Recent and Upcoming Technological Developments in Computed Tomography: High Speed, Low Dose, Deep Learning, Multienergy The advent of computed tomography (CT) has revolutionized radiology, and this revolution is still going on. Starting as a pure head scanner, modern CT systems are now able to perform whole-body examinations within a couple of seconds in isotropic resolution, single-rotation whole-organ perfusion, and temporal resolution to fulfill the needs of cardiac CT. Because of the increasing number of CT examinations in all age groups and overall medical-driven radiation exposure, dose reduction remains a hot topic. Although fast gantry rotation, broad detector arrays, and different dual-energy solutions were main topics in the past years, new techniques such as photon counting detectors, powerful x-ray tubes for low-kV scanning, automated image preprocessing, and machine learning algorithms have moved into focus today. The aim of this article is to give an overview of the technical specifications of up-to-date available CT systems and recent hardware and software innovations for CT systems in the near future. Received for publication April 15, 2019; and accepted for publication, after revision, June 25, 2019. Conflicts of interest and sources of funding: none declared. Correspondence to: Michael M. Lell, MD, Department of Radiology and Nuclear Medicine, Klinikum Nürnberg, Paracelsus Medical University, Prof.-Ernst-Nathan-Straße 1, 90419 Nuernberg, Germany. E-mail: michael.lell@klinikum-nuernberg.de. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Risk of Hypersensitivity Reactions to Iopromide After Intra-Arterial Versus Intravenous Administration: A Nested Case-Control Analysis of 133,331 Patients Objective The aim of this study was to compare the risk of hypersensitivity reactions to iopromide after intra-arterial (IA) administration and intravenous (IV) administration. Materials and Methods Four observational studies were pooled. Almost half of the study population (48.1%) was from Europe, and one quarter each from China (27.6%) and other Asia countries (24.1%). All patients received iopromide either intra-arterially or intravenously for angiographic procedures (mostly cardio-angiography) or contrast-enhanced computed tomography. A nested case-control analysis, including a multivariable logistic regression model, was performed. Cases were defined by patients with a typical and unequivocal hypersensitivity (assumed non–IgE-mediated) reaction; controls were patients without any recorded reaction. The primary target variable is the odds ratio of having a hypersensitivity reaction after IA versus IV administration. Results A total of 133,331 patients met the inclusion criteria, 105,460 and 27,871 patients received iopromide IV or IA, respectively. Hypersensitivity reactions were recorded for 822 patients, and 132,509 patients served as controls. Major risk factors for hypersensitivity reactions were method of injection (IV vs IA), age (18 to <50 years vs ≥65 years), history of allergy or previous contrast media reaction (all P < 0.001), and asthma (P = 0.005). A total of 766 patients (0.7%) and 56 patients (0.2%) were recorded with hypersensitivity reactions after IV or IA administration, respectively (P < 0.0001). Adjusted odds ratio (IA vs IV) was 0.23 (95% confidence interval, 0.16–0.32) for all countries together: for China only, 0.22 (0.11–0.44); for all countries without China, 0.36 (0.25–0.53). Most frequent reactions were erythema/urticaria/rash, pruritus, and cough/sneezing. Conclusions Hypersensitivity reactions to iopromide were significantly less frequently recorded after IA administrations. This could be related to the delayed and diluted arrival of iopromide to the lungs. Received for publication May 28, 2019; and accepted for publication, after revision, July 14, 2019. Conflicts of interest and sources of funding: J.E., A.M., P.L., and K.V. are employees of Bayer; R.K. is a statistician for PAREXEL and paid for his service. Correspondence to: Jan Endrikat, MD, PhD, Bayer AG, Müllerstraße 178, 13353 Berlin, Germany. E-mail: jan.endrikat@bayer.com. This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially without permission from the journal. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Spectral Photon-Counting Computed Tomography for Coronary Stent Imaging: Evaluation of the Potential Clinical Impact for the Delineation of In-Stent Restenosis Objectives In-stent restenosis (ISR) is one of the main long-term complications after coronary stent placement, and the ability to evaluate ISR noninvasively using coronary computed tomography (CT) angiography remains challenging. For this application, spectral photon-counting CT (SPCCT) has the potential to increase image quality and reduce artifacts due to its advanced detector technology. Our study aimed to verify the technical and clinical potential of a novel SPCCT prototype using an ISR phantom setup. Materials and Methods Soft plaque-like restenosis (45 HU; approximately 50% of the stent lumen) were inserted into 10 different coronary stents (3 mm diameter), which were placed in a vessel phantom and filled with a contrast agent (400 HU). A research prototype SPCCT and a clinical dual-layer CT (DLCT; IQon; Philips) with comparable acquisition and reconstruction parameters were used to scan the phantoms. Conventional polyenergetic (PolyE) and monoenergetic (MonoE) images with 4 different energy levels (40, 60, 90, 120 keV) were reconstructed. Qualitative (delineation of the stenosis and adjacent residual lumen using a 5-point Likert scale) and quantitative (image noise, visible lumen diameter, lumen diameter adjacent to the stenosis, contrast-to-noise ratio of the restenosis) parameters were evaluated for both systems. Results The qualitative results averaged over all reconstructions were significantly superior for SPCCT compared with DLCT (eg, subjective rating of the best reconstruction of each scanner: DLCT PolyE: 2.80 ± 0.42 vs SPCCT MonoE 40 keV: 4.25 ± 1.03). Stenosis could be clearly detected in 9 and suspected in 10 of the 10 stents with both SPCCT and DLCT. The residual lumen next to the stenosis was clearly delineable in 7 of 10 stents (0.64 ± 0.11 mm or 34.97% of the measured stent lumen) with SPCCT, while it was not possible to delineate the residual lumen for all stents using DLCT. The measured diameter of the lumen within the stent was significantly higher for SPCCT compared with DLCT in all reconstructions with the best results for the MonoE 40 keV images (SPCCT: 1.80 ± 0.17 mm; DLCT: 1.50 ± 0.31 mm). The image noise and the contrast-to-noise ratio were better for DLCT than for SPCCT (contrast-to-noise ratio: DLCT MonoE 40: 31.58 ± 12.54; SPCCT MonoE 40: 4.64 ± 1.30). Conclusions Spectral photon-counting CT allowed for the noninvasive evaluation of ISR with reliable results regarding the residual lumen for most tested stents and the clear identification or suspicion of stenosis for all stents. In contrast, the residual lumen could not be detected for a single stent using DLCT. Received for publication April 30, 2019; and accepted for publication, after revision, July 14, 2019. Grischa Bratke and Tilman Hickethier contributed equally to this work. Conflicts of interest and sources of funding: G. Pahn, P. Coulon, and Y. Yagil are currently working for Philips. For the remaining authors, none were declared. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Correspondence to: Grischa Bratke, MD, Department of Radiology, University of Cologne, Kerpenerstr. 62, 50937 Cologne, Germany. E-mail: grischa.bratke@uk-koeln.de. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Dual-Energy Low-keV or Single-Energy Low-kV CT for Endoleak Detection?: A 6-Reader Study in an Aortic Aneurysm Phantom Objectives The aim of this study was to compare image quality, conspicuity, and endoleak detection between single-energy low-kV images (SEIs) and dual-energy low-keV virtual monoenergetic images (VMIs+) in computed tomography angiography of the aorta after endovascular repair. Materials and Methods An abdominal aortic aneurysm phantom simulating 36 endoleaks (2 densities; diameters: 2, 4, and 6 mm) in a medium- and large-sized patient was used. Each size was scanned using single-energy at 80 kVp (A) and 100 kVp (B), and dual-energy at 80/Sn150kVp for the medium (C) and 90/Sn150kVp for the large size (D). VMIs+ at 40 keV and 50 keV were reconstructed from protocols C and D. Radiation dose was 3 mGy for the medium and 6 mGy for the large size. Objective image quality and normalized noise power spectrum were determined. Subjective image quality, conspicuity, and sensitivity for endoleaks were independently assessed by 6 radiologists. Sensitivity was compared using Marascuilo procedure and Fisher exact test. Conspicuities were compared using Wilcoxon-matched pairs test, analysis of variance, and Tukey test. Results The contrast-to-noise-ratio of the aorta was significantly higher for VMI+ compared with SEI (P < 0.001). Noise power spectrum showed a higher noise magnitude and coarser texture in VMI+. Subjective image quality and overall conspicuity was lower for VMI+ compared with SEI (P < 0.05). Sensitivity for endoleaks was overall higher in the medium phantom for SEI (60.9% for A, 62.2% for B) compared with VMI+ (54.2% for C, 49.3% for D) with significant differences between protocols B and D (P < 0.05). In the large phantom, there was no significant difference in sensitivity among protocols (P = 0.79), with highest rates for protocols B (31.4%) and C (31.7%). Conclusions Our study indicates that low-keV VMI+ results in improved contrast-to-noise-ratio of the aorta, whereas noise properties, subjective image quality, conspicuity, and sensitivity for endoleaks were overall superior for SEI. Received for publication May 29, 2019; and accepted for publication, after revision, July 6, 2019. Conflicts of interest and sources of funding: none declared. Correspondence to: Hatem Alkadhi, MD, MPH, EBCR, FESER, Institute for Diagnostic and Interventional Radiology, University Hospital Zurich, Rämistrasse 100, 8091 Zurich, Switzerland. E-mail: hatem.alkadhi@usz.ch. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Image Artifact Management for Clinical Magnetic Resonance Imaging on a 7 T Scanner Using Single-Channel Radiofrequency Transmit Mode Objectives The aim of this work was to devise mitigation strategies for addressing a range of image artifacts on a clinical 7 T magnetic resonance imaging scanner using the regulatory-approved single-channel radiofrequency transmit mode and vendor-supplied radiofrequency coils to facilitate clinical scanning within reasonable scan times. Materials and Methods Optimized imaging sequence protocols were developed for routine musculoskeletal knee and neurological imaging. Sources of severe image nonuniformities were identified, and mitigation strategies were devised. A range of custom-made high permittivity dielectric pads were used to compensate for B1+ and B1− inhomogeneities, and also for magnetic susceptibility-induced signal dropouts particularly in the basal regions of the temporal lobes and in the cerebellum. Results Significant improvements in image uniformity were obtained using dielectric pads in the knee and brain. A combination of small voxels, reduced field of view B0 shimming, and high in-plane parallel imaging factors helped to minimize signal loss in areas of high susceptibility-induced field distortions. The high inherent signal-to-noise ratio at 7 T allowed for high receiver bandwidths and thin slices to minimize chemical shift artifacts. Intermittent artifacts due to radiofrequency inversion pulse limitations (power, bandwidth) were minimized with dielectric pads. A patient with 2 implanted metallic cranial fixation devices located within the radiofrequency transmit field was successfully imaged, with minimal image geometric distortions. Conclusions Challenges relating to severe image artifacts at 7 T using single-channel radiofrequency transmit functionality in the knee and brain were overcome using the approaches described in this article. The resultant high diagnostic image quality paves the way for incorporation of this technology into the routine clinical workflow. Further developmental efforts are required to expand the range of applications to other anatomical areas, and to expand the evidence- and knowledge-base relating to the safety of scanning patients with implanted metallic devices. Received for publication April 24, 2019; and accepted for publication, after revision, June 7, 2019. Conflicts of interest and sources of funding: none declared. Correspondence to: Andrew J. Fagan, PhD, Department of Radiology, Mayo Clinic, 200 First Street SW, Rochester, MN 55905. E-mail: Fagan.Andrew@mayo.edu. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Free-Breathing Dynamic Contrast-Enhanced Imaging of the Upper Abdomen Using a Cartesian Compressed-Sensing Sequence With Hard-Gated and Motion-State-Resolved Reconstruction Objectives The aim of this study was to compare a compressed-sensing free-breathing VIBE (fbVIBE) with a conventional breath-hold VIBE (bhVIBE) for dynamic contrast-enhanced imaging of the upper abdomen. Materials and Methods In total, 70 datasets (bhVIBE, n = 30; fbVIBE n = 40; hard-gated [hg] reconstruction, n = 30; motion-state-resolved [mr] reconstruction, n = 10) were assessed by 2 experienced readers. Both sequences were performed on 1.5-T magnetic resonance imaging scanners. The prototypical fbVIBE sequence acquired a navigation signal along with the imaging data and supported 2 different reconstructions: an hg reconstruction that either accepted or rejected an echo train based on the navigation signal and an mr reconstruction that assigned echo trains to their determined motion states. The hg reconstruction to reduce respiratory motion artifacts was carried out inline on the scanner (duration: approximately 8 minutes on the scanner-integrated CPU). The mr reconstruction delivered better results, but the reconstruction time is multiplied by the number of selected motion states (6 in the current study). Comparable reconstruction times to hg reconstruction can only be achieved on GPU-supported scanners. Therefore, the acquired raw data were selectively reconstructed at a later timepoint (duration: approximately 45 minutes). Welch analysis of variance tests were applied to compare image quality (IQ), delineation of structures, artifacts, and diagnostic confidence, which were rated on Likert-type scales (IQ/delineation of structures/diagnostic confidence: 1 [nondiagnostic] to 5 [perfect]; artifacts: 1 [no artifacts] to 5 [severe artifacts]). Mann-Whitney U tests and Kruskal-Wallis H tests were used to compare the extent of artifacts in older (aged ≥70 years) and younger (aged <70 years) patients. Interobserver agreement was assessed using Cohen κ. Results Mean ratings for IQ/delineation of structures/diagnostic confidence of fb(hg)VIBE (4.2 ± 0.7/4.3 ± 0.8/4.3 ± 0.7; κ = 0.8/0.7/0.6) and fb(mr)VIBE (4.9 ± 0.3/4.9 ± 0.3/4.9 ± 0.3; κ = 0.3/1/0.9) were higher compared with those of bhVIBE (3.7 ± 0.8/3.8 ± 0.8/3.9 ± 0.9; κ = 0.9/0.9/0.9), whereas artifacts of fb(hg)VIBE/fb(mr)VIBE were rated lower (fb[hg]VIBE/fb[mr]VIBE/bhVIBE = 2.2 ± 0.9/1.3 ± 0.5/2.4 ± 0.9; κ = 0.6/0.6/0.9). The IQ of fb(hg)VIBE was rated significantly higher compared with that of bhVIBE (P = 0.03). All parameters were significantly improved by mr reconstruction compared with fb(hg)VIBE and bhVIBE (P < 0.001). In the fb(hg)VIBE cohort, an insignificant trend toward lower artifacts in the younger age group (≥70 years: 2.5 ± 0.9 vs <70 years: 1.9 ± 0.8) was found, whereas significant differences emerged in the bhVIBE cohort (≥70 years: 3 ± 0.9 vs <70 years: 2.1 ± 0.9; P = 0.02). Conclusions Fast fbVIBE using hg and mr reconstructions is technically feasible with improved IQ compared with that of bhVIBE. Free-breathing VIBE may be useful for dynamic contrast-enhanced of the upper abdomen, particularly in older and/or severely ill patients with impaired breath-hold capabilities. Received for publication June 1, 2019; and accepted for publication, after revision, June 28, 2019. Conflicts of interest and sources of funding: M.K., D.M.N. and B.K. are Siemens Healthcare employees and the Institute of Clinical Radiology at the Kantonsspital Baden has research agreements with Siemens. The sequence was provided by Siemens for research purposes without charge. There was no financial funding of this work and no financial relationship between the other authors and Siemens. Correspondence to: Daniel Hausmann, MD, Institute of Radiology, Kantonsspital Baden, Im Ergel 1, 5404 Baden, Switzerland. E-mail: Daniel.hausmann@ksb.ch. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Gadolinium Retention in Erythrocytes and Leukocytes From Human and Murine Blood Upon Treatment With Gadolinium-Based Contrast Agents for Magnetic Resonance Imaging Objectives Being administered intravenously, the tissue that gadolinium-based contrast agents (GBCAs) for magnetic resonance imaging mostly encounter is blood. Herein, it has been investigated how much Gd is internalized by cellular blood components upon the in vitro incubation of GBCAs in human blood or upon intravenous administration of GBCAs to healthy mice. We report results that show how the superb sensitivity of inductively coupled plasma–mass spectrometry (ICP-MS) allows the detection of very tiny amounts of GBCAs entering red blood cells (RBCs) and white blood cells (WBCs). This finding may introduce new insights in the complex matter relative to excretion and retention pathway of administered GBCAs. Materials and Methods The study was tackled by 2 independent approaches. First, human blood was incubated in vitro with 5 mM of GBCAs (gadoteridol, gadobenate dimeglumine, gadodiamide, and gadopentetate dimeglumine) for variable times (30 minutes, 1 hour, 2 hours, and 3 hours) at 37°C. Then, blood cell components were isolated by using the Ficoll Histopaque method, washed 3 times, mineralized, and analyzed by ICP-MS for total Gd quantification. Furthermore, blood components derived from human blood incubated with gadodiamide or gadoteridol underwent UPLC-MS (ultra performance liquid chromatography–mass spectrometry) analysis for determination of the amount of intact Gd-DTPA-BMA and Gd-HPDO3A. Second, the distribution of Gd in the blood components of healthy CD-1 mice was administered intravenously with a single dose (1.2 mmol/kg) of gadodiamide or gadoteridol. Blood samples were separated and processed at different time points (24 hours, 48 hours, 96 hours, and 10 days after GBCA administration). As for human blood, ICP-MS quantification of total Gd and UPLC-MS determination of the amount of intact GBCAs were carried out. Results The amount of Gd taken up by RBCs and WBCs was well detectable by ICP-MS. The GBCAs seem to be able to cross the membrane by diffusion (RBCs) or, possibly, by macropinocytosis (WBCs). Ex vivo studies allowed it to be established that the structure of the different GBCAs were not relevant to determine the amount of Gd internalized in the cells. Although the amount of Gd steadily decreases over time in gadoteridol-labeled cells, in the case of gadodiamide, the amount of Gd in the cells does not decrease (even 10 days after the administration of the GBCA). Moreover, while gadoteridol maintains its structural integrity upon cellular uptake, in the case of gadodiamide, the amount of intact complex markedly decreases over time. Conclusions The detection of significant amounts of Gd in RBCs and WBCs indicates that GBCAs can cross blood cell membranes. This finding may play a role in our understanding of the processes that are at the basis of Gd retention in the tissues of patients who have received the administration of GBCAs. Received for publication June 4, 2019; and accepted for publication, after revision, July 9, 2019. Silvio Aime is the senior author of this study. Conflicts of interest and sources of funding: The study was supported by the “Progetto di Ateneo Compagnia di San Paolo” (CSTO160182), EuroBioimaging Italy CNR, and “Regional project Gadoplus (IR2) Industrializzazione dei Risultati della Ricerca (F.E.S.R. 2014/2020).” E.D.G. was supported by FIRC-AIRC (Fondazione Italiana per la Ricerca sul Cancro AIRC) fellowship. The other authors declare no conflict of interest. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Correspondence to: Eliana Gianolio, PhD, Department of Molecular Biotechnologies and Health Sciences, University of Turin, Via Nizza 52, Turin, Italy. E-mail: eliana.gianolio@unito.it. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Development of Upright Computed Tomography With Area Detector for Whole-Body Scans: Phantom Study, Efficacy on Workflow, Effect of Gravity on Human Body, and Potential Clinical Impact Objectives Multiple human systems are greatly affected by gravity, and many disease symptoms are altered by posture. However, the overall anatomical structure and pathophysiology of the human body while standing has not been thoroughly analyzed due to the limitations of various upright imaging modalities, such as low spatial resolution, low contrast resolution, limited scan range, or long examination time. Recently, we developed an upright computed tomography (CT), which enables whole-torso cross-sectional scanning with 3-dimensional acquisition within 15 seconds. The purpose of this study was to evaluate the performance, workflow efficacy, effects of gravity on a large circulation system and the pelvic floor, and potential clinical impact of upright CT. Materials and Methods We compared noise characteristics, spatial resolution, and CT numbers in a phantom between supine and upright CT. Thirty-two asymptomatic volunteers (48.4 ± 11.5 years) prospectively underwent both CT examinations with the same scanning protocols on the same day. We conducted a questionnaire survey among these volunteers who underwent the upright CT examination to determine their opinions regarding the stability of using the pole throughout the acquisition (closed question), as well as safety and comfortability throughout each examination (both used 5-point scales). The total access time (sum of entry time and exit time) and gravity effects on a large circulation system and the pelvic floor were evaluated using the Wilcoxon signed-rank test and the Mann-Whitney U test. For a large circulation system, the areas of the vena cava and aorta were evaluated at 3 points (superior vena cava or ascending aorta, at the level of the diaphragm, and inferior vena cava or abdominal aorta). For the pelvic floor, distances were evaluated from the bladder neck to the pubococcygeal line and the anorectal junction to the pubococcygeal line. We also examined the usefulness of the upright CT in patients with functional diseases of spondylolisthesis, pelvic floor prolapse, and inguinal hernia. Results Noise characteristics, spatial resolution, and CT numbers on upright CT were comparable to those of supine CT. In the volunteer study, all volunteers answered yes regarding the stability of using the pole, and most reported feeling safe (average rating of 4.2) and comfortable (average rating of 3.8) throughout the upright CT examination. The total access time for the upright CT was significantly reduced by 56% in comparison with that of supine CT (upright: 41 ± 9 seconds vs supine: 91 ± 15 seconds, P < 0.001). In the upright position, the area of superior vena cava was 80% smaller than that of the supine position (upright: 39.9 ± 17.4 mm2 vs supine: 195.4 ± 52.2 mm2, P < 0.001), the area at the level of the diaphragm was similar (upright: 428.3 ± 87.9 mm2 vs supine: 426.1 ± 82.0 mm2, P = 0.866), and the area of inferior vena cava was 37% larger (upright: 346.6 ± 96.9 mm2 vs supine: 252.5 ± 93.1 mm2, P < 0.001), whereas the areas of aortas did not significantly differ among the 3 levels. The bladder neck and anorectal junction significantly descended (9.4 ± 6.0 mm and 8.0 ± 5.6 mm, respectively, both P < 0.001) in the standing position, relative to their levels in the supine position. This tendency of the bladder neck to descend was more prominent in women than in men (12.2 ± 5.2 mm in women vs 6.7 ± 5.6 mm in men, P = 0.006). In 3 patients, upright CT revealed lumbar foraminal stenosis, bladder prolapse, and inguinal hernia; moreover, it clarified the grade or clinical significance of the disease in a manner that was not apparent on conventional CT. Conclusions Upright CT was comparable to supine CT in physical characteristics, and it significantly reduced the access time for examination. Upright CT was useful in clarifying the effect of gravity on the human body: gravity differentially affected the volume and shape of the vena cava, depending on body position. The pelvic floor descended significantly in the standing position, compared with its location in the supine position, and the descent of the bladder neck was more prominent in women than in men. Upright CT could potentially aid in objective diagnosis and determination of the grade or clinical significance of common functional diseases. Received for publication April 26, 2019; and accepted for publication, after revision, July 1, 2019. Conflicts of interest and sources of funding: This study was supported by Canon Medical Systems. This study was supported by Japan Society for the Promotion of Science (JSPS KAKENHI: grant number JP17H04266), Uehara Memorial Foundation, and Canon Medical Systems (Otawara, Japan). Masahiro Jinzaki has received a grant from Canon Medical Systems. Canon Medical Systems has loaned the upright computed tomography machine to Keio University. However, Canon Medical Systems is not involved in the design and conduct of the study; in the collection, analysis, and interpretation of the data; and in the preparation, review, or approval of the manuscript. All other authors have no relationships with industry or other entities. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Correspondence to: Masahiro Jinzaki, MD, PhD, Department of Radiology, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo, 160-8582, Japan. E-mail: jinzaki@rad.med.keio.ac.jp. This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially without permission from the journal. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

A Practical Guide to Artificial Intelligence–Based Image Analysis in Radiology The use of artificial intelligence (AI) is a powerful tool for image analysis that is increasingly being evaluated by radiology professionals. However, due to the fact that these methods have been developed for the analysis of nonmedical image data and data structure in radiology departments is not “AI ready”, implementing AI in radiology is not straightforward. The purpose of this review is to guide the reader through the pipeline of an AI project for automated image analysis in radiology and thereby encourage its implementation in radiology departments. At the same time, this review aims to enable readers to critically appraise articles on AI-based software in radiology. Received for publication May 21, 2019; and accepted for publication, after revision, June 22, 2019. Correspondence to: Thomas Weikert, MD, Department of Radiology, University Hospital Basel, Petersgraben 4, 4031 Basel, Switzerland. E-mail: thomas.weikert@usb.ch. Conflicts of interest and sources of funding: none declared. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Improved Detection of Benign and Malignant Rib Lesions in the Routine Computed Tomography Workup of Oncological Patients Using Automated Unfolded Rib Image Postprocessing Objective The aim of this study was to evaluate the performance of the automated computed tomography (CT) postprocessing software unfolded rib images for improved detection of both benign and malignant rib lesions during routine diagnostic workup of oncological patients. Materials and Methods One thousand eight in-patients and out-patients (63.66 ± 14.25 years; range, 18.67–95.67 years; 405 females and 603 males), undergoing chest CT between July 2018 to January 2019 at our institution, were retrospectively evaluated. Patients underwent chest CT alone or as part of a whole-body CT staging/restaging. The CT protocol consisted of the following: 120 kV; 100 mAs; matrix, 512 × 512; collimation, 0.6 mm; reconstructed section thickness of 3 mm and 1 mm using a soft tissue spatial resolution kernel (I30f) and a sharp kernel (B70f). Both transversal image data sets were used for “conventional” diagnosis including coronal reformates with 3-mm slice thickness. One-millimeter slice thickness image data sets of all patients were additionally directed from the scanner to a computational server where they were automatically postprocessed to 3-dimensional unfolded ribs. The “unfolding” of the rib using the centerline as an axis allows a synchronous display and rotation of all ribs by mouse scrolling. These postprocessed image data sets were evaluated in a separate reading session (approximately 4 weeks later). The readers had no information about the underlying medical history or clinical presentation. They were asked to record the lesion number, site of involvement along the rib (proximal, body, distal), number of the involved ribs, and the character of the lesion in terms of lytic versus sclerotic versus mixed lytic/sclerotic. The standard of reference was 18F-FDG PET, 68Ga-DOMITATE PET/CT, bone scan, or imaging follow-up (>6 months). Results From a total of 1008 evaluated patients, 763 (73.02%) were hemato-oncologic patients. A total of 104 rib lesions were found by transversal CT image reading, whereas the unfolded rib image reading detected 305 lesions. Eighty-nine were classified malignant, and 202 were classified benign. Detection of malignant rib lesions proved significant both for less than 1 cm (P < 0.02) and more than 1 cm in diameter (P < 0.007). The sensitivity, specificity, positive predictive value, and negative predictive value for detection of malignant rib lesions were 97.7%, 98.5%, 96.6%, and 99% for unfolding ribs, and 76.4%, 100%, 92.7%, and 90.5% for conventional (transversal) image reading, respectively. Detection of sclerotic rib lesions and lesions greater than 1 cm in diameter were significantly better (P < 0.01) for the unfolding rib algorithm. Conclusions The “unfolded rib” reformates are significantly superior for rib lesion detection compared with conventional transversal CT scan reading and should therefore be used in all patients, particularly those with an oncologic background. Received for publication April 29, 2019; and accepted for publication, after revision, June 17, 2019. Sorin Dumitru Ioanoviciu and Marius Horger both share equal contribution. Conflicts of interest and sources of funding: J.F. received institutional research support from Siemens Healthcare USA, DePuy, Zimmer, Microsoft, and BTG International, and is a scientific advisor of Siemens Healthcare USA, Alexion Pharmaceuticals, and BTG International. He also received speaker's honorarium from Siemens Healthcare USA and has shared patents with Siemens Healthcare and Johns Hopkins University. M.H. received institutional research support from Siemens Healthineers Germany; he is a scientific advisor of Siemens Healthineers Germany and received speaker's honorarium from Siemens Healthineers Germany. K.E., C.K., and S.D.I. have no conflicts of interest to declare. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Correspondence to: Kaspar Ekert, MD, Department of Radiology, Diagnostic and Interventional Radiology, Eberhard Karls University, Hoppe-Seyler-Straße 3, D-72076 Tübingen, Germany. E-mail: kaspar.ekert@uni-tuebingen.de. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.