Computed tomography (CT) has a long and outstanding tradition, and due to significant technological improvements, CT is the most frequently used cross-sectional imaging modality in today's clinical practice. Since the last review in 2015,1 the development in CT technology has shifted from increasing the gantry rotation speed, adding more rows to the detector panel, and improving the performance of the x-ray tubes toward exciting and potentially disruptive technologies such as photon counting detectors, which provide higher spatial resolution, lower image noise, and increased multienergy capabilities and machine learning algorithms for image preprocessing. Commercially available high-end CT systems approach physical (and rational) limits of the hardware, especially detector, tube, and gantry technology: The maximum detector width in z-direction remained at 16 cm. Gantry rotation time almost remained the same; only a dedicated cardiac scanner with a short tube-detector geometry can rotate 0.01 second faster now. The 0.20 second rotation time, as heralded in 2014 for a whole-body CT system, is still a promise. Low-kV scanning with settings down to 70 kV has become widely available, and 90 or 100 kV can now be considered the new standard. On the other hand, major advances in dose-reduction techniques, such as sub-mSv cardiac CT angiography (CTA) or 0.2 mSv chest CT, made this technique attractive for large-scale screening trials. Each major vendor offers dual-energy or multienergy CT, with a multitude of clinical applications proposed recently. Machine learning algorithms will be used for various tasks, image reconstruction, motion correction, individual dose calculation, and finally screen data sets for pathology, and provide diagnostic suggestions to name a few.
Race for Rows Slows Down
Today's high-end CT systems have 192 to 320 detector rows in z-direction and a maximum detector panel width of 16 cm in the isocenter (about 30 cm physically; Table 1). The detector width of 16 cm is a compromise between the advantage of static or dynamic whole-organ imaging (brain, heart, etc) without table movement and the disadvantage of increased scatter, cone-beam artifacts, heel effect, and the trade-off between spatial resolution and image noise due to large cone angle.2 If the full detector width of 16 cm is used, predominantly sequential mode is applied. Toshiba launched the first system with such a detector array (Aquilion One), and GE followed that path and introduced a 16 cm detector system (Revolution CT) at RSNA 2013. A dedicated cardiac CT system with a maximum field of view of 25 or 16 cm in x-y direction and a detector width of 14 cm at isocenter was announced at RSNA 2018 (CardioGraph; GE Healthcare/Arineta). This system is equipped with 2 tubes—in contrast to the dual-source systems (Siemens)—both tubes are aligned on the same x-y position and expose a single-detector array (Fig. 1). This “stereo CT” technology aims at reducing cone-beam artifacts and seeks to make better use of the applied x-ray dose.3 All CT systems with detector width larger than 8 cm have something in common, that they use the entire detector panel in sequential mode only. If the scan range exceeds the detector width, spiral/helical scan mode is applied, which uses the central detector rows only to minimize overbeaming (eg, 64 × 0.5 mm in the case of the Aquilion One system). Because cardiac CTA is now possible within one heart beat with most high-end systems, one of the major driving forces for more slices has faded away. The centrifugal force physically limits the race for faster gantry rotations, although a rotation time as short as 0.20 second has been promised in 2014 by GE, this has not been realized yet. At the moment, the fastest rotation time is 0.24 second (CardioGraph), which has a source-axis distance of 45 cm and therefore can rotate faster for a given g-force. The downside of this system is a smaller gantry bore and a limited field of view that restricts its application to the heart and the central vessels.
The x-ray tube design and technological principles that were discussed in detail previously1 are still valid today: more x-ray tube power (Table 1) is necessary for both faster scanning and, more importantly, to reduce x-ray exposure when applying low-kV scan protocols. Dedicated prefilters have been introduced to shape the photon spectrum and selectively remove low-energy photons from the beam, which would otherwise be absorbed in the patient rather than reaching the detector. Scanning the chest with 100 kV and tin-filtration has been shown to render chest CT possible at an exposure level of conventional chest radiographs (effective dose, < 0.2 mSv),4–6 well suited for future screening programs. A 0.4 mm tin filter, as implemented in some of the Siemens CT, absorb approximately 90% of all photons emitted, thus tube output needs to be high to generate an appropriate signal at the detector.7 The other reason for the need of high tube output is to utilize low-kV protocols combined with high scan speed, not only in small patients but throughout the patient cohort, including larger patients with high body mass index. Scanning at low tube voltage increases the iodine contrast and allows for a reduction of contrast material (CM) volume in many applications. Therefore, tube voltage selection is a powerful tool to reduce both x-ray and CM exposure. A 100 kilovolt has become the new imaging standard on scanners providing adequate tube output that used to be 120 kV in the past.8–12 A few years ago, Siemens started to offer a selection of tube voltage in 10 kV steps from 70 to 150 kV, but in the meantime, other vendors adapted this feature at least for the low-kV values.
The x-ray tube output is also influenced by the system's cone angle, which relates to the number of detector rows. As illustrated in Figure 2, the anode angle of the x-ray tube has to increase with the cone angle of the CT system. Larger anode angles, however, imply less demagnification of the electron beam. To achieve the same spatial resolution, a narrower and thus less powerful electron beam has to be used for systems with wider z-coverage to avoid anode melting. Less x-ray power, however, limits the possibilities for dose reduction.
GE announced a new Quantix 160 x-ray tube at RSNA 2018, being available in the latest high-end system (Revolution Apex), providing 1300 mA at 70 and 80 kV even at the large cone angle and anode angle for a 16 cm detector panel. It can be anticipated that even higher tube current values can be expected in next-generation x-ray tubes with smaller cone angles.
Recent developments in CT detector technology involve detectors with smaller detector elements and prototype photon counting detectors. Computed tomography systems with (conventional) photon-integrating detector technology use 0.5 mm, 0.6 mm, or 0.625 mm thick detector rows and similar pixel sizes in the lateral direction (Table 1). The only exception being Canon's new Aquilion Precision system, in which the ultra-high-resolution (UHR) detector has a pixel size of 0.25 × 0.25 mm (scaled to isocenter). Smaller septa ensure still acceptable geometrical efficiency. Together with a smaller focal spot, this system is able to provide CT images at higher spatial resolution. The technology is different from the UHR mode introduced by Siemens, which introduces a UHR comb or grid in front of the detector array to increase spatial resolution at the cost of higher patient exposure, because 50% to 75% of photons that already passed the patient are absorbed by this grid.13,14 The Canon system does not have such a dose penalty.
Image noise is reduced if image reconstruction at a given resolution is performed with data from smaller detector elements.15,16 This effect (image noise reduction of 10%–20%) has recently been reported with photon counting detector systems17,18; a similar effect is expected with the Canon system that uses conventional detector material. Besides smaller detector elements, image reconstruction with higher matrix (up to 10242 instead of 5122) is suggested (Table 1).
Conventional CT detector technology, where x-rays are indirectly converted into an electric signal, might be replaced by direct converters in the near future. These direct converters are based on semiconductors that convert an x-ray photon directly into an electrical current. The signal generated by a single x-ray photon is short enough to decay before the next photon arrives, so quantification of individual photons becomes possible. To avoid further photons arriving while the first photon's signal is not yet decayed, a situation called “pileup,”19,20 the pixels of such photon counting detectors are typically smaller than those of conventional detectors. The area under each signal is proportional to the energy of the incoming x-ray photon. Before converting this area into a height, the electric pulses are smoothed before being analyzed and then compared with a threshold voltage. Typically 2 to 4 thresholds are built into a pixel, and thus up to 4 energy levels (or energy bins) can be separated. Photon counting detectors should have multiple potential advantages over conventional detectors21:
- Less noise due to absence of electronic noise
- Less noise due to statistical effects (Swank factor)
- Less noise due to the possibility of statistically optimal energy bin weighting
- Less noise if small pixel data are reconstructed at lower spatial resolution
- Smaller pixels and higher spatial resolution (necessary to avoid pileup)
- Spectral information with typically 2 or 4 energy bins
Less noise can always be balanced against more aggressive dose reduction. Because processing of the counts is purely digital, the aforementioned features can be used retrospectively on demand. This means that specific high-resolution or dual-energy protocols are not necessary any more, as this information can be retrospectively derived from the raw data. In Figure 3, scans of a low contrast phantom with a CT system with conventional detector array and a benchtop photon counting CT with matched imaging parameters are compared.
Although numerous benchtop experiments exist, only one whole-body CT system with a photon counting detector panel where scans of humans have been conducted17,18,22–27 is available, with 3 installations worldwide (Mayo Clinic, Rochester, MN; NIH Clinical Center, Bethesda, MD; DKFZ, Heidelberg, Germany). This system features a dual-source CT gantry with one conventional and one photon-counting detector panel, the later utilizing CdTe sensor material.
It operates in 4 different detector modes (Fig. 4) with the sharp and the UHR mode being the most versatile ones. This versatility, however, comes at the cost of a reduced z-coverage of the prototype system due to data transfer rate restrictions. Such constraints or compromises are justified for a prototype CT system. A final product implementation will certainly need to overcome such restrictions.
Given the high maturity of conventional CT detectors, it is impressive that a noninferiority study proved CounT to be as good as state-of-the art clinical CT systems.24
The special photon counting detector modes, with which one can either read out the macro pixels or the smaller subpixels, make it possible to nicely demonstrate the advantages of smaller pixels (higher spatial resolution or lower noise) over larger pixels, as shown in Figure 5. This confirms the observations reported in Pourmorteza et al17 and Leng et al.18 A direct comparison of image resolution between energy integrating and photon counting is shown in Figure 6.
Tube Current Modulation/Automatic Exposure Control
The technical basis for tube current adaption dates back to 1981.28 Angular tube current modulation (TCM) resulted in 15% to 50% dose reduction, depending on the anatomical region in the x-y plane.29 Online TCM did not only reduce patient exposure, but also homogenized noise distribution and therefore improved image quality.30 The logical advancement of in-plane TCM was longitudinal or z-axis TCM. In analogy to angular modulation that considers different attenuation in-plane (eg, transverse projection vs anteroposterior projection at the level of the shoulders), longitudinal TCM aims to homogenize noise, taking into account the different attenuation of the chest as compared with the abdomen or pelvis. Different solutions were implemented, using either a sinusoidal or attenuation-based online modulation algorithm. Automatic exposure control resembles a group of algorithms that incorporate (3-dimensional) TCM and aim to deliver a predefined image quality across a range of patient sizes derived from the topogram, increasing CTDIvol for large, and decreasing CTDIvol for small patients. Because automatic exposure control algorithms assume that the patient is in the isocenter, correct centering is of importance. To optimize patient positioning, a ceiling-mounted 3D-camera system has been introduced, which can identify the position of the patient and optimize it relative to the gantry coordinates.31 If the scan range goes beyond the range of the topogram, CT systems may act differently using either maximum or minimum mAs setting, or something in between (standard mAs setting or mAs setting at last calculated position). The target image quality is also predicted differently, whereas one approach uses a setting referenced to a standard patient, other approaches use noise indexes.1
Scanning at low-kV setting increases the attenuation of CM. With the implementation of more powerful x-ray tubes, low-kV scanning became practical and increasingly popular. The increased iodine attenuation can either be used to reduce the volume of CM or to reduce radiation exposure (compensating higher image noise by higher contrast), or a combination of both.9,11,32–35 Automatic selection of the tube voltage and adaption of the tube current, using information on the patient's attenuation and accounting for the planned examination type, boosted this technology into routine use. Depending on the examination type, patient size, and tube output at low kV, dose reduction between 10% and 30% has been reported.36–38 However, the availability of new high performance x-ray tubes that can deliver a very high tube current at low tube voltages will result in even higher dose reduction. For pediatric CT angiography, dose reduction up to 70% for head, 77% for thorax, and 34% for abdomen/pelvis has been reported.39 Additional prefilters are in use for some protocols to remove undesired low-energy radiation from the low-kV spectrum to maximize the image quality and to minimize patient dose down to a level of conventional radiography.4,40–42
Low-Dose Computed Tomography and Noise Suppression in Image Reconstruction
Filtered backprojection is more and more being replaced by iterative image reconstruction techniques. These are designed to significantly reduce image noise and, to some extent, reduce image artifacts. This is achieved by adding prior knowledge to the reconstruction algorithm and by improving the forward model used for iterative image reconstruction, which may, for example, also include realistic ray profiles or the correct photon statistics. Iterative reconstruction can be categorized into 2 categories: algorithms that are of image postprocessing type (AIDR 3D, ASIR, IRIS, and iDose) and those that are able to do one or more iterations through raw data domain (FIRST, ASIR-V, Veo, Safire, Admire, IMR). In so far, not much has changed compared with our last review.1Table 2 summarizes some of the properties of those algorithms.
Low-dose (LD) CT protocols and iterative reconstruction algorithms (especially algorithms with iterations in raw data domain) to reduce noise and restore image quality have demonstrated their potential in a large variety of clinical settings.
Chest CT, for example, is a very promising application for LD techniques for a variety of reasons: it is a rather frequent examination, radiation sensitive organs are within the scan range, and there are large anatomical attenuation differences. Ultra-low-dose (ULD) CT has been proposed at a dose level comparable to conventional chest x-ray. The term ULD is not clearly defined; in this manuscript, it is used for CT protocols with less than 0.2 mSv, which is roughly in the order of posterior-anterior and lateral chest radiographs. Hu-Wang et al45 could show that chest CT at a level of 0.14 mSv (range, 0.10–0.20 mSv) with MBIR provides similar cyst quantification as standard-dose CT in patients with lymphangioleiomyomatosis. In a study of patients with cystic fibrosis (CF), Ernst et al46 reported similar Bhalla scores for a routine (mean estimated effective dose for children <18 years, 0.52 mSv; adults, 1.12 mSv) and an ULD (children, 0.04 mSv; adults, 0.05 mSv) chest CT. However, there are certainly limitations, the image quality of the ULD cases shown in this article were clearly inferior compared with the routine CT. The authors stated that the ULD CT protocol with MBIR should only be used in the follow-up of patients without exacerbation. Villanueva-Meyer et al47 demonstrated that pediatric ULD chest CT may be adequate for the exclusion of airway foreign body, but suboptimal for the evaluation of parenchymal lung disease.
Nagatani et al48 evaluated the ability of ULD chest CT for nodule detection in a cohort of 83 patients that received 3 CT scans with different dose settings (120 kV, 0.35 second rotation time, 240 [regular dose; reference]/120 [LD]/20 [ULD] mA; ADIR 3D) consecutively at a single visit. The mean effective dose of their ULD protocol is a little higher (0.29 mSv) than our aforementioned definition. They could show that ULD and LD perform similarly when assessing solid nodules larger than 3 mm and ground glass nodules with 8 mm and more. In a similar setting, Katsura et al49 did not find significant differences between LD ASIR and ultra-LD MBIR for overall sensitivity for ground-glass opacity, partly solid, or solid nodule. Kim et al50 reported overall image quality rated diagnostic in 100% of the examinations with LD protocol (mean effective dose, 1.06 ± 0.11 mSv), 96% with ULD-1 protocol (mean effective dose, 0.44 ± 0.05 mSv), and 88% with ULD-2 protocol (mean effective dose, 0.31 ± 0.03 mSv). All patients with nondiagnostic quality images had a body mass index greater than 25. Further dose finding studies are needed to determine protocols with lowest radiation exposure without sacrificing lesion detection for a variety of different CT systems for lung cancer screening projects.
Recently, a new family of image reconstruction algorithms based on deep learning has been announced (Table 2). Because these algorithms may not only be regarded as dose-reduction approaches but also methods to improve image quality and to reduce artifacts, they will be discussed in the following section.
Deep Learning-Based Image Restoration/Reconstruction
With the general success of neural networks—especially convolutional neural networks (CNNs)—deep learning has quickly found its way into medical imaging.51,52 The intrinsic properties of convolutional layers leverage the local dependencies present in image data. The combination of several (linear) convolutional layers and nonlinear layers essentially makes them multiparametric functions that can universally approximate any function.53 A CNN can be a function that maps noisy CT data into noise-reduced data. It can also be a function that maps CT data with artifacts into those with reduced artifacts. The number of open parameters of such universally approximating functions goes into the millions or billions and needs to be trained using large amounts of data.
During training, these networks learn typical features of CT images and learn to avoid smoothing across such feature boundaries to avoid spatial resolution loss. The variety of features AI networks learn promise that they are potentially much more powerful than iterative restoration or reconstruction techniques, which used handmade priors (total variation minimization, edge preserving priors, etc). The simplest way to train a network for noise reduction is to provide LD images as input and high-dose images of the same patient as output. The LD images can be simulated by adding noise to the raw data (virtual data).
A different approach are unsupervised networks that do not require matched data pairs (input-output) for training. These so-called generative adversarial networks (GANs)54 and conditional GANs (cGANs)55 are among the most popular unsupervised approaches nowadays. To put it in a nutshell, the generator is trained to produce high-dose images from LD images, whereas the discriminator is trained to discriminate between virtual high-dose images and real high-dose images (eg, those from a different scan or a different patient). With GAN approaches, training does not necessarily require paired training data to achieve effective noise reduction.56
Only little detail is known about the commercial implementations. Canon's AiCE approach (https://global.medical.canon/products/computed-tomography/aice_dlr) seems to be trained on LD images reconstructed with FBP as input and high-dose images reconstructed with Canon's iterative reconstruction algorithm as output (Fig. 7). The GE algorithm True Fidelity can only be speculated about at the moment: Ziabari et al57 and other similar references of the same GE group that had been published at conferences late 2018 discuss a neural network that converts FBP images into images that have been reconstructed with GE's rather slow iterative Veo algorithm, also known as MBIR. To improve convergence, the deep learning algorithm outputs the residual Veo minus FBP, which is then added to the FBP image. Two-dimensional, 2.5-dimensional, and 3-dimensional variations of the algorithm are discussed with the 2.5-dimensional version being found to be the optimal compromise between computation speed and image quality. Basically, the Canon as well as the GE deep learning algorithms are of image restoration and not of image reconstruction type because they convert noisy images into noise-reduced images.
The impact of such deep learning–based image reconstruction algorithms in clinical practice needs to be proven.
Other Deep Learning Applications in Image Reconstruction
Besides noise reduction, deep learning can be implemented for artifact reduction. Convolutional neural network–based metal artifact reduction (MAR) approaches have been proposed,58,59 but it has not yet been demonstrated that these algorithms outperform conventional algorithms like NMAR and variants.60–63 Instead, one of the CNN approaches proposes to combine images of several MAR algorithms via deep learning.58 Deep learning can also be used for scatter estimation.64 Deep scatter estimation is a computational highly efficient alternative to Monte Carlo–based scatter estimation; although the reference standard needs minutes or hours to compute the x-ray scatter, deep scatter estimation is performed in less than a second and is also applicable in cases of severe data truncation.64 The same applies to estimating dose distributions that can be used to compute dose and risk more accurately but also to design more dose efficient scan protocols. Whereas Monte Carlo requires hours, deep dose estimation performs in seconds (Fig. 8).65
Most imaging modalities suffer from patient motion. Although broader detector arrays, higher gantry rotation speed, and dual-source technology dramatically reduced acquisition time, temporal resolution is in the order of 125 milliseconds for single and of 63 milliseconds for dual-source systems, which may still not be fast enough to freeze the cardiac motion.66–68 Given velocities up to 50 mm/s and more, residual motion blurring is likely to be present in a nonnegligible amount of cardiac CT examinations.
Acquiring images outside the target phase, for example, one image 10% before and after the target phase allows for estimation of cardiac motion, and once the motion is known, a reconstruction that accounts for that motion can be performed. Snapshot freeze (GE healthcare) is one example of this class of algorithms.69,70 Other approaches that make use of such data redundancies (more than 180 degrees of data) operate in sinogram domain.71,72 If no additional data are available and only a single image can be reconstructed, motion could potentially be estimated analyzing artifacts. One solution to maximize local sharpness is to minimize entropy. The algorithm choses motion direction and velocity is locally to obtain images with lowest entropy, potentially combined with positivity constraints. MAM and PAMoCo are examples for these approach algorithms.73,74Figure 9 shows an example of the PAMoCo algorithm: partial angle images are reconstructed (3 in the illustration, typically more than 10 in the implementation), and before these are combined (by summation), motion vectors are computed that result in a minimal entropy image when shifting the images according to these vectors before combining them.
Dual-Energy Computed Tomography
The introduction of dual-source systems revitalized the rather old idea of dual-energy CT (DECT).75–77 In the meantime, DECT is provided by all major CT manufacturers. There are different implementations of DECT in clinical routine: dual-source, rapid kV switching, dual-layer detector, dual-scan, and split-filter techniques. All approaches have their pro and cons; dual-source systems provide good separation of the spectra and mAs modulation, this approach has a limited field of view, a minimal temporal offset, a more complex scatter correction, and last but not least the necessity of a second x-ray tube, which increases the price of the system.1 Philips' approach utilizes single tube but 2 detector layers, an yttrium-based garnet scintillator on top and a gadolinium-oxysulphide scintillator at the bottom.78 This is currently the only DE implementation, where DE postprocessing can theoretically be performed retrospectively and without any temporal delay between the 2 datasets. In practice, higher tube voltages are preferable to improve spectral separation.78–82 The option to use DE retrospectively is a great advantage over the other DECT realizations. However, dual-layer technology comes at the cost of reduced spectral separation compared with other implementations.83
Photon counting detectors can discriminate 2 or more energy windows in a single-detector layer and are therefore intrinsically suited for dual-energy or multienergy CT applications. They also benefit from the retrospective aspect and thus allow to use the spectral information on demand, regardless of the scan protocol given that the tube voltage and prefiltration are set in an optimal way.
Various clinical applications of DECT have been developed and evaluated in vitro and in vivo,84–101 with virtual nonenhanced imaging, automated bone removal, urinary stone classification, gout imaging, MAR, and cardiac and pulmonary applications being the most relevant. Virtual monoenergetic imaging (VMI) can be used to reduce metal artifacts from orthopedic hardware, but in patients with bilateral implants, VMI was rated inferior to iterative MAR algorithms.102–104 Another implementation of VMI is to optimize iodine enhancement in CTA,93,97,105,106 which can be used to reduce the volume of the contrast material bolus, to recover CTA with suboptimal vascular enhancement107 or to create virtual arterial-phase images from portal venous phase acquisitions.108 Thus, VMI has the potential to reduce radiation or contrast material exposure (Fig. 10).
Although the race for speed and rows has slowed down in the last years, the fascination of CT has not declined at all. New detector concepts will provide higher in-plane resolution and the ability to perform CT at dose levels comparable to conventional radiography opens new horizons for screening programs. Machine learning algorithms for various tasks—image reconstruction, preprocessing, annotating, and even analyzing—are proliferating and will significantly change our daily work.
The authors thank Sabrina Dorn, Elias Eulig, Andreas Henneke, Bernice Hoppel, Jiang Hsieh, Patrik Rogalla, and Stefan Sawall for their support.
1. Lell MM, Wildberger JE, Alkadhi H, et al. Evolution in computed tomography
: the battle for speed and dose. Invest Radiol
2. Li B, Toth TL, Hsieh J, et al. Simulation and analysis of image quality impacts from single source, ultra-wide coverage CT scanner. J Xray Sci Technol
3. Forthmann P, Grass M, Proksa R. Adaptive two-pass cone-beam artifact correction using a FOV-preserving two-source geometry: a simulation study. Med Phys
4. Gordic S, Morsbach F, Schmidt B, et al. Ultralow-dose chest computed tomography
for pulmonary nodule detection: first performance evaluation of single energy scanning with spectral shaping. Invest Radiol
5. Newell JD Jr, Fuld MK, Allmendinger T, et al. Very low-dose (0.15 mGy) chest CT protocols using the COPDGene 2 test object and a third-generation dual-source CT scanner with corresponding third-generation iterative reconstruction
software. Invest Radiol
6. Messerli M, Kluckert T, Knitel M, et al. Ultralow dose CT for pulmonary nodule detection with chest x-ray equivalent dose—a prospective intra-individual comparative study. Eur Radiol
7. Primak AN, Giraldo JC, Eusemann CD, et al. Dual-source dual-energy CT with additional tin filtration: dose and image quality evaluation in phantoms and in vivo. AJR Am J Roentgenol
8. Feuchtner GM, Jodocy D, Klauser A, et al. Radiation dose reduction
by using 100-kV tube voltage in cardiac 64-slice computed tomography
: a comparative study. Eur J Radiol
9. Hendriks BMF, Eijsvoogel NG, Kok M, et al. Optimizing pulmonary embolism computed tomography
in the age of individualized medicine: a prospective clinical study. Invest Radiol
10. Lee KH, Lee JM, Moon SK, et al. Attenuation-based automatic tube voltage selection and tube current modulation for dose reduction
at contrast-enhanced liver CT. Radiology
11. Lell MM, Jost G, Korporaal JG, et al. Optimizing contrast media injection protocols in state-of-the art computed tomographic angiography. Invest Radiol
12. Lurz M, Lell MM, Wuest W, et al. Automated tube voltage selection in thoracoabdominal computed tomography
at high pitch using a third-generation dual-source scanner: image quality and radiation dose performance. Invest Radiol
13. Flohr TG, Stierstorfer K, Suss C, et al. Novel ultrahigh resolution data acquisition and image reconstruction
for multi-detector row CT. Med Phys
14. Meyer M, Haubenreisser H, Raupach R, et al. Initial results of a new generation dual source CT system using only an in-plane comb filter for ultra-high resolution temporal bone imaging. Eur Radiol
15. Kachelriess M, Kalender WA. Presampling, algorithm factors, and noise: considerations for CT in particular and for medical imaging in general. Med Phys
16. Baek J, Pineda AR, Pelc NJ. To bin or not to bin? The effect of CT system limiting resolution on noise and detectability. Phys Med Biol
17. Pourmorteza A, Symons R, Henning A, et al. Dose efficiency of quarter-millimeter photon-counting computed tomography
: first-in-human results. Invest Radiol
18. Leng S, Rajendran K, Gong H, et al. 150-μm Spatial resolution using photon-counting detector computed tomography
technology: technical performance and first patient images. Invest Radiol
19. Taguchi K, Zhang M, Frey EC, et al. Modeling the performance of a photon counting x-ray detector for CT: energy response and pulse pileup effects. Med Phys
20. Wang AS, Harrison D, Lobastov V, et al. Pulse pileup statistics for energy discriminating photon counting x-ray detectors. Med Phys
21. Willemink MJ, Persson M, Pourmorteza A, et al. Photon-counting CT: technical principles and clinical prospects. Radiology
22. Bartlett DJ, Koo CW, Bartholmai BJ, et al. High-resolution chest computed tomography
imaging of the lungs: impact of 1024 matrix reconstruction and photon-counting detector computed tomography
. Invest Radiol
23. Gutjahr R, Halaweish AF, Yu Z, et al. Human imaging with photon counting-based computed tomography
at clinical dose levels: contrast-to-noise ratio and cadaver studies. Invest Radiol
24. Pourmorteza A, Symons R, Sandfort V, et al. Abdominal imaging with contrast-enhanced photon-counting CT: first human experience. Radiology
25. Symons R, Reich DS, Bagheri M, et al. Photon-counting computed tomography
for vascular imaging of the head and neck: first in vivo human results. Invest Radiol
26. von Spiczak J, Mannil M, Peters B, et al. Photon counting computed tomography
with dedicated sharp convolution kernels: tapping the potential of a new technology for stent imaging. Invest Radiol
27. Zhou W, Bartlett DJ, Diehn FE, et al. Reduction of metal artifacts and improvement in dose efficiency using photon-counting detector computed tomography
and tin filtration. Invest Radiol
28. Haaga JR, Miraldi F, MacIntyre W, et al. The effect of mAs variation upon computed tomography
image quality as evaluated by in vivo and in vitro studies. Radiology
29. Greess H, Wolf H, Baum U, et al. Dose reduction
in computed tomography
by attenuation-based on-line modulation of tube current: evaluation of six anatomical regions. Eur Radiol
30. Kalender WA, Wolf H, Suess C. Dose reduction
in CT by anatomically adapted tube current modulation. II. Phantom measurements. Med Phys
31. Saltybaeva N, Schmidt B, Wimmer A, et al. Precise and automatic patient positioning in computed tomography
: avatar modeling of the patient surface using a 3-dimensional camera. Invest Radiol
32. Fleischmann U, Pietsch H, Korporaal JG, et al. Impact of contrast media concentration on low-kilovolt computed tomography
angiography: a systematic preclinical approach. Invest Radiol
33. Lell MM, Fleischmann U, Pietsch H, et al. Relationship between low tube voltage (70 kV) and the iodine delivery rate (IDR) in CT angiography: an experimental in-vivo study. PLoS One
34. Faggioni L, Gabelloni M. Iodine concentration and optimization in computed tomography
angiography: current issues. Invest Radiol
35. Higashigaito K, Husarik DB, Barthelmes J, et al. Computed tomography
angiography of coronary artery bypass grafts: low contrast media volume protocols adapted to tube voltage. Invest Radiol
36. Eller A, May MS, Scharf M, et al. Attenuation-based automatic kilovolt selection in abdominal computed tomography
: effects on radiation exposure and image quality. Invest Radiol
37. Eller A, Wuest W, Kramer M, et al. Carotid CTA: radiation exposure and image quality with the use of attenuation-based, automated kilovolt selection. AJNR Am J Neuroradiol
38. Winklehner A, Goetti R, Baumueller S, et al. Automated attenuation-based tube potential selection for thoracoabdominal computed tomography
angiography: improved dose effectiveness. Invest Radiol
39. Papadakis AE, Damilakis J. Automatic tube current modulation and tube voltage selection in pediatric computed tomography
: a phantom study on radiation dose and image quality. Invest Radiol
40. Runge VM, Marquez H, Andreisek G, et al. Recent technological advances in computed tomography
and the clinical impact therein. Invest Radiol
41. Lell MM, May MS, Brand M, et al. Imaging the parasinus region with a third-generation dual-source CT and the effect of tin filtration on image quality and radiation dose. AJNR Am J Neuroradiol
42. Weis M, Henzler T, Nance JW Jr, et al. Radiation dose comparison between 70 kVp and 100 kVp with spectral beam shaping for non-contrast-enhanced pediatric chest computed tomography
: a prospective randomized controlled study. Invest Radiol
43. Hsieh J. Adaptive streak artifact reduction in computed tomography
resulting from excessive x-ray photon noise. Med Phys
44. Kachelriess M, Watzke O, Kalender WA. Generalized multi-dimensional adaptive filtering for conventional and spiral single-slice, multi-slice, and cone-beam CT. Med Phys
45. Hu-Wang E, Schuzer JL, Rollison S, et al. Chest CT scan at radiation dose of a posteroanterior and lateral chest radiograph series: a proof of principle in lymphangioleiomyomatosis. Chest
46. Ernst CW, Basten IA, Ilsen B, et al. Pulmonary disease in cystic fibrosis: assessment with chest CT at chest radiography dose levels. Radiology
47. Villanueva-Meyer JE, Naeger DM, Courtier JL, et al. Pediatric chest CT at chest radiograph doses: when is the ultralow-dose chest CT clinically appropriate? Emerg Radiol
48. Nagatani Y, Takahashi M, Murata K, et al. Lung nodule detection performance in five observers on computed tomography
(CT) with adaptive iterative dose reduction
using three-dimensional processing (AIDR 3D) in a Japanese multicenter study: comparison between ultra-low-dose CT and low-dose CT by receiver-operating characteristic analysis. Eur J Radiol
49. Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction
technique for ultralow-dose chest CT: comparison of pulmonary nodule detectability with the adaptive statistical iterative reconstruction
technique. Invest Radiol
50. Kim Y, Kim YK, Lee BE, et al. Ultra-low-dose CT of the thorax using iterative reconstruction
: evaluation of image quality and radiation dose reduction
. AJR Am J Roentgenol
51. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature
52. Kim J, Hong J, Park H. Prospects of deep learning for medical imaging. Precis Future Med
53. Hornik K, Stinchcombe M, White H. Multilayer feedforward networks are universal approximators. Neural Netw
54. Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets. In: Ghahramani Z, Welling M, Cortes C, et al, eds. Proceedings of the 27th International Conference on Neural Information Processing Systems, Vol. 2 (NIPS'14)
. Cambridge, MA: MIT Press; 2014:2672–2680.
55. Isola P, Zhu JY, Zhou T, et al. Image-to-image translation with conditional adversarial networks. arXiv:161107004v3 [csCV]
56. Wolterink JM, Leiner T, Viergever MA, et al. Generative adversarial networks for noise reduction
in low-dose CT. IEEE Trans Med Imaging
57. Ziabari A, Ye DH, Srivastava S, et al. 2.5D deep learning for CT image reconstruction
using a multi-GPU implementation; 2018.
58. Zhang Y, Yu H. Convolutional neural network based metal artifact reduction in x-ray computed tomography
. IEEE Trans Med Imaging
59. Park HS, Lee SM, Kim HP, et al. CT sinogram-consistency learning for metal-induced beam hardening correction. Med Phys
60. Meyer E, Raupach R, Lell M, et al. Normalized metal artifact reduction (NMAR) in computed tomography
. Med Phys
61. Meyer E, Raupach R, Lell M, et al. Frequency split metal artifact reduction (FSMAR) in computed tomography
. Med Phys
62. Lell MM, Meyer E, Kuefner MA, et al. Normalized metal artifact reduction in head and neck computed tomography
. Invest Radiol
63. Lell MM, Meyer E, Schmid M, et al. Frequency split metal artefact reduction in pelvic computed tomography
. Eur Radiol
64. Maier J, Eulig E, Vöth T, et al. Real-time scatter estimation for medical CT using the deep scatter estimation: method and robustness analysis with respect to different anatomies, dose levels, tube voltages, and data truncation. Med Phys
65. Maier J, Eulig E, Dorn S, et al. Real-time patient-specific CT dose estimation using a deep convolutional neural network. Proc IEEE MIC
66. Achenbach S, Ropers D, Holle J, et al. In-plane coronary arterial motion velocity: measurement with electron-beam CT. Radiology
67. Vembar M, Garcia MJ, Heuscher DJ, et al. A dynamic approach to identifying desired physiological phases for cardiac imaging using multislice spiral CT. Med Phys
68. Husmann L, Leschka S, Desbiolles L, et al. Coronary artery motion and cardiac phases: dependency on heart rate—implications for CT image reconstruction
69. Andreini D, Pontone G, Mushtaq S, et al. Low-dose CT coronary angiography with a novel IntraCycle motion-correction algorithm in patients with high heart rate or heart rate variability. Eur Heart J Cardiovasc Imaging
70. Pontone G, Andreini D, Bertella E, et al. Impact of an intra-cycle motion correction algorithm on overall evaluability and diagnostic accuracy of computed tomography
coronary angiography. Eur Radiol
71. Kim S, Chang Y, Ra JB. Cardiac motion correction based on partial angle reconstructed images in x-ray CT. Med Phys
72. Kim S, Chang Y, Ra JB. Cardiac motion correction for helical CT scan with an ordinary pitch. IEEE Trans Med Imaging
73. Rohkohl C, Bruder H, Stierstorfer K, et al. Improving best-phase image quality in cardiac CT by motion correction with MAM optimization. Med Phys
74. Hahn J, Bruder H, Rohkohl C, et al. Motion compensation in the region of the coronary arteries based on partial angle reconstructions from short-scan CT data. Med Phys
75. Genant HK, Boyd D. Quantitative bone mineral analysis using dual energy computed tomography
. Invest Radiol
76. Kalender WA, Perman WH, Vetter JR, et al. Evaluation of a prototype dual-energy computed tomographic apparatus. I. Phantom studies. Med Phys
77. Vetter JR, Perman WH, Kalender WA, et al. Evaluation of a prototype dual-energy computed tomographic apparatus. II. Determination of vertebral bone mineral content. Med Phys
78. Rassouli N, Etesami M, Dhanantwari A, et al. Detector-based spectral CT with a novel dual-layer technology: principles and applications. Insights Imaging
79. Abdullayev N, Große Hokamp N, Lennartz S, et al. Improvements of diagnostic accuracy and visualization of vertebral metastasis using multi-level virtual non-calcium reconstructions from dual-layer spectral detector computed tomography
. Eur Radiol
80. Grosse Hokamp N, Abdullayev N, Persigehl T, et al. Precision and reliability of liver iodine quantification from spectral detector CT: evidence from phantom and patient data. Eur Radiol
81. Neuhaus V, Grosse Hokamp N, Zopfs D, et al. Reducing artifacts from total hip replacements in dual layer detector CT: combination of virtual monoenergetic images and orthopedic metal artifact reduction. Eur J Radiol
82. van Ommen F, de Jong HWAM, Dankbaar JW, et al. Dose of CT protocols acquired in clinical routine using a dual-layer detector CT scanner: a preliminary report. Eur J Radiol
83. Faby S, Kuchenbecker S, Sawall S, et al. Performance of today's dual energy
CT and future multi energy CT in virtual non-contrast imaging and in iodine quantification: A simulation study. Med Phys
84. Taguchi K, Itoh T, Fuld MK, et al. "X-Map 2.0" for edema signal enhancement for acute ischemic stroke using non-contrast-enhanced dual-energy computed tomography
. Invest Radiol
85. Sofue K, Itoh T, Takahashi S, et al. Quantification of cisplatin using a modified 3-material decomposition algorithm at third-generation dual-source dual-energy computed tomography
: an experimental study. Invest Radiol
86. Mohammed MF, Marais O, Min A, et al. Unenhanced dual-energy computed tomography
: visualization of brain edema. Invest Radiol
87. Martin SS, Weidinger S, Czwikla R, et al. Iodine and fat quantification for differentiation of adrenal gland adenomas from metastases using third-generation dual-source dual-energy computed tomography
. Invest Radiol
88. Kim H, Goo JM, Kang CK, et al. Comparison of iodine density measurement among dual-energy computed tomography
scanners from 3 vendors. Invest Radiol
89. Khodarahmi I, Haroun RR, Lee M, et al. Metal artifact reduction computed tomography
of arthroplasty implants: effects of combined modeled iterative reconstruction
and dual-energy virtual monoenergetic extrapolation at higher photon energies. Invest Radiol
90. Große Hokamp N, Salem J, Hesse A, et al. Low-dose characterization of kidney stones using spectral detector computed tomography
: an ex vivo study. Invest Radiol
91. Frellesen C, Azadegan M, Martin SS, et al. Dual-energy computed tomography
-based display of bone marrow edema in incidental vertebral compression fractures: diagnostic accuracy and characterization in oncological patients undergoing routine staging computed tomography
. Invest Radiol
92. Diekhoff T, Kotlyarov M, Mews J, et al. Iterative reconstruction
may improve diagnosis of gout: an ex vivo (bio)phantom dual-energy computed tomography
study. Invest Radiol
93. Weiss J, Notohamiprodjo M, Bongers M, et al. Effect of noise-optimized monoenergetic postprocessing on diagnostic accuracy for detecting incidental pulmonary embolism in portal-venous phase dual-energy computed tomography
. Invest Radiol
94. Nute JL, Jacobsen MC, Chandler A, et al. Dual-energy computed tomography
for the characterization of intracranial hemorrhage and calcification: a systematic approach in a phantom system. Invest Radiol
95. May MS, Bruegel J, Brand M, et al. Computed tomography
of the head and neck region for tumor staging-comparison of dual-source, dual-energy and low-kilovolt, single-energy acquisitions. Invest Radiol
96. Mannil M, Ramachandran J, Vittoria de Martini I, et al. Modified dual-energy algorithm for calcified plaque removal: evaluation in carotid computed tomography
angiography and comparison with digital subtraction angiography. Invest Radiol
97. Leithner D, Wichmann JL, Vogl TJ, et al. Virtual monoenergetic imaging and iodine perfusion maps improve diagnostic accuracy of dual-energy computed tomography
pulmonary angiography with suboptimal contrast attenuation. Invest Radiol
98. Bongers MN, Bier G, Kloth C, et al. Improved delineation of pulmonary embolism and venous thrombosis through frequency selective nonlinear blending in computed tomography
. Invest Radiol
99. Wichmann JL, Gillott MR, De Cecco CN, et al. Dual-energy computed tomography
angiography of the lower extremity runoff: impact of noise-optimized virtual monochromatic imaging on image quality and diagnostic accuracy. Invest Radiol
100. Kaemmerer N, Brand M, Hammon M, et al. Dual-energy computed tomography
angiography of the head and neck with single-source computed tomography
: a new technical (split filter) approach for bone removal. Invest Radiol
101. Hwang HJ, Seo JB, Lee SM, et al. Assessment of regional xenon ventilation, perfusion, and ventilation-perfusion mismatch using dual-energy computed tomography
in chronic obstructive pulmonary disease patients. Invest Radiol
102. Andersson KM, Norrman E, Geijer H, et al. Visual grading evaluation of commercially available metal artefact reduction techniques in hip prosthesis computed tomography
. Br J Radiol
103. Andersson KM, Nowik P, Persliden J, et al. Metal artefact reduction in CT imaging of hip prostheses-an evaluation of commercial techniques provided by four vendors. Br J Radiol
104. Higashigaito K, Angst F, Runge VM, et al. Metal artifact reduction in pelvic computed tomography
with hip prostheses: comparison of virtual monoenergetic extrapolations from dual-energy computed tomography
and an iterative metal artifact reduction algorithm in a phantom study. Invest Radiol
105. Canellas R, Digumarthy S, Tabari A, et al. Radiation dose reduction
in chest dual-energy computed tomography
: effect on image quality and diagnostic information. Radiol Bras
106. Chen T, Xiao H. Does dual-energy computed tomography
pulmonary angiography (CTPA) have improved image quality over routine single-energy CTPA? J Med Imaging Radiat Oncol
107. Bae K, Jeon KN, Cho SB, et al. Improved opacification of a suboptimally enhanced pulmonary artery in chest CT: experience using a dual-layer detector spectral CT. AJR Am J Roentgenol
108. Patel AA, Sutphin PD, Xi Y, et al. Arterial phase CTA replacement by a virtual arterial phase reconstruction from a venous phase CTA: preliminary results using detector-based spectral CT. Cardiovasc Intervent Radiol