State of the Art: Iterative CT Reconstruction Techniques

Home , Iterative reconstruction

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/280390600

Article in Radiology · August 2015 Impact Factor: 6.87 · DOI: 10.1148/radiol.2015132766 · Source: PubMed

CITATIONS READS 2 52

10 authors, including:

Uwe Joseph Schoepf Jonathon Leipsic Medical University of South Carolina Providence Health Care

636 PUBLICATIONS 11,837 CITATIONS 360 PUBLICATIONS 5,432 CITATIONS

SEE PROFILE SEE PROFILE

Marco Rengo Andrea Laghi Sapienza University of Rome Sapienza University of Rome

122 PUBLICATIONS 462 CITATIONS 450 PUBLICATIONS 6,019 CITATIONS

SEE PROFILE SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Carlo Nicola De Cecco letting you access and read them immediately. Retrieved on: 31 May 2016 Reviews and Commentary n State of the Art 339 ------

1 Iterative CT Iterative www.rsna.org/rsnarights. effects depends on the specific IR algorithm. In the first In the first effects depends on the specific IR algorithm. bases of IR are section of this contribution, the technical re algorithms available and the currently reviewed briefly In described. are manufacturers the major CT leased by status of their clinical imple the second part, the current of the applied IR algo Regardless mentation is surveyed. attests to the substantial evidence rithm, the available limi traditional potential of IR algorithms for overcoming imaging. tations in CT © RSNA, 2015 Owing to recent advances in computing power, iterative iterative in computing power, advances Owing to recent become a clinically have (IR) algorithms reconstruction imaging. (CT) in computed tomographic option viable about the advantages is accumulating Substantial evidence such methods, established analytical of IR algorithms over image quality IR improves back projection. as filtered so Although all available image processing. cyclic through the common mechanism of artifact reduc lutions share chiefly dose savings, tion and/or potential for radiation the magnitude of these due to image noise suppression,

Reconstruction Techniques Reconstruction State of the Art: of the State

radiology.rsna.org

ust 2015 This copy is for your personal non-commercial use only. To order presentation-ready presentation-ready order To personalonly. use non-commercial for your is copy This

Aug ). Address correspondence to Note: at us contact clients, or colleagues distribution for copies to your

2 2 The Russell H. Morgan Department of The Russell H. Institute for Clinical Radiology, Institute for Clinical Radiology, [email protected] Volume 276: Number 2— 276: Volume RSNA, RSNA, 2015

From the Department of Radiology and Radiological From Current address: Current address:

Ludwig-Maximilians-University Hospital, Munich, Germany. Munich, Ludwig-Maximilians-University Hospital, q 3 Radiology and Johns Hopkins Hospital, Radiological Science, Md. Baltimore, final version accepted May 5. final version accepted May 5. (e-mail: U.J.S. 2 Sciences, Oncology and Pathology, University of Rome University of Rome Oncology and Pathology, Sciences, C.N.D.). A.L., Italy (M.R., Latina, Pontino, Sapienza–Polo 2013; revision requested January Received December 17, April 3; 2014; final revision received March 18; accepted 15, Canada (G.B.); Department of Radiology, University of British Canada (G.B.); Department of Radiology, Canada (J.A.L.); Department of BC, Vancouver, Columbia, Toronto, University of General Hospital, Toronto Radiology, of Radiological and Department Canada (N.S.P.); Ont, Toronto, Science, Medical University of South Carolina, Ashley River Medical University of South Carolina, Science, SC 29425 Charleston, 25 Courtenay Dr, MSC 226, Tower, C.N.D.); Department of J.W.N., F.G.M., U.J.S., (L.L.G., Ont, Toronto, Sunnybrook Health Sciences Centre, Radiology, 1 Radiology: Lucas L. Geyer, MD Geyer, Lucas L. Felix G. Meinel, MD Meinel, Felix G. U. Joseph Schoepf, MD MD Joseph Schoepf, U. Carlo N. De Cecco, MD De Cecco, Carlo N. Jonathon A. Leipsic, MD Leipsic, A. Jonathon MD Paul, Narinder S. PhD MD, Marco Rengo, MD Andrea Laghi, Gorka Bastarrika, MD Gorka Bastarrika, John W. Nance, Jr, MD Jr, Nance, W. John STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

omputed tomographic (CT) tech- 256- (9), and 320-detector (10) single- general technical evolution providing nology has seen remarkable in- source or dual-source CT systems (11). the required computational power. Fur- Cnovations in the past decade However, the increased number of de- thermore, the increasing number of CT that have substantially improved the tector rows and detector technology examinations worldwide and the asso- diagnostic performance of this modal- are only one domain of CT evolution. ciated radiation dose to the population ity and steadily increased its clinical While advances in CT hardware con- have clearly fostered the rediscovery of indications. Since its first clinical in- tinue to expand the boundaries of phys- IR technology as a promising tool to de- troduction by Sir Godfrey Hounsfield ical limitations, increases in computing crease radiation requirements via noise and James Ambrose in 1972 (1,2), the power have opened additional pathways reduction. evolution of CT technology has mainly for improving the performance of this In this contribution, we review the been driven by advances in hardware. modality via enhanced data process- technical bases of IR and describe the During subsequent decades, important ing methods, such as reconstruction currently available algorithms released milestones have included the introduc- techniques. The most prominent exam- by the major CT manufacturers. Fur- tion of electron-beam CT in the mid- ple of recent years is the renaissance ther, we survey the current status of 1980s (3), spiral (helical) CT imaging of iterative reconstruction (IR) CT al- their clinical implementation. Regard- in 1989 (4), and multi–detector row CT gorithms. IR approaches are not new less of the applied IR algorithm, the in 1998 (5–7). Currently, the major CT and were, in fact, the initially proposed available evidence attests to the sub- manufacturers offer a variety of 64- (8), method for data reconstruction in the stantial potential of IR algorithms for early days of CT technology during the overcoming traditional limitations in Essentials 1970s (2). However, due to its mathe- CT imaging. matically demanding properties and the nn Iterative reconstruction (IR) tech- large amount of data in CT imaging, un- niques allow for substantial radia- til recently IR has not been practical for Technical Background tion dose savings through noise clinical purposes. Instead, this recon- The exact underlying computational al- reduction in CT image processing. struction technique became the default gorithms of the currently available IR nn IR can be used to improve image method for nuclear medicine emission algorithms are mostly considered pro- quality and reduce noise through- tomography imaging modalities with prietary and only partly revealed by out the body, particularly in obese lower spatial and temporal resolution, the manufacturers. However, published patients. such as single photon emission CT and data indicate that these algorithms can nn Besides improvements in general positron emission tomography, because differ substantially with respect to the measures of image quality, an in- of the smaller data volumes and less underlying assumptions of data acqui- creasing number of reports are complex data handling (12). The less sition, data processing, system geome- emerging on enhanced diagnostic perfect, albeit much faster, analytical ap- tries, and noise characteristics. Never- accuracy and artifact suppression proach of filtered back projection (FBP) theless, the following sections attempt with use of IR. has become the standard reconstruction to provide an objective description of nn Special attention should be paid to method for diagnostic CT. the currently available IR techniques. quantitative CT imaging applica- FBP has been established in clinical Pertinent Principles of CT Data tions, as the use of IR may alter routine due to its ability to generate Acquisition standards established on the basis CT studies of adequate image quality of prior analytical image recon- in a robust and fast manner. Despite The fundamental goal of CT data ac- struction methods. its overall acceptable performance, CT quisition and reconstruction is to as- studies that are reconstructed with FBP nn Robust data regarding the impact can be affected by high image noise, and safety of IR in the clinical set- artifacts (eg, streak artifacts), or poor Published online ting are available; thus, routine 10.1148/radiol.2015132766 Content code: low-contrast detectability in specific implementation of IR in CT proto- clinical scenarios. For example, data Radiology 2015; 276:339–357 cols should be strongly considered. acquisition with reduced tube output nn While a multitude of reports high- Abbreviations: or CT imaging of obese patients is of- AIDR = adaptive iterative dose reduction light the promise of IR to enhance ten compromised by high image noise; ASIR = adaptive statistical iterative reconstruction diagnostic performance and high-density structures, such as calci- BMI = body mass index reduce radiation at CT, actual ex- fications or stents, result in blooming CNR = contrast-to-noise ratio amples of adjustment to lower artifacts; metallic implants or bone FBP = filtered back projection radiation dose settings to fully structures might lead to severe streak IR = iterative reconstruction IRIS = iterative reconstruction in image space implement the benefits of IR algo- artifacts. These particular shortcom- SAFIRE = sinogram-affirmed iterative reconstruction rithms in daily clinical practice are ings of FBP likely have driven the re- still limited. naissance of IR algorithms along with Conflicts of interest are listed at the end of this article.

340 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

sign an attenuation value to each voxel Figure 1 of a three-dimensional volume. Data acquisition is performed by transmit- ting a fan of photons in multiple angles through the body to an array of detec- tors. The data at each detector repre- sent the sum of the attenuation of all tissues through which the beam has passed; this is the “raw data.” Recon- struction algorithms use the raw data to determine attenuation values for each voxel; differences between reconstruction techniques involve determining how this attenuation value is assigned in the final image. There are two major classes of reconstruction algorithms: analytical and iterative.

Analytical Image Reconstruction–FBP Knowledge of the basic properties of FBP is crucial to understand the ben- Figure 1: Simplified schematic of CT data reconstruction. Traditionally, several simplifications concerning efits of IR. Analytical reconstruction the data acquisition process are made in the context of FBP: pencil-beam geometry of the x-ray, focal spot algorithms such as FBP are based on as an infinitely small point, intensity measured on a point located at the detector cell center. Regarding a the assumption that both the measure- single x-ray, photons with a known intensity are transmitted from the x-ray source through an object to the ment process and the projection data detector. According to the law of attenuation, the transmitted intensity decreases exponentially due to ab- are represented by continuous func- sorption within the object resulting in a lower measured intensity. Multiple x-rays result in the measurement tions. In a simplified model, the x-ray of intensity profiles in the CT detector. By preprocessing, intensity values are transformed into attenuation beam is collimated to a pencil shape values (projection data). Then, projection data are filtered using different reconstruction algorithms (kernels) and moved subsequently parallel to a to create specific image characteristics for soft-tissue or high-contrast visualization. Finally, the measured linear x-ray detector array. Then, the projection data are propagated into the image domain (back projection). Multiple projections are needed to x-ray source is rotated by an angle a solve the mathematical system with multiple equations and variables to generate the final CT image. and the process is repeated. The in- tensities measured at the detector are mathematically described as an but also increases image noise. Differ- and speed. A major limiting feature of integral function for a specific anglea ent kernels enable optimized depiction FBP is that it fails to account for image and a particular linear shift position of of soft-tissue or high-contrast struc- noise that results from Poisson statisti- the x-ray tube (Fig 1). The reconstruc- tures, such as bone or lung tissue. It cal variations in photon number across tion process is the solution of the re- is a characteristic of FBP that image the image plane; practically speaking, sulting integral equations by inversion sharpness and image noise are directly this means that a reduction in radia- (back projection). The back projection coupled: The sharper the image, the tion dose translates into an increase that describes the propagation of the higher the image noise. With the evo- in image noise. High image noise in- measured projection data into the im- lution of CT hardware, adaptations, terferes with the delineation and low- age domain is traditionally combined such as interpolation methods or use contrast detectability of a structure, with a filter component (eg, Ram-Lak of the Feldkamp algorithm or other so that certain minimal radiation dose filter). The filter compensates for the three-dimensional methods, have been requirements need to be fulfilled to effect of the so-called low-pass blur applied to compensate for fan-beam generate a diagnostic CT data set. that occurs because of the differ- and cone-beam geometries, respec- Lowering image noise by choosing ent numbers of projections passing tively. Those approaches, however, still “smoother” kernels for image recon- through the center and the periphery remain approximations and interpola- struction will result in impaired spatial of an object. In clinical practice, fur- tions to satisfy underlying assumptions resolution with use of a conventional ther variations of the filter (kernels) such as a point x-ray source, a pencil FBP technique. can be chosen, which are contingent x-ray beam, and the point of detec- upon a compromise between spatial tor elements, which are prerequisites Iterative Image Reconstruction resolution and image noise. Increasing for the implementation of the Radon In general, the process of data acquisi- compensation of the low-pass blur in- transform. The main advantages of tion can be described by the following creases the “sharpness” of the image, this approach consist in its robustness formula: p = Hf + n, where the measured

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 341 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 2 the so-called data term combined with a regularization term (or prior term). While the data term is a fitting model of the observed projection data, the regularization term often incorporates the nonuniformities of the CT system, such as noise. In a so-called statistical IR, a weighting term is introduced into the data term that assigns low weight to data with high statistical uncertainty (high noise) and high weight to data with low statistical uncertainty (low noise). Data fitting can be mathematically achieved by different statistical methods, such as maximum likelihood, least squares, or maximum a posteri- ori estimators. Variations of both the data and the regularization term result in different characteristics, mainly affecting the handling of image noise and artifacts.

Hybrid Algorithms So-called hybrid algorithms com- Figure 2: Schematic representation of the principle steps of iterative image algorithms. Following the CT bine both analytical and iterative acquisition process (measured projections), a first image estimate is generated. An x-ray beam is simulated methods in different combinations. via forward projection to obtain simulated projection data, which are then compared with the measured In one arrangement, the initial image projection data. In case of discrepancy, the first image estimate is updated based on the characteristics of is generated by the use of analytical the underlying algorithm. This correction of image and projection data is repeated until a condition predefined methods (raw data domain), and iter- by the algorithm is satisfied and the final image is generated. ative methods are focused to optimize image characteristics, for example, projection data p is related to the real performance: simultaneous iterative noise, in the image domain. In an- data f (attenuation coefficient) through reconstruction technique and simulta- other pairing, an iterative algorithm a projection process H and the ad- neous algebraic reconstruction tech- can be directly implemented into the ditional noise n. The image recon- nique. For further details, we refer to reconstruction process to focus on struction averages the solution of this previous literature (13–15). However, image improvements of an initial im- equation that can be achieved by two as computational power was limited in age estimate that is generated by an mathematically different iterative con- the early days of CT technology, IR al- analytical method. In the literature, cepts: algebraic algorithms and statisti- gorithms were not practical for clinical the term hybrid IR usually refers to cal algorithms. application. algorithms that mainly decrease image The principle of iterative image algo- The example mentioned above also noise by iterative methods. In con- rithms is based on six key steps (Fig 2). illustrates that the complexity of IR trast, the term model-based iterative For a better understanding of the the- algorithms rapidly increased when ad- reconstruction usually refers to algo- ory and complexity of iterative image ditional components of the data acqui- rithms that implement models of the reconstruction, Figure 3 illustrates a sition process or image characteristics acquisition process, image statistics, simplified model. Leaving additional were integrated. In addition to different and system geometry. However, we noise n aside, the algebraic algorithm H sources of image noise (eg, statistical find it important to emphasize that the solves a simple system of linear equa- photon distribution, electronic noise), clinical performance of IR algorithms tions, where the projection value is the geometry of modern CT systems is not necessarily related to the com- the sum of two attenuation coefficients (eg, shape and size of the detector and plexity of the method. Consequently, along each projection line. In the early the focal spot, distances between the we do not further distinguish between 1970s, the first IR algorithm—algebraic x-ray tube, the isocenter, the detector, both approaches in the clinical section reconstruction technique—was imple- etc) contributes substantially to the of the manuscript. Currently commer- mented, disregarding the additional projection process. cially available IR algorithms are uti- noise n. Later on, two modified algo- Basically, the mathematical model lizing a broad spectrum of the above- rithms were developed to improve the of IR methods consists of two parts: described principles.

342 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 3

Figure 3: Simplified model of an algebraic IR cycle. Four different attenuation coefficients, in a 23 2 pixel matrix, are represented by five projections(p) at three different angles (two acquired in horizontal, two in vertical, and one in oblique directions). The matrix is succes- sively updated by stepwise back projection. The corrected attenuation coefficients can be used to generate synthesized projection data (P’) via forward projection. A subsequent cycle can be initiated until a stop criterion is satisfied.

dation of image quality, with a somewhat data from the physics of the interaction Vendor-specific IR Approaches unfamiliar, almost “plastic” texture to the of x-rays with matter. Similar to other images (22,23). IR solutions, parts of model-based iter- GE Healthcare Veo.—Veo, initially introduced as ative reconstruction can be initialized Adaptive statistical iterative reconstruc- model-based iterative reconstruction, with a FBP reconstruction to facilitate tion.—In late 2008, GE Healthcare is the second-generation IR algorithm a relatively fast convergence. Then, (Waukesha, Wis) introduced their first introduced by GE Healthcare (24). The all voxels of the image volume are up- hybrid adaptive statistical iterative re- calculation process of Veo is complex dated within one complete iterative construction (ASIR) algorithm for clinical and exceeds the scope of this article. cycle. This extensive modeling and its use (16). ASIR, unlike FBP, performs re- We therefore refer to the publications complexity are demanding on computa- construction of the CT data sets by mod- by Yu et al (25) and Thibault et al (26) tional power and time; currently, recon- eling the system statistics in the process for more detailed descriptions. In brief, struction times range between 10 and (17–21), using information obtained from this algorithm incorporates an extensive 90 minutes depending on the number the FBP algorithm as a building block for three-dimensional model of the data ac- of images, which roughly equals 0.2 to each individual image reconstruction. quisition process, including system op- 0.5 images per second (27). This po- The ASIR model integrates matrix al- tics (eg, geometry of the x-ray source, tential delay between data acquisition gebra to convert the measured value of cone-beam shape, detector characteris- and availability of images for interpre- each pixel to a new estimate of the pixel tics), in addition to the models of the tation has to be considered in clinical value. This pixel value is then evaluated statistical noise and the prior term. The practice, for example, for emergent in- and is compared with the ideal value that model of the system optics describes dications. is predicted with noise modeling. The how each element of a scanned object process is repeated in successive itera- is projected onto the detector, disre- Philips Healthcare tive steps until the final estimated and garding the simplified assumptions of iDose4.—In 2010, Philips Healthcare ideal pixel values ultimately converge. FBP. Veo assumes a three-dimensional (Best, the Netherlands) introduced their ASIR is blended with traditional FBP in volume of each voxel element and takes approach to IR techniques with iDose4. 10% increments according to user pref- a focal spot with known dimensions, as The iDose4 reconstruction algorithm first erence. However, similar to other IR well as an active area of the detector, analyzes the projection data, identifying techniques, a higher percentage of ASIR into account. Veo also models the sta- and correcting the noisiest CT measure- in the reconstruction can result in degra- tistical distribution of the measured ments (poor signal-to-noise ratio or very

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 343 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

low photon counts). A model including (Forchheim, Germany) released their reconstruction of a typical thorax exam- the photon statistics is applied to each first-generation iterative reconstruction ination of 30 cm in 15 seconds. projection for the detection of very noisy algorithm, iterative reconstruction in Advanced modeled iterative recon measurements. Through an iterative image space (IRIS) (31). This approach struction.—Recently, Siemens released process, the noisy data are penalized is based in the image domain, where an their third-generation IR algorithm. Ad- and edges are preserved. This process initial image is reconstructed from the vanced modeled iterative reconstruc- ensures that the attenuation gradients raw data upon which three to five iteration comprises three modifications, of underlying structures are retained, tions of the algorithm are then applied, compared with previous algorithms thus preserving spatial resolution while with the goal of reducing noise and en- (36): (a) the use of a weighted FBP allowing a substantial noise reduction. hancing object contrast step-by-step. in the loop, which aims at improved The noise left after this process is prop- Sinogram-affirmed iterative recon removal of artifacts based on geomet- agated to the image space; however, the struction.—Sinogram-affirmed iterative rically nonexact reconstruction opera- propagated noise is highly localized and reconstruction (SAFIRE) is the second- tors; (b) computations commence with can be removed to support the desired generation IR algorithm released by up to two iterations, with the goal of level of dose reduction. This technique Siemens Healthcare in 2010, which in- removing geometric imperfections such aims at preventing photon starvation corporates an IR technique that utilizes as cone-beam artifacts; (c) the statis- artifacts (streaks, bias) before image both raw data and image data iterations tical modeling performs a local signal- creation and maintaining image quality with up to five strength levels available to-noise ratio analysis to decompose while avoiding the artificial appearance for adaptation of the regularization data into information and noise ac- of images that has been typical of earlier term to control for image impression cording to the model. Compared with generation IR techniques. and noise reduction. The strength is SAFIRE, the analysis incorporates not From an operative point of view, the not related to the number of iteration only nearest-neighbor data but also a percentage of dose reduction (from 0% loops (32). larger area. to 80%) should be chosen before the Similar to traditional IR, SAFIRE acquisition, while iDose4 reconstruc- performs an initial reconstruction using Toshiba Medical Systems tion levels (from 0 to 7) can be selected a weighted FBP, after which two differ- In the initial IR algorithm developed before the scan or after the acquisition ent correction loops are introduced into by Toshiba Medical Systems (Otawara, of raw data. To maintain the same im- the reconstruction process. In the first Japan), adaptive iterative dose reduc- age quality and noise of FBP, a propor- loop, new synthetic raw data (from a tion (AIDR), the image noise reduction tional iDose4 level should be selected forward projection) are compared with occurred in the reconstruction (image) according to the chosen percentage of the original raw data to derive correc- domain. This IR technique required that dose reduction (28). However, a higher tion projections that are then used to the original high-noise images undergo iDose4 level can be applied to increase reconstruct a correction image. The de- several loops of iteration to reduce the image resolution. Consequently, the tected deviations are again reconstruct- image noise until the desired noise user is able to prioritize a goal of dose ed using the weighted FBP, and the loop level is achieved (37,38). More recently reduction, image quality improvement, is repeated a number of times depend- this technique has been replaced by an or a weighted compromise between ing on the scan mode. The second cor- AIDR system using a three-dimensional both goals (29). rection loop occurs in image space, processing algorithm (AIDR 3D) (39). Iterative model reconstruction.— where noise is removed from the im- This reconstruction algorithm is based Iterative model reconstruction is the age through a statistical optimization on IR performed not only in the recon- second-generation IR algorithm intro- process. Noise can be locally estimated struction domain but also in the raw- duced by Philips Healthcare in 2012. In and removed by using a dynamic raw- data domain. In the raw-data domain, contrast to iDose4, iterative model re- database noise model that, during each AIDR 3D processing takes into consid- construction aims at accounting for not iteration, predicts the variance of the eration the quantum noise derived from only the noise behavior of the image but image noise in different directions the amount of x-ray photons that reach also the data statistics, image statistics, in each image pixel and adjusts the the detector and the electrical noise and system models during its iterative space-variant regularization function from the CT system; the algorithm also cycle (30). To our knowledge, however, correspondingly. Noise reduction oc- uses the raw-data domain in a model as of the time of this writing further de- curs almost solely in image space, thus accounting for the specific scanner tails about this very recently introduced reducing the requirement to return to geometry and a statistical noise model algorithm have not yet been made avail- raw data space. The corrected image to reduce noise (37,40). Finally, in an able in the imaging literature. is compared with the original, and the effort to maintain the noise granularity process is repeated a number of times and render the image more “natural,” Siemens Healthcare depending on the examination type a weighted blending is applied to the Iterative reconstruction in image (33,34). SAFIRE can reconstruct up to original reconstruction and the output space.—In 2008, Siemens Healthcare 20 images per second (35), allowing the of this iterative process (41).

344 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Clinical Applications acquisition, the technologist simply to-noise ratio, contrast-to-noise ratio selects the appropriate IR reconstruc- (CNR), low-contrast visualization, and The current evidence on the clinical tion algorithm (kernel) and the desired spatial resolution. Ghetti et al (44,45) implementation of IR into CT protocols strength level (ASIR, iDose4, SAFIRE), used Catphan (The Phantom Labo- shows substantial promise for major if applicable. Except for Veo—which is ratory, Greenwich, NY ), a specific improvements in image quality, chiefly currently only available with a standard phantom for the evaluation of medi- noise reduction—with subsequent radi- soft-tissue kernel—all IR algorithms cal imaging equipment, and dedicated ation dose reduction—and artifact sup- can be combined with a specific re- three-dimensional spatial resolution pression. The former suggests the op- construction algorithm, for example, phantoms to compare FBP with IRIS portunity for substantial radiation dose soft tissue, bone, lung, et cetera. In and SAFIRE, respectively, and found savings by mitigating the contrarian re- general, the choice of the IR algorithm that both Siemens products preserved lationship between dose and noise that and its strength level, if available, influ- spatial resolution while decreasing governs the use of FBP. However, exact ences the image impression and noise image noise when radiation dose was estimates of dose reduction in clinical characteristics. As a result, the selec- kept constant. Attenuation values were practice are difficult to derive from pub- tion of the institutionally preferred unchanged between the IR and FBP al- lished data because studies comparing IR technique is a specific clinical task gorithms, which led to proportionally various IR algorithms of the same ven- germane to the individual preferences increased CNR and low-contrast res- dor are scarce and comparisons of dif- for image quality: For instance, a more olution when the iterative techniques ferent vendors are largely missing. This aggressive noise reduction may be ben- were used. SAFIRE has the ability to may be partly due to the fact that virtu- eficial for the detection of low-contrast vary the iterative strength level from 1 ally all IR techniques are vendor-specif- structures such as hypovascular liver to 5, and the authors found that image ic with limited applicability to other CT lesions, whereas CT angiography ex- noise reduction increased from 10% systems. As expectations concerning aminations may benefit from strategies to 60% with increased strength. Simi- image quality can substantially differ aiming at improving spatial resolution lar studies comparing FBP with AIDR between institutions, careful attention or decreasing artifacts, rather than (37) and iDose (46) also found that to specific data acquisition protocols noise reduction. With first-generation the iterative products preserved spa- is required when reviewing literature IR algorithms in particular, a substantial resolution while decreasing image about IR technology. tial noise reduction might be associated noise when identical acquisition pa- Furthermore, recent data indi- with an “oversmoothing” of the image, rameters were used. iDose4 provided cate that also acquisition parameters leading to a blotchy appearance of the image noise reduction of 11%–55%, of FBP protocols could be improved IR-reconstructed studies (22,23). In depending on the iDose level, which by means of quality management, addition, second-generation IR algo- resulted in improved low-contrast res- such as periodical audits (42). Con- rithms allow for a more effective arti- olution compared with FBP acquired sequently, the impact of IR in terms fact reduction, such as streak or metal at the same dose level and equivalent of radiation dose reduction might be artifacts. However, these effects are not low-contrast resolution compared with partly overestimated. One should also necessarily related to the complexity of FBP acquired at lower doses. AIDR, keep in mind that there is neither an the IR algorithm. A practical consider- which does not provide the option official definition of terms such as low ation is that because of the possibly un- to select between different iterative dose or even ultra-low dose, which are familiar overall image impression, radi- strengths, provided noise reductions of frequently used in connection with IR ologists might initially feel inclined to 35%–44%. A more recent study by Mi- techniques, nor a consensus on the best reject the routine implementation of IR eville el al (47) compared ASIR, Veo, indicator (CT dose index, dose-length algorithms and question their diagnos- iDose4, and FBP and reported that the product, size-specific dose estimate, tic accuracy compared with traditional model-based IR product, Veo, resulted etc) of radiation dose (43). Moreover, FBP techniques. Commonly, however, in superior image quality compared the current literature on IR technol- such concerns will be dispelled after an with the other three techniques, par- ogy in CT imaging is mainly focused initial learning curve and increasing fa- ticularly with respect to spatial reso- on image quality and radiation dose. miliarity with the IR image impression. lution. The authors further describe However, investigations of the diagnos- improved low-contrast detectability tic performance of IR and the resulting Phantom Experiments with this algorithm, even at decreased clinical outcome are largely missing. Initial reports on the various IR prod- radiation dose levels. ucts were largely based on studies in In addition to more visually recog- Practical Considerations which phantom models were used. nizable methods, a valuable parameter From a practitioner’s point of view, Given the purported benefits of IR, for the evaluation of image quality with the integration of IR does not intro- studies generally compared some IR is the noise power spectrum (NPS). duce major workflow changes as com- combination of various image-quality The NPS graphically represents both pared with the use of FBP. After data parameters, including noise, signal- image noise (defined by the area under

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 345 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

the curve) and noise texture (reflected Figure 4 in the shape of the curve). Most IR algorithms affect NPS graphs similarly—the area under the curve (noise) is reduced while the peak of the curve shifts to- ward lower frequencies. This particular effect is specifically reported with the use of several IR techniques, including ASIR and Veo (47), IRIS (44), SAFIRE (45), and iDose4 (46). The perceived consequence is an unfamiliar visual appearance that is commonly described as plastic-like, paint-brushed, blurry, blotchy, or over-smoothed, which can be objectified by NPS analysis (45). IR- FBP blending and variable user-speci- fied iterative strength levels are two techniques the vendors have utilized in Figure 4: Coronal reformations of contrast-enhanced CT study of the head in a patient with subcutaneous their attempt to mitigate this unfamiliar abscess in the left cheek. Compared with, A, FBP, the visualization of the abscess formation is improved by noise and often undesired textural alteration. reduction with use of the, B, Veo IR algorithm. The surrounding abscess membrane is clearly depicted (arrow). Overall, phantom models have ver- ified the feasibility of IR techniques in reducing image noise while maintaining Figure 5 other image quality parameters, and these studies have led to an ever-ex- panding body of literature demonstrat- ing similar findings in vivo.

Head and Neck IR use in the head and neck has shown utility in decreasing dose, improving image quality, and mitigating artifacts (Fig 4). At least two studies demonstrated a reduction in radiation dose associated with cervical spine CT to a level comparable to that of conventional radiography in trauma patients (48,49). Radiation dose decreases in the brain ranging from 20% to 40% have been shown by using ASIR, iDose, IRIS, and SAFIRE (50–54). CT brain perfusion imaging, traditionally associated with relatively high radiation doses, may be well suited to exploit this benefit, with Figure 5: Transverse sections from a CT angiographic study of the carotid arteries in a patient with metallic initial studies showing dose reduc- dental hardware. Compared with, A, FBP, there is a reduction of metal artifacts with use of the, B, Veo IR algorithm. tions of 20% (55). Model-based IR has shown improved delineation of small arteries that are traditionally difficult or paranasal sinuses, often necessitate (63). Finally, model-based IR prod- to image, including arteries within thinner section reconstructions that ucts (such as Veo) or those containing the posterior fossa (56), the anterior are subject to increased noise second- specific metal artifact-reduction algo- spinal artery (57), and the artery of ary to quantum mottle; again, this has rithms can substantially reduce com- Adamkiewicz (58). SAFIRE utilization shown to be an area in which the noise- mon artifacts in head and neck imag- in cervical spine CT has resulted in reduction properties of IR techniques ing, such as photon starvation caused better depiction of the intervertebral may be put to good use (61,62). ASIR by the shoulders and streak artifact disks and ligaments (59,60). Fine bone has been shown to enhance the delin- caused by dental hardware (64,65) structures, such as the temporal bone eation between white and gray matter (Fig 5).

346 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Thorax Figure 6 Currently available IR products have consistently been shown to allow radiation dose reductions without compromising the diagnostic image quality of routine chest CT, with reported dose reductions ranging from 27% to 80% (28,66–73). One group described that Veo was able to depict pulmonary nodules despite radiation dose reductions to levels comparable to those of conventional chest radiographs (74). Studies evaluating IR in thin-section pulmonary CT examinations have shown similar results, with comparable or superior image quality of IR compared with traditional FBP reconstructions (73,75,76) (Fig 6). One study concluded that CT images reconstructed with IR result in better visual scores than conventional FBP reconstruction for the assessment of lung architecture, such as interlobu- lar septa, the centrilobular region, and small bronchi/bronchioles; IR was also superior at delineating pathologic findings such as reticulations, tiny nodules, altered attenuation patterns, and bron- chiectasis (77). Similar findings have been observed in a pediatric patient population with cystic fibrosis (78). While IR may allow greater con- sistency of emphysema quantification at low-dose CT (79), another study showed that quantitative measures of emphysema and air trapping are substantially influenced by IR algorithms (80). This highlights an important point regarding IR—anything that significantly alters image reconstruction has the potential to influence quantitative methods, potentially arriving at diverging results when compared with standards set by using FBP. Fortunately, Figure 6: Coronal reformations of nonenhanced chest CT study (effective dose, 1.1 mSv) reconstructed this seems not to be the case in one of with, A, C, FBP and, B, D, iDose. In the study using IR, the overall image quality is improved and beam-hard- the most typical applications of IR in ening artifacts at the level of the shoulders are reduced. (Image courtesy of D. Utsunomiya, MD, Faculty of thoracic CT, pulmonary nodule assess- Life Sciences, Kumamoto University, Kumamoto, Japan.) ment. Both phantom-based (81,82) and in vivo studies (29) have suggested that CT imaging with IR maintains diagnos- estimated radiation savings of 25%– lung nodule volumetry is robust and tic accuracy compared with FBP in the 40% (66,70) (Fig 7). While no stud- reproducible throughout a wide range identification and characterization of ies have compared the accuracy of IR of tube voltage and tube current-time ground glass opacities, part-solid nod- versus FBP CT pulmonary angiography product exposure settings. Further- ules, and solid nodules, while allowing in the detection of pulmonary emboli, more, IR utilization does not negatively a dose reduction of approximately 75% researchers have shown preservation affect the performance of lung nod- (84,85). of reader confidence, with no change ule computer-aided detection systems IR has shown considerable poten- in the rate of nondiagnostic studies (83). From a qualitative standpoint, tial in CT pulmonary angiography, with (86). Indeed, there may be a role for

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 347 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 7

Figure 7: Transverse sections from a CT pulmonary angiographic study displayed at the level of the pulmonary trunk. Compared with, A, FBP, the use of, B, ASIR results in noise reduction and slightly enhanced CT attenuation. A further enhancement of the visual image quality impression is achieved with, C, Veo. image quality optimization with use of at a dose reduction of 25% compared In addition to noise and dose re- IR in CT pulmonary angiographic ex- with FBP have been reported for ASIR duction, early evidence suggests that aminations independent of radiation (70), and equivalent diagnostic accu- IR products may have a role in reduc- dose, for example, in larger patients racy and image quality with dose re- ing beam-hardening and blooming ar- in whom high noise levels limit the ductions up to 72% have been dem- tifacts associated with coronary artery interpretability of small subsegmental onstrated (98). Clinical observational stents and heavily calcified vessels. pulmonary arteries. Researchers have studies after implementation of ASIR Reductions in measured stent volumes advocated the use of higher strength report 44%–54% reductions in effec- indicating less blooming artifacts and IR in these patients to reduce the like- tive dose with cardiac CT applications image noise have been reported along lihood of a noninterpretable examina- (87,89). Reduced-dose protocols using with improved in-stent visualization tion. Of note, while there is concern IRIS have shown improved image qual- (93,103–105), and the noise-reduction regarding the undesirable textural ity compared with routine acquisitions properties of IR may allow increased effects seen in images reconstructed using FBP, with dose savings up to 62% utilization of high-resolution (ultra- with high-strength IR, the effect on (93); likewise, SAFIRE demonstrated thin section, usually 0.23-mm spatial image quality has been reported to be improved image quality compared with resolution) coronary CT in-stent eval- more modest for vascular structures FBP with simulated dose reductions of uation. Traditionally, these examina- than for the pulmonary parenchyma 50%–80% (35,99). Radiation reductions are limited by high levels of noise itself (70,87). tions of 55%–63% have been reported secondary to photon starvation. Early without compromising image quality studies have shown improvements in Cardiac Imaging using both fixed- and adaptive-dose noise, blooming artifacts, in-stent vi- Like other applications, the major dem- protocols with iDOSE reconstructions sualization, and diagnostic accuracy onstrated benefit of IR in coronary CT (90,91,100). with use of IR in conjunction with angiography to date has been a reduc- Body mass index (BMI)-adaptive high-resolution reconstruction kernels tion in image noise without substan- reconstructions using predefined ac- and acquisitions (106–108). Likewise, tial effects on attenuation compared quisition settings based on patient Renker et al (109) compared IRIS to with FBP reconstructions of the same body habitus offer a potential solu- FBP in patients with Agatston scores data. This results in improvements in tion (92,98), while other groups have of 400 or greater and showed that subjective image quality and vessel proposed patient-specific adaptive- IRIS resulted in significantly lower im- assessment (22,35,87–96) (Fig 8). dose procedures that adjust scan set- age noise (P = .011–.035) and calci- These findings were reported for all tings on the basis of allowable image fication volume (P = .019 and .026), commercially available IR products ex- noise (70,101). In this regard, Yin et significantly higher subjective image cept for Veo, with which experiences al (102) recently demonstrated that quality (P = .031 and .042), and sig- are limited to date. For coronary CT in a population with a broad range of nificantly improved per-segment diag- angiographic studies, AIDR has shown BMI values, the use of IR after apply- nostic accuracy for detection of signifi- improved objective and subjective im- ing a 50% tube current reduction for cant stenoses (P = .0001), with overall age quality with simulated half-dose every selected kilovoltage preserves diagnostic accuracy of 95.9% for IRIS, acquisitions compared with full-dose image quality and diagnostic accuracy compared with 91.8% by using FBP. FBP reconstructions of the same data at coronary CT angiography compared Note that these reductions in calcium (97). Likewise, improved image quality with standard FBP. volume may be a relevant consider-

348 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 8 FBP reconstructions. In contrast, IR does not appear to significantly alter the analysis of plaque composition and plaque burden quantification (88,112,113).

Abdomen and Pelvis A number of advantages have been demonstrated with use of IR in abdominal and pelvic imaging. Princi- ple among them is noise reduction, which allows concomitant dose reduction. Routine abdominal and pelvic CT performed with commercially available IR software allows equivalent to improved subjective and objective image quality at dose reductions of 25%–50% compared with full-dose FBP (114–120). Newer IR algorithms have been shown to allow diagnostic quality acquisitions with dose-reduced protocols performed with only 50 mA, providing 75% dose reductions in selected patients compared with standard acquisition parameters (121,122) (Figs 9, 10). CT angiography.—Like IR in coronary CT angiography, body CT angiographic applications should allow significant dose reductions while maintaining diagnostic image quality (32,123), and model-based IR products may lead to improved accuracy when measuring vascular diameter and evaluating vessel wall attenuation (124). Liver CT.—IR application in liver CT has been validated in both phantom and in-vivo studies, which have shown that dose reductions between 41% and 50% are possible without sacrificing image quality (125,126). This may be particularly relevant in hepatic perfusion imaging, which has traditionally Figure 8: A, B, Three-dimensional volume-rendered reconstructions, C, D, transverse sections, and, E, F, involved relatively high radiation dose oblique maximum intensity projections of a coronary CT angiographic study (80 kVp, 250 mA). Images are values. A recent study showed that the reconstructed by using AIDR (A, C, E) and FBP (B, D, F). There is a reduction in image noise and improve- use of AIDR allowed a 45% dose re- ment in image quality with AIDR compared with FBP reconstructions. duction by applying a tube current of 120 mAs instead of the standard 250 mAs without affecting image quality or ation when performing nonenhanced blooming artifacts when ASIR was quantitative hepatic perfusion values coronary artery calcium scoring ex- used compared with FBP (110,111); (127). aminations, another time-honored the practical implication is that cal- While IR can increase CNR by re- quantitative CT imaging application. cium scoring using IR may result in ducing noise alone, there has also been One study showed decreased noise incorrect risk stratification, as the interest in improving small lesion con- but also decreased Agatston and volu- population-based studies that pro- spicuity by further enhancing CNR with metric calcium scores due to reduced vided calcium score nomograms used combined low-dose acquisition and IR

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 349 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 9

Figure 9: Coronal reformations of contrast-enhanced abdominal CT study in a patient with liver metastasis (arrowhead). Beam-hardening artifacts (arrow) due to metallic clips are seen. Images reconstructed with, A, FBP and SAFIRE strengths, B, 1, C, 3, and, D, 5. Increasing the IR strength reduces image noise and beam- hardening artifacts.

Figure 10

Figure 10: Coronal reformations of CT angiographic study of the abdominal vasculature (effective dose, 2.8 mSv) reconstructed with, A, FBP and, B, iDose4. The IR algorithm reduces image noise (B) and improves the image quality of three-dimensional volume-rendered reconstructions (C). (Image courtesy of C. Liang, MD, and Z. Liu, MD, Guangdong General Hospital, Guangzhou, China.) techniques (Fig 11). Lower kilovoltage in which significantly improved image vs 50 mAs) studies using ASIR have scans increase attenuation (contrast), quality of volume-rendered images has equivalent image quality with de- while IR mitigates the increased noise been demonstrated (132). creased noise compared with those associated with these low-dose scans; CT enterography.—Two studies on using FBP (135), and half-dose re- this approach has been shown to be patients with Crohn disease have shown constructions using SAFIRE produce effective in increasing CNR in low- that CT enterography with IR can lead equivalent image quality as full-dose dose arterial, portal, and late vascular to statistically significant dose reduc- reconstructions using FBP (136). Fur- phase image acquisitions (128). While tions between 35% and 50% without thermore, a porcine colon phantom several studies have demonstrated no loss of image quality or observer con- study comparing FBP, ASIR, and Veo significant improvement in the detec- fidence compared with FBP (133,134). reconstructions at various radiation tion of small low-contrast liver lesions CT colonography.—CT colonogra- dose acquisitions showed statistically (129,130), this strategy has been phy is becoming a widely recognized significant improvement in per-polyp shown to be effective in improving screening tool for colorectal cancer. detection sensitivity with use of ASIR the detection of hypervascular liver As for all screening application in a and significantly reduced image noise lesions such as hepatocellular carci- priori healthy populations, ionizing with both IR techniques (137) (Figs noma (131). IR-based improvements radiation exposure should be kept 12, 13). in subjective image quality and CNR to an absolute minimum. Studies CT urography.—Initial experiences also manifest in CT portovenography, have shown that reduced-dose (25 (138,139) have demonstrated that the

350 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 11

Figure 11: Transverse sections from contrast-enhanced abdominal CT study reconstructed with, A, FBP and SAFIRE strengths, B, 1, C, 3, and, D, 5. With increasing IR strength, a reduction in image noise is observed, allowing for better delineation of regional changes in hepatic perfusion (arrows). in-vivo applications of IR to CT urogra- medical and lay press (140,141). Pe- be well suited as an additional tool to phy allow a significant radiation dose re- diatric patients are thought to have provide diagnostic image quality with duction between 45% and 84% without increased radiation sensitivity of their the lowest doses possible (32,35). reducing image quality and without af- immature tissues, generally have a An increasing number of studies are fecting diagnostic confidence. Kulkarni longer life expectancy and therefore reporting that noise, CNR, signal- et al (139) demonstrated that even more time to develop stochastic ef- to-noise ratio, and subjective image when the dose was reduced from 9.9 fects, and often undergo repeat diag- quality are significantly improved with to 1.8 mGy, images were still deemed nostic testing. It is no surprise then the use of IR algorithms in pediatric diagnostically acceptable for reliable that there have been strong efforts in patients (46,143–146). In pediatric detection of urinary stones. Further- limiting ionizing radiation exposure in cardiac CT, Han et al (145) reported more, this extreme dose reduction did this population, as, for instance, prom- a significant noise reduction of 34%, not impair the evaluation of the rest of inently exemplified by the Image Gently and substantial increase of 41% and the abdomen and pelvis for noncalculus campaign (142). A number of CT tech- 56% in CNR and signal-to-noise ratio, findings. niques have been developed and are respectively, with the use of SAFIRE emerging, including high-pitch, low- compared with FBP. Depending on the IR in Pediatric Imaging tube-voltage protocols; patient-specific body region, the potential dose reduc- The potential stochastic effects of ion- protocol optimization; increased and tion provided by the implementation izing radiation in pediatric patients more effective use of shielding de- of IR techniques ranges between 22% have raised considerable concern over vices; and an overall emphasis on staff and 48% without impairing diagnostic the past several years within both the training and attention. IR appears to confidence (51,147–149).

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 351 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

Figure 12

Figure 12: Transverse sections of CT colonography study (80 mAs, 100 kVp) reconstructed with, A, 40% ASIR and, B, FBP. Application of the ASIR algorithm (A) to low-dose CT colonography reduces image noise and improves overall image quality compared with FBP (B).

Figure 13 reduction using iDose compared with FBP in obese patients (BMI  30 kg/ m2) undergoing CT pulmonary angiography. Moreover, the implementation of iDose level 5 provided noise values that were comparable to those in non- obese control subjects (average BMI, 22 kg/m2). In coronary CT angiography, Wang et al (155) showed a potential dose reduction of 50% facilitated by the use of SAFIRE in obese patients without sacrificing image quality. This is consistent with the results of other studies that have shown radiation dose reductions of 32% and 50% in abdominal-pelvic CT imaging (157) and coronary CT angiography, respectively (158).

IR in Emergency Radiology Figure 13: Three-dimensional surface-rendered endoluminal displays of CT colonography study (80 mAs, IR algorithms have also been deemed 100 kVp) reconstructed with, A, 40% ASIR and, B, FBP. Note that image noise causes a mildly speckled appearance of the colon wall on FBP image (B), which is reduced on ASIR image (A). effective in emergency conditions. They are able to reduce total radiation dose without any loss in image quality in IR in Obese Patients reconstruction techniques using dual- applications that included acute aor- CT examinations in the obese popula- source CT (152,153). Unfortunately, tic syndrome (20% decrease in dose tion are challenging. Noise is intrin- both techniques result in significantly length product using ASIR compared sically higher secondary to reduced increased radiation dose (P < .01). The with FBP) (123) and trauma surveys of photon transmission and scatter, com- noise-reduction properties of IR may the brain, cervical spine, chest, abdo- promising both image quality and di- hold particular appeal in the evaluation men, and pelvis (20% dose decrease agnostic accuracy (150,151). Several of obese individuals, with IR applica- using ASIR compared with FBP) (49). techniques can improve image quality tions showing reductions in both image Importantly, one study also demon- in obese patients, such as BMI-adaptive noise and radiation dose in this popu- strated that IR implementation (in this scan protocols (generally using higher lation (35,154,155). Kligerman et al study, iDose4) does not significantly de- kilovoltage acquisitions) and half-scan (156) demonstrated a significant noise lay reconstruction time or speed (159).

352 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

7. Hu H. Multi-slice helical CT: scan and recon- Summary pears safe to predict that in the fore- seeable future IR will replace traditional struction. Med Phys 1999;26(1):5–18. Increases in available and affordable analytical methods as the preferred CT 8. Zhang D, Li X, Liu B. Objective character- computer power have fostered the de- image reconstruction method. ization of GE discovery CT750 HD scanner: gemstone spectral imaging mode. Med Phys velopment of a variety of IR algorithms Disclosures of Conflicts of Interest: L.L.G. Ac- 2011;38(3):1178–1188. and their application to diagnostic CT tivities related to the present article: disclosed 9. Hsiao EM, Rybicki FJ, Steigner M. CT coro- imaging. While the specific algorithms no relevant relationships. Activities not related nary angiography: 256-slice and 320-detector differ, the clinical basis for the benefits to the present article: received payment for row scanners. Curr Cardiol Rep 2010;12(1): of IR implementation primarily involves lectures including service on speakers bureaus 68–75. (year 2012) from GE Healthcare (Germany). image noise reduction, which leads to Other relationships: disclosed no relevant rela- 10. Rybicki FJ, Otero HJ, Steigner ML, et al. Initial improved objective and subjective im- tionships. U.J.S. Activities related to the pre- evaluation of coronary images from 320-detec- sent article: disclosed no relevant relationships. tor row computed tomography. Int J Cardiovasc age quality compared with those using Imaging 2008;24(5):535–546. FBP reconstructions. Decreased noise Activities not related to the present article: reports grants from Bayer, Bracco, GE, Me- alone results in improved image quality 11. Flohr TG, McCollough CH, Bruder H, et al. drad, Siemens, personal fees from Bayer, GE, First performance evaluation of a dual-source in previously challenging areas, such as Siemens, nonfinancial support from Bayer, Me- CT (DSCT) system. Eur Radiol 2006;16(2): ultra-high-resolution imaging and the drad, Siemens. Other relationships: disclosed 256–268. evaluation of obese individuals. Perhaps no relevant relationships. F.G.M. disclosed no relevant relationships. J.W.N. disclosed no rel- 12. Brooks RA, Di Chiro G. Principles of com- more important, the noise-reduction evant relationships. G.B. Activities related to puter assisted tomography (CAT) in radio- properties of IR techniques hold poten- the present article: disclosed no relevant rela- graphic and radioisotopic imaging. Phys Med Biol 1976;21(5):689–732. tial to enable designing CT acquisition tionships. Activities not related to the present protocols at reduced radiation dose article: received speaker honorarium from Bayer 13. Andersen AH, Kak AC. Simultaneous alge- and Siemens Healthcare. Other relationships: braic reconstruction technique (SART): a su- levels without sacrificing image qual- disclosed no relevant relationships. J.A.L. Ac- perior implementation of the art algorithm. ity, which is particularly attractive in tivities related to the present article: received Ultrason Imaging 1984;6(1):81–94. screening examinations (eg, lung and fees from GE Healthcare. Activities not related to the present article: received payment for ser- 14. Gordon R, Bender R, Herman GT. Algebraic colorectal cancer), perfusion studies, reconstruction techniques (ART) for three- vice on speakers bureau from GE Healthcare. dimensional electron microscopy and x-ray pediatric imaging, and for repeat ex- Other relationships: disclosed no relevant rela- photography. J Theor Biol 1970;29(3):471–481. aminations. Besides enhancements in tionships. N.S.P. Activities related to the pre- general measures of image quality, an sent article: disclosed no relevant relationships. 15. Gilbert P. Iterative methods for the three- increasing body of evidence describes Activities not related to the present article: in- dimensional reconstruction of an object from stitution received research grant from Toshiba projections. J Theor Biol 1972;36(1):105–117. improvements in the diagnostic accu- Medical Systems, author received payment for 16. Hsieh J. Adaptive statistical iterative recon- racy of various CT imaging applications, service on speakers bureau from Toshiba Med- struction: GE white paper. Waukesha, Wis: ical Systems. Other relationships: research col- for example, via artifact reduction, by GE Healthcare, 2008. use of IR techniques. However, while laboration Carestream Healt. M.R. disclosed no relevant relationships. A.L. disclosed no relevant 17. Cheng L, Chen Y, Fang T, Tyan J. Fast itera- an ever greater number of scientific re- relationships. C.N.D.C. disclosed no relevant re- tive adaptive reconstruction in low-dose CT ports highlight the substantial promise lationships. imaging. In: 2006 IEEE International Confer- of IR techniques to enhance the diag- ence on Image Processing, 2006; 889–892. nostic performance and to incur drastic 18. Hara AK, Paden RG, Silva AC, Kujak JL, reductions in radiation requirements References Lawder HJ, Pavlicek W. Iterative reconstruction technique for reducing body radiation at CT, major evidence confirming such 1. Ambrose J. Computerized transverse axial dose at CT: feasibility study. AJR Am J Roent- scanning (tomography). II. Clinical applica- initial findings in larger patient popula- genol 2009;193(3):764–771. tions is still mostly lacking, along with tion. Br J Radiol 1973;46(552):1023–1047. 19. Liu YJ, Zhu PP, Chen B, et al. A new iterative 2. Hounsfield GN. Computerized transverse ax- professional society guidelines on the algorithm to reconstruct the refractive index. ial scanning (tomography). I. Description of Phys Med Biol 2007;52(12):L5–L13. appropriate routine implementation of system. Br J Radiol 1973;46(552):1016–1022. IR techniques. However, we will likely 20. Nuyts J, De Man B, Dupont P, Defrise M, 3. Boyd DP, Lipton MJ. Cardiac computed to- Suetens P, Mortelmans L. Iterative recon- observe a gradual development of a mography. Proc IEEE 1983;71(3):298–307. more meaningful evidence base and struction for helical CT: a simulation study. 4. Kalender WA, Seissler W, Klotz E, Vock P. Phys Med Biol 1998;43(4):729–737. of pertinent guidelines with the more Spiral volumetric CT with single-breath-hold 21. Paden RG, Pavlicek W, Peter MB, Boltz TF II, widespread availability of IR algorithms technique, continuous transport, and continu- Park K, Langan DA. Adaptive statistical itera- ous scanner rotation. Radiology 1990;176(1): and increasing familiarity of the medi- tive reconstruction (ASIR) for CT dose reduc- 181–183. cal imaging community with their use. tion of head and body examinations: a phan- Refinement of current IR techniques 5. Klingenbeck-Regn K, Schaller S, Flohr T, tom study [abstr]. In: Radiological Society of will likely result in more effective and Ohnesorge B, Kopp AF, Baum U. Subsecond North America Scientific Assembly and Annual multi-slice computed tomography: basics and Meeting Program. Oak Brook, Ill: Radiological universal clinical adoption, and emerg- applications. Eur J Radiol 1999;31(2):110–124. Society of North America, 2008, 455. ing and future algorithms promise to 6. McCollough CH, Zink FE. Performance eval- 22. Leipsic J, Labounty TM, Heilbron B, et al. expand the role of IR to novel diagnos- uation of a multi-slice CT system. Med Phys Adaptive statistical iterative reconstruction: tic imaging applications. Overall, it ap- 1999;26(11):2223–2230. assessment of image noise and image qual-

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 353 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

ity in coronary CT angiography. AJR Am J traditional filtered back projection. Eur Radiol physical metrics and 4-alternative forced- Roentgenol 2010;195(3):649–654. 2011;21(10):2130–2138. choice detectability experiments—a phantom approach. Phys Med 2013;29(1):99–110. 23. Mueck FG, Körner M, Scherr MK, et al. Up- 36. Gordic S, Morsbach F, Schmidt B, et al. Ul- grade to iterative image reconstruction (IR) in tralow-dose chest computed tomography for 48. Geyer LL, Körner M, Hempel R, et al. Eval- abdominal MDCT imaging: a clinical study for pulmonary nodule detection: first performance uation of a dedicated MDCT protocol using detailed parameter optimization beyond ven- evaluation of single energy scanning with iterative image reconstruction after cervical dor recommendations using the adaptive sta- spectral shaping. Invest Radiol 2014;49(7): spine trauma. Clin Radiol 2013;68(7):e391– tistical iterative reconstruction environment 465–473. e396. (ASIR). Rofo 2012;184(3):229–238. 37. Gervaise A, Osemont B, Lecocq S, et al. CT 49. Maxfield MW, Schuster KM, McGillicuddy 24. Thibault J-B. Veo: CT model-based iterative image quality improvement using adaptive it- EA, et al. Impact of adaptive statistical itera- reconstruction. http://www.gehealthcare.com. erative dose reduction with wide-volume active reconstruction on radiation dose in eval- Published 2010. Accessed September 2013. quisition on 320-detector CT. Eur Radiol 2012; uation of trauma patients. J Trauma Acute 22(2):295–301. Care Surg 2012;73(6):1406–1411. 25. Yu Z, Thibault JB, Bouman CA, Sauer KD, Hsieh J. Fast model-based X-ray CT reconstruction 38. Tatsugami F, Matsuki M, Nakai G, et al. The 50. Kilic K, Erbas G, Guryildirim M, Arac M, Ilgit using spatially nonhomogeneous ICD optimi- effect of adaptive iterative dose reduction on E, Coskun B. Lowering the dose in head CT zation. IEEE Trans Image Process 2011;20(1): image quality in 320-detector row CT coro- using adaptive statistical iterative reconstruc- 161–175. nary angiography. Br J Radiol 2012;85(1016): tion. AJNR Am J Neuroradiol 2011;32(9): e378–e382. 1578–1582. 26. Thibault JB, Sauer KD, Bouman CA, Hsieh J. A three-dimensional statistical approach to im- 39. Irwan R, Nakanishi S, Blum A. AIDR 3D: 51. Kilic K, Erbas G, Guryildirim M, et al. proved image quality for multislice helical CT. reduces dose and simultaneously improves Quantitative and qualitative comparison of Med Phys 2007;34(11):4526–4544. image quality. https://www.toshiba-medical. standard-dose and low-dose pediatric head eu/eu/wp-content/uploads/sites/2/2014/10/ computed tomography: a retrospective study 27. Nelson RC, Feuerlein S, Boll DT. New iterative AIDR-3D-white-paper1.pdf. Published 2011. assessing the effect of adaptive statistical iter- reconstruction techniques for cardiovascular Accessed September 2013. ative reconstruction. J Comput Assist Tomogr computed tomography: how do they work, and 2013;37(3):377–381. what are the advantages and disadvantages? 40. Yang Z, Zamyatin AA, Akino N. Effective J Cardiovasc Comput Tomogr 2011;5(5): data-domain noise and streak reduction for 52. Korn A, Bender B, Fenchel M, et al. Sinogram 286–292. x-ray CT. In: Kachelriess M, Rafecas M, eds. affirmed iterative reconstruction in head CT: The 11th International Meeting on Fully 3D improvement of objective and subjective im- 28. Laqmani A, Buhk JH, Henes FO, et al. Im- Image Reconstruction in Radiology and Nu- age quality with concomitant radiation dose pact of a 4th generation iterative reconstruc- clear Medicine. Potsdam, Germany, 2011; reduction. Eur J Radiol 2013;82(9):1431–1435. tion technique on image quality in low-dose 290–293. http://www.fully3d.org/2011/Fully- 53. Korn A, Fenchel M, Bender B, et al. Iterative computed tomography of the chest in immu- 3D2011Proceedings.pdf. nocompromised patients. Rofo 2013;185(8): reconstruction in head CT: image quality of 749–757. 41. Yang Z, Silver MD, Yasuhiro N. Adpative routine and low-dose protocols in comparison weight anisotropic diffusion for computed to- with standard filtered back-projection. AJNR 29. Willemink MJ, Borstlap J, Takx RA, et al. The mography denoising. In: Kachelriess M, Rafe- Am J Neuroradiol 2012;33(2):218–224. effects of computed tomography with itera- cas M, eds. The 11th International Meeting on 54. Wu TH, Hung SC, Sun JY, et al. How far can tive reconstruction on solid pulmonary nodule Fully 3D Image Reconstruction in Radiology the radiation dose be lowered in head CT with volume quantification. PLoS ONE 2013;8(2): and Nuclear Medicine. Potsdam, Germany, iterative reconstruction? analysis of imaging e58053. 2011; 210–213. http://www.fully3d.org/2011/ quality and diagnostic accuracy. Eur Radiol Fully3D2011Proceedings.pdf. 30. Mehta D, Thompson R, Morton T, Dhanant- 2013;23(9):2612–2621. wari A, Shefer E. Iterative model reconstruc- 42. Tack D, Jahnen A, Kohler S, et al. Multidetec- 55. Lin CJ, Wu TH, Lin CH, et al. Can iterative tion: simultaneously lowered computed to- tor CT radiation dose optimisation in adults: reconstruction improve imaging quality for mography radiation dose and improved image short- and long-term effects of a clinical audit. lower radiation CT perfusion? initial experi- quality. Med Phys Int 2013;1(2):147–155. Eur Radiol 2014;24(1):169–175. ence. AJNR Am J Neuroradiol 2013;34(8): 31. Grant K, Flohr T. Iterative reconstruction in 43. Bankier AA, Kressel HY. Through the Look- 1516–1521. image space (IRIS). http://www.usa.siemens. ing Glass revisited: the need for more 56. Machida H, Takeuchi H, Tanaka I, et al. Im- com/healthcare. Published 2010. Accessed meaning and less drama in the reporting of proved delineation of arteries in the posterior September 2013. dose and dose reduction in CT. Radiology fossa of the brain by model-based iterative re- 2012;265(1):4–8. 32. Winklehner A, Karlo C, Puippe G, et al. Raw construction in volume-rendered 3D CT angi- data-based iterative reconstruction in body 44. Ghetti C, Ortenzia O, Serreli G. CT iterative ography. AJNR Am J Neuroradiol 2013;34(5): CTA: evaluation of radiation dose saving po- reconstruction in image space: a phantom 971–975. tential. Eur Radiol 2011;21(12):2521–2526. study. Phys Med 2012;28(2):161–165. 57. Machida H, Tanaka I, Fukui R, et al. Improved 33. Baumueller S, Winklehner A, Karlo C, et al. 45. Ghetti C, Palleri F, Serreli G, Ortenzia O, delineation of the anterior spinal artery with Low-dose CT of the lung: potential value of iter- Ruffini L. Physical characterization of a new model-based iterative reconstruction in CT ative reconstructions. Eur Radiol 2012;22(12): CT iterative reconstruction method operat- angiography: a clinical pilot study. AJR Am J 2597–2606. ing in sinogram space. J Appl Clin Med Phys Roentgenol 2013;200(2):442–446. 2013;14(4):4347. 34. Grant K, Raupach R. SAFIRE: Sinogram af- 58. Nishida J, Kitagawa K, Nagata M, Yamazaki firmed iterative reconstruction. http://www. 46. Noël PB, Fingerle AA, Renger B, Münzel D, A, Nagasawa N, Sakuma H. Model-based it- usa.siemens.com/healthcare. Published 2012. Rummeny EJ, Dobritz M. Initial performance erative reconstruction for multi-detector row Accessed September 2013. characterization of a clinical noise-suppress- CT assessment of the Adamkiewicz artery. ing reconstruction algorithm for MDCT. AJR Radiology 2014;270(1):282–291. 35. Moscariello A, Takx RA, Schoepf UJ, et al. Am J Roentgenol 2011;197(6):1404–1409. Coronary CT angiography: image quality, di- 59. Becce F, Ben Salah Y, Verdun FR, et al. Com- agnostic accuracy, and potential for radiation 47. Miéville FA, Gudinchet F, Brunelle F, Bo- puted tomography of the cervical spine: com- dose reduction using a novel iterative image chud FO, Verdun FR. Iterative reconstruction parison of image quality between a standard- reconstruction technique-comparison with methods in two different MDCT scanners: dose and a low-dose protocol using filtered

354 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

back-projection and iterative reconstruction. 72. Yamada Y, Jinzaki M, Hosokawa T, et al. Dose 84. Higuchi K, Nagao M, Matsuo Y, et al. Detec- Skeletal Radiol 2013;42(7):937–945. reduction in chest CT: comparison of the tion of ground-glass opacities by use of hybrid adaptive iterative dose reduction 3D, adap- iterative reconstruction (iDose) and low-dose 60. Omoumi P, Verdun FR, Ben Salah Y, et al. tive iterative dose reduction, and filtered back 256-section computed tomography: a phan- Low-dose multidetector computed tomog- projection reconstruction techniques. Eur J Ra- tom study. Radiol Phys Technol 2013;6(2): raphy of the cervical spine: optimization of diol 2012;81(12):4185–4195. 299–304. iterative reconstruction strength levels. Acta Radiol 2014;55(3):335–344. 73. Ichikawa Y, Kitagawa K, Nagasawa N, Mu- 85. Katsura M, Matsuda I, Akahane M, et al. rashima S, Sakuma H. CT of the chest with Model-based iterative reconstruction tech- 61. Schulz B, Beeres M, Bodelle B, et al. Perfor- model-based, fully iterative reconstruction: nique for ultralow-dose chest CT: comparison mance of iterative image reconstruction in CT comparison with adaptive statistical iterative of pulmonary nodule detectability with the of the paranasal sinuses: a phantom study. reconstruction. BMC Med Imaging 2013;13(1): adaptive statistical iterative reconstruction AJNR Am J Neuroradiol 2013;34(5):1072– 27. technique. Invest Radiol 2013;48(4):206–212. 1076. 74. Neroladaki A, Botsikas D, Boudabbous S, 86. Yuan R, Shuman WP, Earls JP, et al. Reduced 62. Niu YT, Mehta D, Zhang ZR, et al. Radiation Becker CD, Montet X. Computed tomography iodine load at CT pulmonary angiography with dose reduction in temporal bone CT with it- of the chest with model-based iterative recon- dual-energy monochromatic imaging: compar- erative reconstruction technique. AJNR Am J struction using a radiation exposure similar to ison with standard CT pulmonary angiogra- Neuroradiol 2012;33(6):1020–1026. chest x-ray examination: preliminary observa- phy—a prospective randomized trial. Radiology 63. Rapalino O, Kamalian S, Kamalian S, et al. tions. Eur Radiol 2013;23(2):360–366. 2012;262(1):290–297. Cranial CT with adaptive statistical iterative 75. Wang H, Tan B, Zhao B, Liang C, Xu Z. Raw- 87. Leipsic J, Labounty TM, Heilbron B, et al. reconstruction: improved image quality with data-based iterative reconstruction versus Estimated radiation dose reduction using concomitant radiation dose reduction. AJNR filtered back projection: image quality of low- adaptive statistical iterative reconstruction in Am J Neuroradiol 2012;33(4):609–615. dose chest computed tomography examina- coronary CT angiography: the ERASIR study. 64. Katsura M, Sato J, Akahane M, et al. Com- tions in 87 patients. Clin Imaging 2013;37(6): AJR Am J Roentgenol 2010;195(3):655–660. parison of pure and hybrid iterative re- 1024–1032. construction techniques with conventional 88. Fuchs TA, Fiechter M, Gebhard C, et al. CT filtered back projection: image quality assess- 76. Botelho MP, Agrawal R, Gonzalez-Guindalini coronary angiography: impact of adapted ment in the cervicothoracic region. Eur J Ra- FD, et al. Effect of radiation dose and itera- statistical iterative reconstruction (ASIR) on diol 2013;82(2):356–360. tive reconstruction on lung lesion conspicuity coronary stenosis and plaque composition at MDCT: does one size fit all? Eur J Radiol analysis. Int J Cardiovasc Imaging 2013;29(3): 65. Morsbach F, Wurnig M, Kunz DM, et al. Metal 2013;82(11):e726–e733. 719–724. artefact reduction from dental hardware in carotid CT angiography using iterative re- 77. Prakash P, Kalra MK, Ackman JB, et al. Dif- 89. Gosling O, Loader R, Venables P, et al. A constructions. Eur Radiol 2013;23(10):2687– fuse lung disease: CT of the chest with adap- comparison of radiation doses between state- 2694. tive statistical iterative reconstruction tech- of-the-art multislice CT coronary angiography nique. Radiology 2010;256(1):261–269. with iterative reconstruction, multislice CT 66. Pontana F, Pagniez J, Flohr T, et al. Chest coronary angiography with standard filtered computed tomography using iterative recon- 78. Miéville FA, Berteloot L, Grandjean A, et al. back-projection and invasive diagnostic coro- struction vs filtered back projection. I. Evalu- Model-based iterative reconstruction in pedi- nary angiography. Heart 2010;96(12):922–926. ation of image noise reduction in 32 patients. atric chest CT: assessment of image quality Eur Radiol 2011;21(3):627–635. in a prospective study of children with cystic 90. Hou Y, Liu X, Xv S, Guo W, Guo Q. Com- fibrosis. Pediatr Radiol 2013;43(5):558–567. parisons of image quality and radiation dose 67. Singh S, Kalra MK, Gilman MD, et al. Adap- between iterative reconstruction and filtered tive statistical iterative reconstruction tech- 79. Nishio M, Matsumoto S, Ohno Y, et al. back projection reconstruction algorithms in nique for radiation dose reduction in chest Emphysema quantification by low-dose CT: 256-MDCT coronary angiography. AJR Am J CT: a pilot study. Radiology 2011;259(2):565– potential impact of adaptive iterative dose Roentgenol 2012;199(3):588–594. 573. reduction using 3D processing. AJR Am J Roentgenol 2012;199(3):595–601. 91. Hou Y, Xu S, Guo W, Vembar M, Guo Q. The 68. Hwang HJ, Seo JB, Lee HJ, et al. Low-dose optimal dose reduction level using iterative re- chest computed tomography with sinogram- 80. Mets OM, de Jong PA, van Ginneken B, Gi- construction with prospective ECG-triggered affirmed iterative reconstruction, iterative etema HA, Lammers JW. Quantitative com- coronary CTA using 256-slice MDCT. Eur J Ra- reconstruction in image space, and filtered puted tomography in COPD: possibilities and diol 2012;81(12):3905–3911. back projection: studies on image quality. J limitations. Lung 2012;190(2):133–145. Comput Assist Tomogr 2013;37(4):610–617. 92. Park EA, Lee W, Kim KW, et al. Iterative recon- 81. Willemink MJ, Leiner T, Budde RP, et al. Sys- struction of dual-source coronary CT angiog- 69. Kalra MK, Woisetschläger M, Dahlström N, tematic error in lung nodule volumetry: effect raphy: assessment of image quality and radia- et al. Sinogram-affirmed iterative reconstruc- of iterative reconstruction versus filtered back tion dose. Int J Cardiovasc Imaging 2012;28(7): tion of low-dose chest CT: effect on image projection at different CT parameters. AJR 1775–1786. quality and radiation dose. AJR Am J Roent- Am J Roentgenol 2012;199(6):1241–1246. genol 2013;201(2):W235–W244. 93. Renker M, Ramachandra A, Schoepf UJ, et al. 82. Wielpütz MO, Lederlin M, Wroblewski J, et Iterative image reconstruction techniques: ap- 70. Leipsic J, Nguyen G, Brown J, Sin D, Mayo al. CT volumetry of artificial pulmonary nod- plications for cardiac CT. J Cardiovasc Comput JR. A prospective evaluation of dose reduc- ules using an ex vivo lung phantom: influence Tomogr 2011;5(4):225–230. tion and image quality in chest CT using adap- of exposure parameters and iterative recon- tive statistical iterative reconstruction. AJR struction on reproducibility. Eur J Radiol 2013; 94. Schuhbaeck A, Achenbach S, Layritz C, et al. Am J Roentgenol 2010;195(5):1095–1099. 82(9):1577–1583. Image quality of ultra-low radiation exposure coronary CT angiography with an effective 71. Vardhanabhuti V, Loader RJ, Mitchell GR, 83. Yanagawa M, Honda O, Kikuyama A, et al. dose ,0.1 mSv using high-pitch spiral ac- Riordan RD, Roobottom CA. Image quality as- Pulmonary nodules: effect of adaptive statis- quisition and raw data-based iterative recon- sessment of standard- and low-dose chest CT tical iterative reconstruction (ASIR) technique struction. Eur Radiol 2013;23(3):597–606. using filtered back projection, adaptive statis- on performance of a computer-aided detection tical iterative reconstruction, and novel mod- (CAD) system-comparison of performance be- 95. Yoo RE, Park EA, Lee W, et al. Image qual- el-based iterative reconstruction algorithms. tween different-dose CT scans. Eur J Radiol ity of adaptive iterative dose reduction 3D AJR Am J Roentgenol 2013;200(3):545–552. 2012;81(10):2877–2886. of coronary CT angiography of 640-slice CT:

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 355 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

comparison with filtered back-projection. Int computed tomography for evaluation of coro- 119. Singh S, Kalra MK, Hsieh J, et al. Abdominal J Cardiovasc Imaging 2013;29(3):669–676. nary artery stents: comparison to standard- CT: comparison of adaptive statistical itera- definition 64-detector row computed tomogra- tive and filtered back projection reconstruc- 96. Yin WH, Lu B, Hou ZH, et al. Detection of phy. J Cardiovasc Comput Tomogr 2009;3(4): tion techniques. Radiology 2010;257(2):373– coronary artery stenosis with sub-milliSievert 246–251. 383. radiation dose by prospectively ECG-triggered high-pitch spiral CT angiography and iterative 108. Oda S, Utsunomiya D, Funama Y, et al. Im- 120. Fletcher JG, Krueger WR, Hough DM, et al. reconstruction. Eur Radiol 2013;23(11):2927– proved coronary in-stent visualization using Pilot study of detection, radiologist confidence 2933. a combined high-resolution kernel and a and image quality with sinogram-affirmed iter- hybrid iterative reconstruction technique at ative reconstruction at half-routine dose level. 97. Chen MY, Steigner ML, Leung SW, et al. Sim- 256-slice cardiac CT-Pilot study. Eur J Radiol J Comput Assist Tomogr 2013;37(2):203–211. ulated 50 % radiation dose reduction in coro- 2013;82(2):288–295. nary CT angiography using adaptive iterative 121. Kalra MK, Woisetschläger M, Dahlström N, dose reduction in three-dimensions (AIDR3D). 109. Renker M, Nance JW Jr, Schoepf UJ, et al. et al. Radiation dose reduction with sinogram Int J Cardiovasc Imaging 2013;29(5): Evaluation of heavily calcified vessels with affirmed iterative reconstruction technique for 1167–1175. coronary CT angiography: comparison of it- abdominal computed tomography. J Comput erative and filtered back projection image Assist Tomogr 2012;36(3):339–346. 98. Pontone G, Andreini D, Bartorelli AL, et al. reconstruction. Radiology 2011;260(2):390– Feasibility and diagnostic accuracy of a low 399. 122. Singh S, Kalra MK, Do S, et al. Comparison radiation exposure protocol for prospective of hybrid and pure iterative reconstruction ECG-triggering coronary MDCT angiography. 110. Gebhard C, Fiechter M, Fuchs TA, et al. Coro- techniques with conventional filtered back Clin Radiol 2012;67(3):207–215. nary artery calcium scoring: influence of adap- projection: dose reduction potential in the ab- tive statistical iterative reconstruction using domen. J Comput Assist Tomogr 2012;36(3): 99. Takx RA, Schoepf UJ, Moscariello A, et al. 64-MDCT. Int J Cardiol 2013;167(6):2932– 347–353. Coronary CT angiography: comparison of a 2937. novel iterative reconstruction with filtered 123. Cornfeld D, Israel G, Detroy E, Bokhari J, back projection for reconstruction of low-dose 111. Kurata A, Dharampal A, Dedic A, et al. Mojibian H. Impact of adaptive statistical iter- CT-Initial experience. Eur J Radiol 2013;82(2): Impact of iterative reconstruction on CT ative reconstruction (ASIR) on radiation dose 275–280. coronary calcium quantification. Eur Radiol and image quality in aortic dissection studies: a qualitative and quantitative analysis. AJR 100. Hosch W, Stiller W, Mueller D, et al. Reduc- 2013;23(12):3246–3252. Am J Roentgenol 2011;196(3):W336–W340. tion of radiation exposure and improvement 112. Stolzmann P, Schlett CL, Maurovich-Horvat P, of image quality with BMI-adapted prospec- et al. Variability and accuracy of coronary CT 124. Suzuki S, Machida H, Tanaka I, Ueno E. Mea- tive cardiac computed tomography and itera- angiography including use of iterative recon- surement of vascular wall attenuation: com- tive reconstruction. Eur J Radiol 2012;81(11): struction algorithms for plaque burden assess- parison of CT angiography using model-based 3568–3576. ment as compared with intravascular ultra- iterative reconstruction with standard filtered 101. Shen J, Du X, Guo D, et al. Noise-based tube sound-an ex vivo study. Eur Radiol 2012;22(10): back-projection algorithm CT in vitro. Eur J current reduction method with iterative re- 2067–2075. Radiol 2012;81(11):3348–3353. construction for reduction of radiation expo- 113. Takx RA, Willemink MJ, Nathoe HM, et al. 125. Husarik DB, Marin D, Samei E, et al. Radi- sure in coronary CT angiography. Eur J Radiol The effect of iterative reconstruction on quan- ation dose reduction in abdominal computed 2013;82(2):349–355. titative computed tomography assessment of tomography during the late hepatic arterial 102. Yin WH, Lu B, Li N, et al. Iterative reconstruc- coronary plaque composition. Int J Cardio- phase using a model-based iterative recon- tion to preserve image quality and diagnostic vasc Imaging 2014;30(1):155–163. struction algorithm: how low can we go? In- vest Radiol 2012;47(8):468–474. accuracy at reduced radiation dose in coronary 114. Deák Z, Grimm JM, Treitl M, et al. Filtered CT angiography: an intraindividual compar- back projection, adaptive statistical iterative 126. Mitsumori LM, Shuman WP, Busey JM, ison. JACC Cardiovasc Imaging 2013;6(12): reconstruction, and a model-based iterative re- Kolokythas O, Koprowicz KM. Adaptive sta- 1239–1249. construction in abdominal CT: an experimental tistical iterative reconstruction versus filtered 103. Ebersberger U, Tricarico F, Schoepf UJ, et al. clinical study. Radiology 2013;266(1):197–206. back projection in the same patient: 64 chan- CT evaluation of coronary artery stents with nel liver CT image quality and patient radia- 115. Desai GS, Thabet A, Elias AY, Sahani DV. iterative image reconstruction: improvements tion dose. Eur Radiol 2012;22(1):138–143. Comparative assessment of three image re- in image quality and potential for radiation construction techniques for image quality and 127. Negi N, Yoshikawa T, Ohno Y, et al. Hepatic CT dose reduction. Eur Radiol 2013;23(1):125– radiation dose in patients undergoing abdomi- perfusion measurements: a feasibility study 132. nopelvic multidetector CT examinations. Br J for radiation dose reduction using new image 104. Eisentopf J, Achenbach S, Ulzheimer S, et al. Radiol 2013;86(1021):20120161. reconstruction method. Eur J Radiol 2012; Low-dose dual-source CT angiography with 81(11):3048–3054. 116. Karpitschka M, Augart D, Becker HC, Reiser iterative reconstruction for coronary artery M, Graser A. Dose reduction in oncological 128. Marin D, Nelson RC, Schindera ST, et al. Low- stent evaluation. JACC Cardiovasc Imaging staging multidetector CT: effect of iterative tube-voltage, high-tube-current multidetector 2013;6(4):458–465. reconstruction. Br J Radiol 2013;86(1021): abdominal CT: improved image quality and de- 105. Gebhard C, Fiechter M, Fuchs TA, et al. 20120224. creased radiation dose with adaptive statisti- Coronary artery stents: influence of adaptive cal iterative reconstruction algorithm—initial 117. May MS, Wüst W, Brand M, et al. Dose re- statistical iterative reconstruction on image clinical experience. Radiology 2010;254(1): duction in abdominal computed tomography: quality using 64-HDCT. Eur Heart J Cardio- 145–153. intraindividual comparison of image quality of vasc Imaging 2013;14(10):969–977. full-dose standard and half-dose iterative re- 129. Dobeli KL, Lewis SJ, Meikle SR, Thiele DL, 106. Funama Y, Oda S, Utsunomiya D, et al. Coro- constructions with dual-source computed to- Brennan PC. Noise-reducing algorithms do not nary artery stent evaluation by combining mography. Invest Radiol 2011;46(7):465–470. necessarily provide superior dose optimisation iterative reconstruction and high-resolution for hepatic lesion detection with multidetector 118. Prakash P, Kalra MK, Kambadakone AK, kernel at coronary CT angiography. Acad Ra- CT. Br J Radiol 2013;86(1023):20120500. et al. Reducing abdominal CT radiation diol 2012;19(11):1324–1331. dose with adaptive statistical iterative re- 130. Baker ME, Dong F, Primak A, et al. Contrast- 107. Min JK, Swaminathan RV, Vass M, Gallagher construction technique. Invest Radiol 2010; to-noise ratio and low-contrast object resolution S, Weinsaft JW. High-definition multidetector 45(4):202–210. on full- and low-dose MDCT: SAFIRE versus fil-

356 radiology.rsna.org n Radiology: Volume 276: Number 2—August 2015 STATE OF THE ART: Iterative CT Reconstruction Techniques Geyer et al

tered back projection in a low-contrast object low can we go? Radiology 2012;265(1):158– 150. Raff GL, Gallagher MJ, O’Neill WW, Goldstein phantom and in the liver. AJR Am J Roentgenol 166. JA. Diagnostic accuracy of noninvasive coro- 2012;199(1):8–18. nary angiography using 64-slice spiral com- 140. Pearce MS, Salotti JA, Little MP, et al. Radi- puted tomography. J Am Coll Cardiol 2005; 131. Hur S, Lee JM, Kim SJ, Park JH, Han JK, Choi ation exposure from CT scans in childhood 46(3):552–557. BI. 80-kVp CT using iterative reconstruction in and subsequent risk of leukaemia and brain image space algorithm for the detection of hy- tumours: a retrospective cohort study. Lancet 151. Alkadhi H, Scheffel H, Desbiolles L, et al. pervascular hepatocellular carcinoma: phantom 2012;380(9840):499–505. Dual-source computed tomography coronary and initial clinical experience. Korean J Radiol angiography: influence of obesity, calcium 2012;13(2):152–164. 141. Mathews JD, Forsythe AV, Brady Z, et al. load, and heart rate on diagnostic accuracy. Cancer risk in 680,000 people exposed Eur Heart J 2008;29(6):766–776. 132. Matsuda I, Hanaoka S, Akahane M, et al. to computed tomography scans in child- Adaptive statistical iterative reconstruction for hood or adolescence: data linkage study 152. Leschka S, Stinn B, Schmid F, et al. Dual volume-rendered computed tomography por- of 11 million Australians. BMJ 2013;346: source CT coronary angiography in severely tovenography: improvement of image quality. f2360. obese patients: trading off temporal resolution Jpn J Radiol 2010;28(9):700–706. and image noise. Invest Radiol 2009;44(11): 142. Goske MJ, Applegate KE, Boylan J, et al. The 720–727. 133. Kambadakone AR, Chaudhary NA, Desai GS, Image Gently campaign: working together Nguyen DD, Kulkarni NM, Sahani DV. Low- to change practice. AJR Am J Roentgenol 153. Chinnaiyan KM, McCullough PA, Flohr TG, dose MDCT and CT enterography of patients 2008;190(2):273–274. Wegner JH, Raff GL. Improved noninvasive with Crohn disease: feasibility of adaptive coronary angiography in morbidly obese pa- statistical iterative reconstruction. AJR Am J 143. Scheffel H, Stolzmann P, Schlett CL, et al. tients with dual-source computed tomogra- Roentgenol 2011;196(6):W743–W752. Coronary artery plaques: cardiac CT with phy. J Cardiovasc Comput Tomogr 2009;3(1): model-based and adaptive-statistical itera- 35–42. 134. Lee SJ, Park SH, Kim AY, et al. A prospective tive reconstruction technique. Eur J Radiol comparison of standard-dose CT enterogra- 2012;81(3):e363–e369. 154. Renker M, Geyer LL, Krazinski AW, Silverman phy and 50% reduced-dose CT enterography JR, Ebersberger U, Schoepf UJ. Iterative image with and without noise reduction for evalu- 144. Utsunomiya D, Weigold WG, Weissman reconstruction: a realistic dose-saving method ating Crohn disease. AJR Am J Roentgenol G, Taylor AJ. Effect of hybrid iterative re- in cardiac CT imaging? Expert Rev Cardiovasc 2011;197(1):50–57. construction technique on quantitative Ther 2013;11(4):403–409. and qualitative image analysis at 256-slice 135. Flicek KT, Hara AK, Silva AC, Wu Q, Pe- prospective gating cardiac CT. Eur Radiol 155. Wang R, Schoepf UJ, Wu R, et al. Image ter MB, Johnson CD. Reducing the radiation 2012;22(6):1287–1294. quality and radiation dose of low dose coro- dose for CT colonography using adaptive sta- nary CT angiography in obese patients: sino- tistical iterative reconstruction: a pilot study. 145. Han BK, Grant KL, Garberich R, Sedlmair gram affirmed iterative reconstruction versus AJR Am J Roentgenol 2010;195(1):126– M, Lindberg J, Lesser JR. Assessment of an filtered back projection. Eur J Radiol 2012; 131. iterative reconstruction algorithm (SAFIRE) 81(11):3141–3145. on image quality in pediatric cardiac CT data- 136. Fletcher JG, Grant KL, Fidler JL, et al. Vali- sets. J Cardiovasc Comput Tomogr 2012;6(3): 156. Kligerman S, Mehta D, Farnadesh M, Jeudy dation of dual-source single-tube reconstruc- 200–204. J, Olsen K, White C. Use of a hybrid iteration as a method to obtain half-dose images to tive reconstruction technique to reduce image evaluate radiation dose and noise reduction: 146. Tricarico F, Hlavacek AM, Schoepf UJ, et al. noise and improve image quality in obese pa- phantom and human assessment using CT Cardiovascular CT angiography in neonates tients undergoing computed tomographic pul- colonography and sinogram-affirmed iterative and children: image quality and potential monary angiography. J Thorac Imaging 2013; reconstruction (SAFIRE). J Comput Assist To- for radiation dose reduction with iterative 28(1):49–59. mogr 2012;36(5):560–569. image reconstruction techniques. Eur Radiol 2013;23(5):1306–1315. 157. Desai GS, Uppot RN, Yu EW, Kambadakone 137. Yoon MA, Kim SH, Lee JM, et al. Adaptive sta- AR, Sahani DV. Impact of iterative recon- tistical iterative reconstruction and Veo: assess- 147. Brady SL, Yee BS, Kaufman RA. Character- struction on image quality and radiation dose ment of image quality and diagnostic perfor- ization of adaptive statistical iterative recon- in multidetector CT of large body size adults. mance in CT colonography at various radiation struction algorithm for dose reduction in CT: Eur Radiol 2012;22(8):1631–1640. doses. J Comput Assist Tomogr 2012;36(5): a pediatric oncology perspective. Med Phys 158. Gebhard C, Fuchs TA, Fiechter M, et al. 596–601. 2012;39(9):5520–5531. Image quality of low-dose CCTA in obese 138. Juri H, Matsuki M, Itou Y, et al. Initial ex- 148. Singh S, Kalra MK, Shenoy-Bhangle AS, et al. patients: impact of high-definition computed perience with adaptive iterative dose reduc- Radiation dose reduction with hybrid iterative tomography and adaptive statistical iterative tion 3D to reduce radiation dose in computed reconstruction for pediatric CT. Radiology reconstruction. Int J Cardiovasc Imaging tomographic urography. J Comput Assist To- 2012;263(2):537–546. 2013;29(7):1565–1574. mogr 2013;37(1):52–57. 149. Vorona GA, Zuccoli G, Sutcavage T, Clayton BL, 159. Willemink MJ, Schilham AM, Leiner T, Mali 139. Kulkarni NM, Uppot RN, Eisner BH, Sahani Ceschin RC, Panigrahy A. The use of adaptive WP, de Jong PA, Budde RP. Iterative recon- DV. Radiation dose reduction at multidetector statistical iterative reconstruction in pediatric struction does not substantially delay CT im- CT with adaptive statistical iterative recon- head CT: a feasibility study. AJNR Am J Neuro- aging in an emergency setting. Insights Imag- struction for evaluation of urolithiasis: how radiol 2013;34(1):205–211. ing 2013;4(3):391–397.

Radiology: Volume 276: Number 2—August 2015 n radiology.rsna.org 357