence interval [CI]: -0.178 to 0.047) and full-dose IR (difference: -0.123; 95% CI: -0.182 to 0.053) (p < 0.001). CONCLUSION: DLIR enables a 50% dose reduction for detecting low-contrast hepatic metastases while maintaining comparable image quality to full-dose FBP and IR. KEY POINTS: • Non-inferiority study showed that deep learning image reconstruction (DLIR) can reduce the dose to oncological patients with low-contrast lesions without compromising the diagnostic information. • Radiation dose levels for DLIR can be reduced to 50% of full-dose FBP and IR for detecting low-contrast hepatic metastases, while maintaining comparable image quality. • The reduction of radiation by 70% by DLIR is clinically acceptable but insufficient for detecting small low-contrast hepatic metastases (< 1 cm).

Visual ensemble selection of deep convolutional neural networks for 3D segmentation of breast tumors on dynamic contrast enhanced MRI.

OBJECTIVES: To develop a visual ensemble selection of deep convolutional neural networks (CNN) for 3D segmentation of breast tumors using T1-weighted dynamic contrast-enhanced (T1-DCE) MRI. METHODS: Multi-center 3D T1-DCE MRI (n = 141) were acquired for a cohort of patients diagnosed with locally advanced or aggressive breast cancer. Tumor lesions of 111 scans were equally divided between two radiologists and segmented for training. The additional 30 scans were segmented independently by both radiologists for testing. Three 3D U-Net models were trained using either post-contrast images or a combination of post-contrast and subtraction images fused at either the image or the feature level. Segmentation accuracy was evaluated quantitatively using the Dice similarity coefficient (DSC) and the Hausdorff distance (HD95) and scored qualitatively by a radiologist as excellent, useful, helpful, or unacceptable. Based on this score, a visual ensemble approach selecting the best segmentation among these three models was proposed. RESULTS: The mean and standard deviation of DSC and HD95 between the two radiologists were equal to 77.8 ± 10.0% and 5.2 ± 5.9 mm. Using the visual ensemble selection, a DSC and HD95 equal to 78.1 ± 16.2% and 14.1 ± 40.8 mm was reached. The qualitative assessment was excellent (resp. excellent or useful) in 50% (resp. 77%). CONCLUSION: Using subtraction images in addition to post-contrast images provided complementary information for 3D segmentation of breast lesions by CNN. A visual ensemble selection allowing the radiologist to select the most optimal segmentation obtained by the three 3D U-Net models achieved comparable results to inter-radiologist agreement, yielding 77% segmented volumes considered excellent or useful. KEY POINTS: • Deep convolutional neural networks were developed using T1-weighted post-contrast and subtraction MRI to perform automated 3D segmentation of breast tumors. • A visual ensemble selection allowing the radiologist to choose the best segmentation among the three 3D U-Net models outperformed each of the three models. • The visual ensemble selection provided clinically useful segmentations in 77% of cases, potentially allowing for a valuable reduction of the manual 3D segmentation workload for the radiologist and greatly facilitating quantitative studies on non-invasive biomarker in breast MRI.

Performance of 1-mm non-gated low-dose chest computed tomography using deep learning-based noise reduction for coronary artery calcium scoring.

OBJECTIVE: To investigate performance of 1-mm, sharp kernel, low-dose chest computed tomography (LDCT) for coronary artery calcium scoring (CACS) using deep learning (DL)-based denoising technique. METHODS: This retrospective, intra-individual comparative study consisted of four image datasets of 131 participants who underwent LDCT and calcium CT on the same day between January and February 2020; 1-mm LDCT with DL, 1-mm LDCT with iterative reconstruction (IR), 3-mm LDCT, and calcium CT. CACS from calcium CT were considered as reference and CACS were categorized as 0, 1-10, 11-100, 101-400, and > 400. We compared CACS from LDCTs with that from calcium CT. RESULTS: Mean CACS was 104.8 ± 249.1 and proportion of positive CACS was 45% (59/131). CACS from LDCT images tended to be underestimated than those from calcium CT: 1-mm LDCT with DL (93.5 ± 249.6, p = 0.002), 1-mm LDCT with IR (94.7 ± 249.9, p < 0.001), and 3-mm LDCT (90.3 ± 245.3, p = 0.004). All LDCT datasets showed excellent agreement with calcium CT: intraclass correlation coefficient (ICC) = 0.961 (95% confidence interval (CI), 0.945-0.972) for DL, 0.969 (95% CI, 0.956-0.978) for IR, and 0.952 (95% CI, 0.932-0.966) for 3-mm LDCT; weighted kappa for CACS classification, 0.930 (95% CI, 0.893-0.966) for 1-mm LDCT with DL, 0.908 (95% CI, 0.866-0.950) for 1-mm LDCT with IR, and 0.846 (95% CI, 0.780-0.912) for 3-mm LDCT. The accuracy of CACS classification of 1-mm LDCT with DL (90%) tended to be better than 1-mm LDCT with IR (87%) and 3-mm LDCT (84.7%) (p = 0.10). CONCLUSION: DL-based noise reduction algorithm can offer reliable calcium scores in 1-mm LDCT reconstructed with sharp kernel. KEY POINTS: • Deep learning (DL)-based noise reduction enables calcium scoring at 1-mm, sharp kernel reconstructed low-dose chest CT (LDCT). • Both iterative reconstruction and DL-based noise reduction underestimated calcium score, but agreement were excellent with those from calcium CT. • Accuracy of categorical classification of calcium scoring tended to be highest in 1-mm LDCT with DL compared to 1-mm LDCT with IR and 3-mm LDCT (90%, 87%, and 84.7%, p = 0.10).

Abdominal fat quantification using convolutional networks.

OBJECTIVES: To present software for automated adipose tissue quantification of abdominal magnetic resonance imaging (MRI) data using fully convolutional networks (FCN) and to evaluate its overall performance-accuracy, reliability, processing effort, and time-in comparison with an interactive reference method. MATERIALS AND METHODS: Single-center data of patients with obesity were analyzed retrospectively with institutional review board approval. Ground truth for subcutaneous (SAT) and visceral adipose tissue (VAT) segmentation was provided by semiautomated region-of-interest (ROI) histogram thresholding of 331 full abdominal image series. Automated analyses were implemented using UNet-based FCN architectures and data augmentation techniques. Cross-validation was performed on hold-out data using standard similarity and error measures. RESULTS: The FCN models reached Dice coefficients of up to 0.954 for SAT and 0.889 for VAT segmentation during cross-validation. Volumetric SAT (VAT) assessment resulted in a Pearson correlation coefficient of 0.999 (0.997), relative bias of 0.7% (0.8%), and standard deviation of 1.2% (3.1%). Intraclass correlation (coefficient of variation) within the same cohort was 0.999 (1.4%) for SAT and 0.996 (3.1%) for VAT. CONCLUSION: The presented methods for automated adipose-tissue quantification showed substantial improvements over common semiautomated approaches (no reader dependence, less effort) and thus provide a promising option for adipose tissue quantification. CLINICAL RELEVANCE STATEMENT: Deep learning techniques will likely enable image-based body composition analyses on a routine basis. The presented fully convolutional network models are well suited for full abdominopelvic adipose tissue quantification in patients with obesity. KEY POINTS: • This work compared the performance of different deep-learning approaches for adipose tissue quantification in patients with obesity. • Supervised deep learning-based methods using fully convolutional networks  were suited best. • Measures of accuracy were equal to or better than the operator-driven approach.

Improved image quality and dose reduction in abdominal CT with deep-learning reconstruction algorithm: a phantom study.

OBJECTIVES: To assess the impact of a new artificial intelligence deep-learning reconstruction (Precise Image; AI-DLR) algorithm on image quality against a hybrid iterative reconstruction (IR) algorithm in abdominal CT for different clinical indications. METHODS: Acquisitions on phantoms were performed at 5 dose levels (CTDIvol: 13/11/9/6/1.8 mGy). Raw data were reconstructed using level 4 of iDose4 (i4) and 3 levels of AI-DLR (Smoother/Smooth/Standard). Noise power spectrum (NPS), task-based transfer function (TTF) and detectability index (d') were computed: d' modelled detection of a liver metastasis (LM) and hepatocellular carcinoma at portal (HCCp) and arterial (HCCa) phases. Image quality was subjectively assessed on an anthropomorphic phantom by 2 radiologists. RESULTS: From Standard to Smoother levels, noise magnitude and average NPS spatial frequency decreased and the detectability (d') of all simulated lesions increased. For both inserts, TTF values were similar for all three AI-DLR levels from 13 to 6 mGy but decreased from Standard to Smoother levels at 1.8 mGy. Compared to the i4 used in clinical practice, d' values were higher using the Smoother and Smooth levels and close for the Standard level. For all dose levels, except at 1.8 mGy, radiologists considered images satisfactory for clinical use for the 3 levels of AI-DLR, but rated images too smooth using the Smoother level. CONCLUSION: Use of the Smooth and Smoother levels of AI-DLR reduces the image noise and improves the detectability of lesions and spatial resolution for standard and low-dose levels. Using the Smooth level is apparently the best compromise between the lowest dose level and adequate image quality. KEY POINTS: • Evaluation of the impact of a new artificial intelligence deep-learning reconstruction (AI-DLR) on image quality and dose compared to a hybrid iterative reconstruction (IR) algorithm. • The Smooth and Smoother levels of AI-DLR reduced the image noise and improved the detectability of lesions and spatial resolution for standard and low-dose levels. • The Smooth level seems the best compromise between the lowest dose level and adequate image quality.

Impact of different metal artifact reduction techniques in photon-counting computed tomography head and neck scans in patients with dental hardware.

OBJECTIVES: Metal artifacts remain a challenge in computed tomography. We investigated the potential of photon-counting computed tomography (PCD-CT) for metal artifact reduction using an iterative metal artifact reduction (iMAR) algorithm alone and in combination with high keV monoenergetic images (140 keV) in patients with dental hardware. MATERIAL AND METHODS: Consecutive patients with dental implants were prospectively included in this study and received PCD-CT imaging of the craniofacial area. Four series were reconstructed (standard [PCD-CTstd], monoenergetic at 140 keV [PCD-CT140keV], iMAR corrected [PCD-CTiMAR], combination of iMAR and 140 keV monoenergetic [PCD-CTiMAR+140keV]). All reconstructions were assessed qualitatively by four radiologists (independent and blinded reading on a 5-point Likert scale [5 = excellent; no artifact]) regarding overall image quality, artifact severity, and delineation of adjacent and distant anatomy. To assess signal homogeneity and evaluate the magnitude of artifact reduction, we performed quantitative measures of coefficient of variation (CV) and a region of interest (ROI)-based relative change in artifact reduction [PCD-CT/PCD-CTstd]. RESULTS: We enrolled 48 patients (mean age 66.5 ± 11.2 years, 50% (n = 24) males; mean BMI 25.2 ± 4.7 kg/m2; mean CTDIvol 6.2 ± 6 mGy). We found improved overall image quality, reduced artifacts and superior delineation of both adjacent and distant anatomy for the iMAR vs. non-iMAR reconstructions (all p < 0.001). No significant effect of the different artifact reduction approaches on CV was observed (p = 0.42). The ROI-based analysis indicated the most effective artifact reduction for the iMAR reconstructions, which was significantly higher compared to PCD-CT140keV (p < 0.001). CONCLUSION: PCD-CT offers highly effective approaches for metal artifact reduction with the potential to overcome current diagnostic challenges in patients with dental implants. CLINICAL RELEVANCE STATEMENT: Metallic artifacts pose a significant challenge in CT imaging, potentially leading to missed findings. Our study shows that PCD-CT with iMAR post-processing reduces artifacts, improves image quality, and can possibly reveal pathologies previously obscured by artifacts, without additional dose application. KEY POINTS: • Photon-counting detector CT (PCD-CT) offers highly effective approaches for metal artifact reduction in patients with dental fillings/implants. • Iterative metal artifact reduction (iMAR) is superior to high keV monoenergetic reconstructions at 140 keV for artifact reduction and provides higher image quality. • Signal homogeneity of the reconstructed images is not affected by the different artifact reduction techniques.

Lung shrinking assessment on HRCT with elastic registration technique for monitoring idiopathic pulmonary fibrosis.

OBJECTIVES: Evaluation and follow-up of idiopathic pulmonary fibrosis (IPF) mainly rely on high-resolution computed tomography (HRCT) and pulmonary function tests (PFTs). The elastic registration technique can quantitatively assess lung shrinkage. We aimed to investigate the correlation between lung shrinkage and morphological and functional deterioration in IPF. METHODS: Patients with IPF who underwent at least two HRCT scans and PFTs were retrospectively included. Elastic registration was performed on the baseline and follow-up HRCTs to obtain deformation maps of the whole lung. Jacobian determinants were calculated from the deformation fields and after logarithm transformation, log_jac values were represented on color maps to describe morphological deterioration, and to assess the correlation between log_jac values and PFTs. RESULTS: A total of 69 patients with IPF (male 66) were included. Jacobian maps demonstrated constriction of the lung parenchyma marked at the lung base in patients who were deteriorated on visual and PFT assessment. The log_jac values were significantly reduced in the deteriorated patients compared to the stable patients. Mean log_jac values showed positive correlation with baseline percentage of predicted vital capacity (VC%) (r = 0.394, p < 0.05) and percentage of predicted forced vital capacity (FVC%) (r = 0.395, p < 0.05). Additionally, the mean log_jac values were positively correlated with pulmonary vascular volume (r = 0.438, p < 0.01) and the number of pulmonary vascular branches (r = 0.326, p < 0.01). CONCLUSIONS: Elastic registration between baseline and follow-up HRCT was helpful to quantitatively assess the morphological deterioration of lung shrinkage in IPF, and the quantitative indicator log_jac values were significantly correlated with PFTs. KEY POINTS: • The elastic registration on HRCT was helpful to quantitatively assess the deterioration of IPF. • Jacobian logarithm was significantly reduced in deteriorated patients and mean log_jac values were correlated with PFTs. • The mean log_jac values were related to the changes of pulmonary vascular volume and the number of vascular branches.

Robustness of dual-energy CT-derived radiomic features across three different scanner types.

OBJECTIVES: To investigate the robustness of radiomic features between three dual-energy CT (DECT) systems. METHODS: An anthropomorphic body phantom was scanned on three different DECT scanners, a dual-source (dsDECT), a rapid kV-switching (rsDECT), and a dual-layer detector DECT (dlDECT). Twenty-four patients who underwent abdominal DECT examinations on each of the scanner types during clinical follow-up were retrospectively included (n = 72 examinations). Radiomic features were extracted after standardized image processing, following ROI placement in phantom tissues and healthy appearing hepatic, splenic and muscular tissue of patients using virtual monoenergetic images at 65 keV (VMI65keV) and virtual unenhanced images (VUE). In total, 774 radiomic features were extracted including 86 original features and 8 wavelet transformations hereof. Concordance correlation coefficients (CCC) and analysis of variances (ANOVA) were calculated to determine inter-scanner robustness of radiomic features with a CCC of ≥ 0.9 deeming a feature robust. RESULTS: None of the phantom-derived features attained the threshold for high feature robustness for any inter-scanner comparison. The proportion of robust features obtained from patients scanned on all three scanners was low both in VMI65keV (dsDECT vs. rsDECT:16.1% (125/774), dlDECT vs. rsDECT:2.5% (19/774), dsDECT vs. dlDECT:2.6% (20/774)) and VUE (dsDECT vs. rsDECT:11.1% (86/774), dlDECT vs. rsDECT:2.8% (22/774), dsDECT vs. dlDECT:2.7% (21/774)). The proportion of features without significant differences as per ANOVA was higher both in patients (51.4-71.1%) and in the phantom (60.6-73.4%). CONCLUSIONS: The robustness of radiomic features across different DECT scanners in patients was low and the few robust patient-derived features were not reflected in the phantom experiment. Future efforts should aim to improve the cross-platform generalizability of DECT-derived radiomics. KEY POINTS: • Inter-scanner robustness of dual-energy CT-derived radiomic features was on a low level in patients who underwent clinical examinations on three DECT platforms. • The few robust patient-derived features were not confirmed in our phantom experiment. • Limited inter-scanner robustness of dual-energy CT derived radiomic features may impact the generalizability of models built with features from one particular dual-energy CT scanner type.

Quantitative and qualitative assessments of deep learning image reconstruction in low-keV virtual monoenergetic dual-energy CT.

OBJECTIVES: To evaluate a novel deep learning image reconstruction (DLIR) technique for dual-energy CT (DECT) derived virtual monoenergetic (VM) images compared to adaptive statistical iterative reconstruction (ASIR-V) in low kiloelectron volt (keV) images. METHODS: We analyzed 30 venous phase acute abdominal DECT (80/140 kVp) scans. Data were reconstructed to ASIR-V and DLIR-High at four different keV levels (40, 50, 74, and 100) with 1- and 3-mm slice thickness. Quantitative Hounsfield unit (HU) and noise assessment were measured within the liver, aorta, fat, and muscle. Subjective assessment of image noise, sharpness, texture, and overall quality was performed by two board-certified radiologists. RESULTS: DLIR reduced image noise by 19.9-35.5% (p < 0.001) compared to ASIR-V in all reconstructions at identical keV levels. Contrast-to-noise ratio (CNR) increased by 49.2-53.2% (p < 0.001) in DLIR 40-keV images compared to ASIR-V 50 keV, while no significant difference in noise was identified except for 1 and 3 mm in aorta and for 1-mm liver measurements, where ASIR-V 50 keV showed 5.5-6.8% (p < 0.002) lower noise levels. Qualitative assessment demonstrated significant improvement particularly in 1-mm reconstructions (p < 0.001). Lastly, DLIR 40 keV demonstrated comparable or improved image quality ratings when compared to ASIR-V 50 keV (p < 0.001 to 0.22). CONCLUSION: DLIR significantly reduced image noise compared to ASIR-V. Qualitative assessment showed that DLIR significantly improved image quality particularly in thin sliced images. DLIR may facilitate 40 keV as a new standard for routine low-keV VM reconstruction in contrast-enhanced abdominal DECT. KEY POINTS: • DLIR enables 40 keV as the routine low-keV VM reconstruction. • DLIR significantly reduced image noise compared to ASIR-V, across a wide range of keV levels in VM DECT images. • In low-keV VM reconstructions, improvements in image quality using DLIR were most evident and consistent in 1-mm sliced images.

The effect of deep learning reconstruction on abdominal CT densitometry and image quality: a systematic review and meta-analysis.

OBJECTIVE: To determine the difference in CT values and image quality of abdominal CT images reconstructed by filtered back-projection (FBP), hybrid iterative reconstruction (IR), and deep learning reconstruction (DLR). METHODS: PubMed and Embase were systematically searched for articles regarding CT densitometry in the abdomen and the image reconstruction techniques FBP, hybrid IR, and DLR. Mean differences in CT values between reconstruction techniques were analyzed. A comparison between signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of FBP, hybrid IR, and DLR was made. A comparison of diagnostic confidence between hybrid IR and DLR was made. RESULTS: Sixteen articles were included, six being suitable for meta-analysis. In the liver, the mean difference between hybrid IR and DLR was - 0.633 HU (p = 0.483, SD ± 0.902 HU). In the spleen, the mean difference between hybrid IR and DLR was - 0.099 HU (p = 0.925, SD ± 1.061 HU). In the pancreas, the mean difference between hybrid IR and DLR was - 1.372 HU (p = 0.353, SD ± 1.476 HU). In 14 articles, CNR was described. In all cases, DLR showed a significantly higher CNR. In 9 articles, SNR was described. In all cases but one, DLR showed a significantly higher SNR. In all cases, DLR showed a significantly higher diagnostic confidence. CONCLUSIONS: There were no significant differences in CT values reconstructed by FBP, hybrid IR, and DLR in abdominal organs. This shows that these reconstruction techniques are consistent in reconstructing CT values. DLR images showed a significantly higher SNR and CNR, compared to FBP and hybrid IR. KEY POINTS: CT values of abdominal CT images are similar between deep learning reconstruction (DLR), filtered back-projection (FBP), and hybrid iterative reconstruction (IR). DLR results in improved image quality in terms of SNR and CNR compared to FBP and hybrid IR images. DLR can thus be safely implemented in the clinical setting resulting in improved image quality without affecting CT values.

Intra-scan inter-tissue variability can help harmonize radiomics features in CT.

OBJECTIVE: We studied the repeatability and the relative intra-scan variability across acquisition protocols in CT using phantom and unenhanced abdominal series. METHODS: We used 17 CT scans from the Credence Cartridge Radiomics Phantom database and 20 unenhanced multi-site non-pathologic abdominal patient series for which we measured spleen and liver tissues. We performed multiple measurements in extracting 9 radiomics features. We defined a "tandem" as the measurement of a given tissue (or material) by a given radiomics. For each tandem, we assessed the proportion of the variability attributable to repetitions, acquisition protocols, material, or patient. We analyzed the distribution of the intra-scan correlation between pairs of tandems and checked the impact of correlation coefficient greater than 0.90 in comparing paired and unpaired differences. RESULTS: The repeatability of radiomics features depends on the measured material; 56% of tandems were highly repeatable. Histogram-derived radiomics were generally less repeatable. Nearly 60% of relative radiomics measurements had a correlation coefficient higher than 0.90 allowing paired measurements to improve reliability in detecting the difference between two materials. The analysis of liver and spleen tissues showed that measurement variability was negligible with respect to other variabilities. As for phantom data, we found that gray level zone length matrix (GLZLM)-derived radiomics and gray level co-occurrence matrix (GLCM)-derived radiomics were the most correlating features. For these features, relative intra-scan measurements improved the detection of different materials or tissues. CONCLUSIONS: We identified radiomics features for which the intra-scan measurements between tissues are linearly correlated. This property represents an opportunity to improve tissue characterization and inter-site harmonization. KEY POINTS: • The repeatability of radiomics features on CT depends on the measured material or tissue. • Some tandems of radiomics features/tissues are linearly affected by the variability of acquisition protocols on CT. • Relative intra-scan measurements are an opportunity for improving quantitative imaging on CT.

Comparison of detection of trauma-related injuries using combined "all-in-one" fused images and conventionally reconstructed images in acute trauma CT.

OBJECTIVES: To compare the accuracy of lesion detection of trauma-related injuries using combined "all-in-one" fused (AIO) and conventionally reconstructed images (CR) in acute trauma CT. METHODS: In this retrospective study, trauma CT of 66 patients (median age 47 years, range 18-96 years; 20 female (30.3%)) were read using AIO and CR. Images were independently reviewed by 4 blinded radiologists (two residents and two consultants) for trauma-related injuries in 22 regions. Sub-analyses were performed to analyze the influence of experience (residents vs. consultants) and body region (chest, abdomen, skeletal structures) on lesion detection. Paired t-test was used to compare the accuracy of lesion detection. The effect size was calculated (Cohen's d). Linear mixed-effects model with patients as the fixed effect and random forest models were used to investigate the effect of experience, reconstruction/image processing, and body region on lesion detection. RESULTS: Reading time of residents was significantly faster using AIO (AIO: 266 ± 72 s, CR: 318 ± 113 s; p < 0.001; d = 0.46) while no significant difference was observed in the accuracy of lesion detection (AIO: 93.5 ± 6.0%, CR: 94.6 ± 6.0% p = 0.092; d = - 0.21). Reading time of consultants showed no significant difference (AIO: 283 ± 82 s, CR: 274 ± 95 s; p = 0.067; d = 0.16). Accuracy was significantly higher using CR; however, the difference and effect size were very small (AIO 95.1 ± 4.9%, CR: 97.3 ± 3.7%, p = 0.002; d = - 0.39). The linear mixed-effects model showed only minor effect of image processing/reconstruction for lesion detection. CONCLUSIONS: Residents at the emergency department might benefit from faster reading time without sacrificing lesion detection rate using AIO for trauma CT. KEY POINTS: • Image fusion techniques decrease the reading time of acute trauma CT without sacrificing diagnostic accuracy.