Technological developments of X-ray computed tomography over half a century: User ’ s influence on protocol optimization

Since the introduction of Computed Tomography (CT), technological improvements have been impressive. At the same time, the number of adjustable acquisition and reconstruction parameters has increased substantially. Overall, these developments led to improved image quality at a reduced radiation dose. However, many pa- rameters are interrelated and part of automated algorithms. This makes it more complicated to adjust them individually and more difficult to comprehend their influence on CT protocol adjustments. Moreover, the user ’ s influence in adapting protocol parameters is sometimes limited by the manufacturer ’ s policy or the user ’ s knowledge. As a consequence, optimization can be a challenge. A literature search in Embase, Medline, Cochrane, and Web of Science was performed. The literature was reviewed with the objective to collect information regarding technological developments in CT over the past five decades and the role of the associated acquisition and reconstruction parameters in the optimization process.


Introduction
Computed tomography (CT) has fundamentally changed the practice of medicine and continues to expand our knowledge about diseases and management of major health challenges [1]. Consequently, the number of CT scans performed worldwide is constantly increasing. The number of CT exams obtained per year in the United States was around 3 million at the early eighties, increasing with approximately a factor of 20-62 million performed CT exams in 2007 [2]. The rapid increase of CT use was seen in European countries as well, and where previously CT scans of mainly adults were performed, there is an increase of CT's performed in pediatric patients [3,4]. Especially the latter are believed to benefit from technological innovations such as high-speed CT scanning that improve diagnostic capabilities in these patients. But, just like in adults, these scans should always be justified [2,5]. Without a doubt, CT scanning is the biggest contributor of radiation exposure to the collective effective dose of medical examinations worldwide [3,6]. Dutch researchers found that the increase in CT exams was not primarily due to the growth and aging of the Dutch population, but can be explained by its easy accessibility, associated technological developments and capabilities. In parallel with the increase of performed CT scans, public awareness and concerns about medical radiation exposure increased [7,8]. The radiology community is aware of this fear and technological developments for radiation dose optimization have always been at their attention. However, optimization of a scanning protocol with respect to image quality (IQ) and radiation dose is a delicate procedure, mainly due to the interrelation of parameters. Furthermore, system properties and accompanying data acquisition techniques changed and expanded over the years. In this paper, we present an overview of the technological developments during the evolution of CT and the accompanying user's influence for protocol optimization. Finally, a future outlook on technological developments is given. restriction. The full search syntax is provided in the Appendix A of the supplementary material. Duplicates were removed and reference lists of included articles and review articles were searched for additional articles. First, articles were screened on title and abstract. Non-original research articles, e.g. case-reports, and original research articles not containing information on image quality and radiation dose regarding CT were excluded. Inclusion, exclusion, and screening of all articles was performed by one author (RBo). Selection criteria were articles containing information regarding key technological developments in CT and the accompanied influence of those developments on image quality and/or radiation dose. After the search, we continued to prospectively add recent articles of which we thought that they supported the text.

System properties
The user's influence and choices in protocol optimization depend on the CT scanner's technological capabilities and system properties. Main technological developments of system properties, acquisition, and reconstruction parameters are presented in Table 1 and are discussed below. An overview of the evolution of CT scanners and the technical advances in CT, is illustrated in Fig. 1.

Translation-rotation and slip ring technology
Initially, CT images were acquired by the translation-rotation method in the "first and second" generation CT scanners. Within this method, data was acquired by the x-ray tube and detector moving in a linear translatory pathway and was repeated with small rotational increments [9]. The third generation CT scanners have a wide fan beam and detectors that rotated slowly around the patient, requiring multiple breath holds to complete an axial CT exam. There was a high chance in missing abnormalities due to the multiple breath holds (Fig. 1a). Slip ring technology introduced in 1987 allowed continuous rotation of the tube and detectors by transferring electrical energy to the rotating gantry part and transmission of measured data to the computer system [10]. As the fourth generation scanners, with a stationary detector ring, did not get widely accepted, all currently available CT scanners are third generation scanners by design. Therefore, we will only briefly comment on special scanner concepts like electron beam CT and dynamic spatial reconstructor.

Detectors
The total beam collimation in the longitudinal, or z-direction, in the first-generation CT scanners was limited to one detector of 8-13 mm in width, but detector size decreased rapidly to 2-8 mm [11,12]. With the introduction of spiral multi-detector CT (MDCT) in 1998 (also known as multi-slice CT), the individual detector elements became even smaller, down to 0.25 mm per detector element nowadays [13], resulting in improved spatial resolution. Moreover, it provided more and fast longitudinal coverage since multiple detector elements were combined ( Fig. 1) [14]. Currently, for several CT manufacturers the total beam collimation is up to 160 mm with multiple detectors in the z-direction, allowing dynamic data acquisition of e.g. the entire brain or heart without table movement [15,16]. Another positive outcome of an increased total collimation, is the decrease of the overbeaming effect: The collimated x-ray beam is always wider than the total detector width because of the penumbra, which does not contribute to the image reconstruction, but does increase radiation dose. Although the impact of overbeaming on radiation dose was reduced with increased total collimation, overranging dose increases with increasing collimation and pitch values [17]. Therefore, a dynamic collimator was introduced in 2009 to reduce the amount of pre-and post-spiral dose which are irrelevant for image reconstruction and is automatically applied [18].
Another approach to detector developments were improved detector efficiency to increase radiation dose efficacy, and the introduction of dual layer detectors. These detectors can measure x-ray attenuation for low and high energy photons separately in two different detector layers, enabling material identification and quantification [19].

X-ray tube
With the introduction of spiral CT, the x-ray tubes had to be redesigned again to cope with overheating problems because of the need for increased tube output [20]. The introduction of a periodic motion of the focal spot in the z-direction resulted in doubling measurement positions in the longitudinal direction per rotation; thereby increasing spatial resolution and eliminating aliasing artifacts [21]. This multifan measurement technique is commonly known as z-flying focal spot and "double-dynamic" focus and applied by several vendors Recent developments also include an additional tin filtration within the x-ray tube, which is of particular use in e.g. unenhanced CT high contrast studies of the chest and sinus [22,23] and is currently applied by one vendor.

Dual source CT (DSCT)
CT scanners with multiple x-ray sources can provide fast imaging and improved temporal resolution (TR). The dynamic spatial reconstructor was one of the first attempts to introduce such a CT system but was never used in clinical routine [24]. In 2005, the first DSCT with two tubes and two corresponding detectors was introduced, demonstrating improved TR and dual energy imaging capabilities in clinical practice which was widely accepted [25].

Acquisition parameters
The main developments in acquisition parameters and how they influence image acquisition are discussed next.

Tube current
Within the first-generation CT scanners, the user could set tube current (mA value) depending on the accompanying tube voltage [26,27]. Tube current was constant during a scan and this remained so for almost twenty years.

Automated tube current modulation (ATCM)
ATCM was introduced end '90 s as part of the automatic exposure control (AEC) [28]. Early strategies consisted of online angular tube current modulation only, where nowadays it is often applied in combination with tube current adaptation in longitudinal directions. Some strategies enabled users to set customizable quality levels to achieve a constant noise level, whereby tube current is adjusted for the chosen scan and reconstruction parameters. Algorithms within the latest systems may suggest adjustments to average tube current and image noise based on a user defined dose index and patient diameter, accounting for the use of iterative reconstructions (IR) and used tube voltage. Another strategy was to have the ATCM system measure the attenuation from patients in a specific protocol, using this as a standard protocol body attenuation. The user can determine a noise reference or set the tube current to individual patient habitus. A different approach of fully ATCM is adaptation to different anatomical regions and patient sizes by setting a target tube current level for a standard-size reference patient [29]. The user may set different tube current modulation schemes for different patient sizes and anatomical regions.
A high level of awareness by the users for optimal positioning of the patient in the CT scanners is of utmost importance [30,31]. Both radiation dose and IQ may be affected when the CT localizer radiograph, which is used by the AEC, is made with the patient positioned off-center [30,31].

Tube voltage
Within the first and second generation CT scanners, the user was able to set the peak tube voltage in the range of 100-140 kilovolt peak (kVp) [27,32,33]. These high voltages are much appreciated when imaging thick patients, or to reduce metal implant artifacts, however radiation dose is likely to be increased. Lowering the tube voltage requires tube current to be increased, and this was often limited by tube power early on.

Automatic tube voltage selection
Changing the tube voltage in predefined scan protocols, requires understanding of its influence on signal-to-noise ratio and contrast-tonoise ratio (CNR). Therefore, it could be challenging for users to understand how to perceive an improved IQ, or even the same IQ while reducing radiation dose, when changing the tube voltage. It was until the '10 s that integrated automatic tube voltage selection and accompanying tube current adjustment became fully integrated into the AEC. Currently, it is available in most CT systems [34]. The main goal of automated tube voltage selection is to control the CNR and thereby minimize radiation dose. The user can define settings for the anatomical region and exam type with or without contrast.

Dual energy imaging
Dual energy, or so-called spectral imaging, can add tissue information to the CT image (e.g. discriminate bone from iodine-enhanced tissue). The possibility of determining the atomic number of the materials within a slice was already discussed by Sir Hounsfield in the seventies [12]. First attempts were done by a double scan: once with a high tube voltage and once with a low tube voltage and in parallel by a rapid kV switching technique. However, clinical use was rather limited due to needed technological improvements and high costs. The introduction of DSCT in 2005 allowed the acquisition of nearly simultaneous dual-energy data by using two tubes (Fig. 1f) [35]. A few years later, this was also made possible with the introduction of an improved rapid tube voltage switching technique [36]. TwinBeam CT and Dual layer spectral CT are the latest technologies to acquire dual energy datasets [19,37].

Sequential scanning
Sequential CT imaging represents scanning with a stationary scanner table while the x-ray tube is rotated around the patient. After the scan, the patient is transported with a predefined incremental step. Then the next acquisition is performed and the process is repeated to the end of the scan range.

Spiral scanning
CT entered a new era with spiral CT (also known as helical scanning) in the late 1980s [38,39]. The scanner table was able to travel at a constant speed through the gantry, i.e. the table feed, with the tube rotating, allowing the acquisition of volumetric data. It also introduced the concept of pitch (the ratio between the table feed per full rotation and total beam collimation) which can be adjusted by the user. With single-detector spiral CT and a reduced rotation time, scan time was reduced. However, scans were restricted to single organs. A complete thorax-abdomen scan required at least two breath holds (Fig. 1b). The introduction of MDCT (Fig. 1c) gave the user the choice to scan with a small detector row width (e.g. 4 × 1 mm) to increase spatial resolution (=detail) or to scan with a large detector size, e.g. 4 × 2.5 mm, to reduce scan time (=volume). Spiral scanning with a 16-row MDCT allowed isotropic datasets of large volumes and an increase in quality of the post-processing images, as demonstrated in Fig. 1b-f. DSCT made scanning at a pitch >2 possible by filling the sampling gaps of one detector with data of the second detector, providing clinical advantages in (cardio)vascular, trauma and pediatric patients due to increased scan speed (Fig. 1f) [40,41].

Rotation time and temporal resolution
Gantry rotation time directly affects TR as data from at least a 180degree rotation are needed to reconstruct an image. Faster gantry rotation times result in improved TR with less motion artifacts and improved clarity of lung and cardiac imaging [42,43]. Gantry rotation times have decreased from 5 to 40 seconds in rotation-translation systems in the seventies to 0.24− 0.30 seconds for the current CT systems [26]. Until today, most single source scanners still cannot reach the 50-100 ms TR of electron beam CT scanners. Those scanners were especially proposed for cardiac CT because they were able to reach good TR thanks to its scanning without mechanical motion [44]. It was until the introduction of DSCT to achieve similar TR with up to 66 ms with 3rd generation CT systems [45].

Electrocardiogram (ECG) synchronization and ECG-guided dose modulation
Cardiac motion limited imaging of the heart in the early years of CT. However, in 1977 there were considerable achievements in technology reducing cardiac motion by ECG-gated reconstruction and provided "stop-action" cardiac CT scans [46,47]. However, acquiring data for a single slice took up to 12 s. Multi-detector spiral CT reduced exam time, enabled reducing contrast volume, improved spatial resolution and ECG-gated coronary CT angiography became feasible in clinical practice. Especially from the 64-row MDCT on, robust low heart rate (HR) cardiac CT was possible (Fig. 1e). [48,49]. At first, only a retrospective spiral scan mode with full dose, during the whole R-R interval at low pitch values was provided [50]. Later on, by introducing adaptive algorithms which can react to heart rate variability and simple arrhythmia, a dose reduction was achieved [51]. When the heart rhythm has complex arrhythmia, often a retrospective protocol is preferred for ECG gated data editing possibilities. However, such a protocol requires a low pitch for oversampling to ensure enough data for reconstruction is available at the expense of a high(er) radiation dose. While a prospective sequential scanning technique might have stack artifacts, a single heart beat scan mode such as a high-pitch prospective scan or a scan with a wide area detector does not. However, both single heart beat techniques require a low and stable heart rate [52].

Reconstruction parameters
Some of the steps in the reconstruction process are not, or to a less degree, adjustable by the user. All of the choices made within the reconstruction process directly influence IQ. We will highlight the main technological developments in reconstruction techniques.

Image reconstruction technique
Within the first CT systems, images were reconstructed with a simple iterative reconstruction method known as algebraic reconstruction [53]. However, due to the lack of computing power, this technique was soon replaced by filtered backprojection (FBP) [54]. FBP images are reconstructed by a convolution method or a direct Fourier algorithm. This second group incorporated interpolation in the Fourier plane, followed by inverse Fourier transformation. Convoluting the attenuation profiles with a so-called kernel and the backprojection of the modified profiles into the image plane to create the final image, is the method known as filtered backprojection. It is an analytical solution of the reconstruction problem. Where FBP was the most widely used CT image reconstruction technique for decades, nowadays mainly IR techniques are applied [55].

Iterative reconstruction technique
Computing power by the late '00 made IR techniques feasible in clinical routine [55]. IR techniques developed rapidly in three steps: Firstly, IR reconstruction was mainly done in the image domain on an initial image reconstructed from the raw data, secondly it went to sinogram-based or so-called hybrid reconstructions. Thirdly, reconstruction algorithms developed to full model-based IR techniques [56]. However, most algorithms remain a "black-box" lacking specific details.

Matrix and FoV
Within the first-generation CT scanners, the image matrix size was limited to 80 × 80 pixels and one could only adjust the window level and width. Nowadays 512 × 512 is the most commonly used image matrix size but CT scanners with sizes up to 2048 × 2048 are available [57].
Extended field of view (FoV) reconstructions allow visualization of skin and tissue outside the primary FoV. This is of importance for PET-CT attenuation correction and radiotherapy CT dose calculations [58].

Cardiac reconstructions
The multiple ECG cycles acquired for cardiac CT in the late seventies were needed for acceptable effective TR with the aid of multi-segment reconstruction. Despite long acquisition time and extensive motion artifacts, the cardiac outline and fat grooves could be sharply visualized. Nowadays, mono-segment reconstruction is often used, but bi-or multisegment reconstruction techniques are still available to make scanning of coronaries at higher heartrates feasible. These methods could improve the TR by a factor of 2 by combining two or more heart beats for one reconstruction, but at the cost of a very low pitch value and consequently an increased radiation dose [50]. A disadvantage of multi-segment reconstruction is a possible creation of blurry images [59]. Even though vendors developed motion reduction algorithms, motion free imaging primarily depends on heart rate and gantry rotation time [60,61].

Image enhancement tools
Several tools to improve IQ are developed and can be manually selected or are integrated into the reconstruction process. The most often used tools are noise and artifact reduction algorithms.

Noise reduction
Recently, noise reduction algorithms are implemented in several reconstruction processes, mostly running in the background e.g. in repeated low dose imaging during dynamic CT perfusion, in order to improve spatial resolution and CNR [62]. Sometimes it can be manually applied by the user e.g. to improve CNR in monoenergetic image reconstruction of dual energy data [63].

Artifact reduction
Artifacts are defined as artificial structures, which deviate from reality. Examples are artifacts occurring from voluntary and involuntary patient motion or beamhardening. Nowadays, motion correction algorithms are often used in CT perfusion of the head and body to correct for subtle head displacement or the breathing state during the acquisition times. The corrections are applied on already reconstructed image data and mainly done in post-processing software. Whereas most of the algorithms for beam hardening correction or metal artifact reduction use iterative algorithms and therefore have to be applied on raw data [64].

Scanning protocol optimization
Technological developments generally resulted in a reduction of radiation dose per exam and improved IQ [65,66]. Both radiation dose and IQ are dictated by the ALARA (As Low as Reasonably Achievable) principle. With the introduction of diagnostic reference levels for CT in 1996, a practical tool came available to promote radiation dose optimization for specific diagnostic tasks [67]. With that, reference levels for CT exams were introduced around the globe [68][69][70][71]. Together with the technological developments it contributed to the decrease of effective dose for a CT exam [72]. However, the diagnostic reference levels are general guidelines and do not apply to optimization for an individual patient. In the meantime, the user is one of the "protocol optimization factors" or may be even the most important factor in the optimization process. The user's contribution to the optimization process depends on the user himself and on the technological developments. All stakeholders, e.g. radiologist, medical physicist, and radiographer should work together and consider the whole optimization process as a team effort. In the next paragraph we will discuss the optimization process. Some optimization steps are highlighted by a single case study (Fig. 2), which covers a wide area of technological developments over more than a decade. Note: As there are several CT manufacturers, so are (subtle) differences in their approaches in the technological developments in system, acquisition, and reconstruction parameters. Generalizations should come in only if features are significantly similar in all or most common vendors.
The whole scan protocol optimization process strives for optimization for an individual patient, taking the specific organ region and the referral question into account. Some technological developments have a direct effect on radiation dose applied to a patient (e.g. tube current). Other developments, like iterative reconstructions or automatic adaptation of tube voltage, are dependent on the user's motivation, acceptation and awareness. Benefits of the increased and evolved technologies are known, but the technological developments were and could be misunderstood or misused, leading to excessive radiation dose to the patient [73,74]. Thereby, awareness of radiation dose and the possible risks are not always known [75].
Within the optimization process, the user's influence has increased, while automated tools were integrated to assist in the optimization process. This does not mean that changing a parameter will lead to an automatic compensation in other features/parameters, for example to maintain image quality. Many of the acquisition and reconstruction parameters are interrelated, making them more complicated to adjust individually and more difficult to comprehend, especially when they are part of automated algorithms and, likely, in the near future with artificial intelligence. Nevertheless, technological improvements and automated tools, combined with attention to the human side by the radiographer, will lead to the optimal scanning procedure. For example, automatization might speed up the scanning and reconstruction process, while the main focus of the radiographer is on the patient itself. In the meantime, adjusting parameters is like slotted dials: On the road to optimization, regardless of whether the adjustments have been made by humans or artificial intelligence, an adjustment of an acquisition or reconstruction parameter will have a direct influence on image quality and, directly or indirectly, on radiation dose as well due to their interrelation (Fig. 3). Within this light, it is mandatory to focus first on diagnostic optimization, which can be defined, and achieved, by the determination of the minimally acceptable IQ for diagnosis and thus of the lower limit of the diagnostic reference level. Minimally acceptable IQ is set by the desired image contrast, spatial resolution, and the amount of artifacts accepted [76,77]. The second step will be technological optimization, defined as parameter selection to achieve this preferred lower limit IQ at the lowest reasonable dose. Fig. 2 shows an example how a thoracic scanning protocol was technologically optimized. Users should be aware that diagnostic and technological optimization outcomes may vary between different CT scanners and institutions with different IQ preferences [71,78]. The impact of a change in acquisition and reconstruction parameters on IQ and radiation dose, together with considerations for protocol optimization is illustrated in Table 2. This table is used as a guidance for the next paragraphs to discuss the impact of adaptation of a single parameter on IQ and radiation dose.

Acquisition parameters
Protocol optimization for every individual patient can be obtained by adaptation of a single or multiple acquisition parameters. Every parameter demands its own consideration for optimization ( Table 2, "considerations for CT protocol optimization"). For instance, when objects have slight attenuation differences such as in soft tissue studies, image noise impairs contrast resolution. So again, it is essential to determine the tolerance level of noise in the CT images as Sir Hounsfield already stated in 1976 [79]. An increase in noise is not problematic in objects with high intrinsic contrast, e.g. bone and air ways. [57,79].
Adaptation of the tube voltage will have different effects and depends on whether or not iodinated contrast material is used ( Table 2, "acquisition parameters", "image quality", "radiation dose") and on the general strategy for using automatic tube voltage selection [80]. X-ray attenuation by objects such as bone and iodine contrast strongly depends on the photon energy due to their high atomic number. Therefore, when iodine material is used, an improved CNR is possible, e.g. to better depict enhancing masses, at a low tube voltage with a dose similar to a high tube voltage scan (Fig. 4AB) [81]. On the other hand, for scanning protocol optimization in e.g. young patients, the user may consider a reduction of radiation dose while maintaining CNR (Fig. 4C) [82]. While the main goal of automatic tube voltage selection is to control the CNR and thereby minimize radiation dose, sometimes the user should adjust the proposed parameters by the scanner software for an individual patient, instead of following the general strategy for automatic tube voltage selection. Thus, in some cases the referral question or individual patient demands for a higher radiation dose. For example, the user may also want to apply a higher contrast volume or flow since a high tube voltage decreases iodine contrast enhancement (Fig. 5AB).
The presence of high attenuating materials such as a hip prosthesis (Fig. 5CD), warrants an increased tube voltage to decrease artifacts when no metal artifact reduction techniques are available ( Table 2, "risk of artifacts").
Continuing with the parameter adaptation shown in Table 2: In general, tube current adaptation is not dependent on the use of iodine material, but rather on the noise tolerated in the images. Modulation of the tube current is used throughout most of the CT scanning protocols. Its use changes the relative dependencies in individual exposure parameters. For example, changing the pitch or rotation time often does not affect the patient's dose, as a change in tube current compensates for the change in other parameters [83]. However, when using ATCM, one should be aware that specific parameters, like slice thickness, kernel, and tube voltage, may affect the behavior of ATCM and that this differs between vendors [84].
Considerations to increase TR and the pitch mostly depend on the need of decreasing motion artifacts ( Fig. 2B and 2C), mainly when Fig. 3. Graphical illustration by slotted dials, demonstrating the balance between optimization of a scanning protocol with respect to image quality and radiation dose. Changing system properties or parameters, input of human or artificial intelligence will influence both radiation dose and image quality.
imaging the heart or scanning non-cooperative patients. However, a higher pitch value often demands a higher tube current, especially in scanners that keep the noise and dose level constant. Faster rotation times may increase artifacts. Therefore, in cases of cooperative patients, the user may decrease the pitch to decrease the overranging effect. Moreover, this will also lead to increased IQ in e.g. bone exams, especially when the structures are angulated relative to scan plane [18,85].

Reconstruction parameters
CT protocol optimization is also obtained by adaptation of single, or multiple, reconstruction parameters (Table 2). In image reconstruction, when selecting the level of smoothing (minimal, moderate, or maximum), the user can improve low contrast detectability by reducing the amount of noise. The other way around, edge-enhancement filters improve spatial resolution, by "sharpening up" the CT image and are especially useful in bone or lung exams [86]. Other filters may increase Table 2 Overview of general impact when adapting acquisition and reconstruction parameters currently used.  Note. : Increase is demonstrated with the "+", decrease with the "− ", and (almost) equal effect with the "≈". dFoV = display field of view. Data in paranthesis 1=beamhardening; 2=streak; 3=windmill/spiral; 4=stair-step; 5=motion/breathing/pulsation; 6=partial volume effect/ blooming; a=active flying focal spot.
metal to tissue transition such as stent lumen by reducing blooming effects [87]. An improved spatial resolution comes with an increased noise level and reduced soft tissue contrast. Within iterative reconstruction techniques, careful considerations in iterative strength, also known as level or scale, and accompanied dose adjustments are mandatory [88,89]. For instance, higher iterative strength can have an effect on contrast and spatial resolution, but also on image appearance [90]. The image texture might be blurred and a high iterative strength can give rise to a noiseless image appearance. These kind of images are often evaluated as too smooth or artificial, neither are often desired by users [91]. Reliable diagnostic quality and statistically significant dose reductions can be achieved in adult and pediatric patients using IR [92,93]. However, negative effects as low contrast detectability are reported as well [94].
Spatial resolution may increase with increased matrix size thanks to a decrease of the voxel size. In general this will be accompanied by an increase of noise (Table 2). Moreover, users should also be aware that image data size increases with increased matrix size. Adaptation of the FoV is also related to the voxel size: Increasing or decreasing the FoV will directly influence voxel size. Thereby, it may affect IQ: a smaller FoV may increase spatial resolution, but decrease contrast resolution due to increase of noise. Balancing between optimization of a protocol with respect to IQ and radiation dose, e.g. the increase of spatial resolution, at the cost of image noise, the user may also want to adjust the slice thickness to restore the signal-to-noise ratio. For example, when an increase in contrast resolution is required, noise levels can be lowered by increasing slice thickness (Fig. 6AB). Simultaneously, spatial resolution will decrease due to the partial-volume effect (Fig. 6CD).  4. (a-c) Axial CT images of the same human abdomen acquired with equal CTDIvol and contrast injection protocol. Window width and level were 300/30. Images made with two days in between. (a) Demonstrating an increased contrast to noise ratio (CNR) when applying a lower tube voltage of 80 kVp compared to the CNR observed in (b) with 120 kVp. (c) CT image (maximum intensity projection, coronal view) of the heart of a thirteen-year-old boy. Reduced radiation dose in coronary CT angiography when applying low tube voltage (70 kV, a total dose length product of 8.2 mGy*cm, and a SSDE of 0.77 mGy).

Future outlook and conclusion
CT is still evolving, even in its middle age, and bringing new technological developments and new diagnostic strategies for healthcare. Users should not only be at the forefront in embracing latest technologies, but also be proactive on the road to highly optimized protocols.
Currently, photon counting CT (PCCT) and artificial intelligence (AI) promise to bring a new revolution in CT [55,72] (Fig. 1g). PCCT is expected to provide intrinsic spectral sensitivity, high spatial resolution, less noise and artifacts with better low-signal performance, and less characteristic energy-weighting [55,95]. PCCT opens the possibility of achieving multi-energy imaging in every scan, similar to dual energy CT, but using a single tube voltage. Where dual layer CT uses a single tube voltage too, PCCT is able to count the number of all incoming photons and measure its energy. Improved iodine contrast visibility may even require less radiation dose, or lower iodine contrast material injection [96,97].
AI is already applied within clinical protocols for instance in artifact reduction and image reconstruction [95]. As such the application of AI resembles IR: its application can be used to reduce radiation dose while maintaining IQ or increase IQ without increasing radiation exposure [55]. Both PCCT and AI are one of the latest technological developments in almost five decades of CT, but certainly will not be the last to be introduced and demanding an adjustment of the optimization process.
In conclusion, technological developments in CT have led to an increased number of processes for protocol optimization. Consequently, it is necessary that users are aware of those developments, their operation, and how they are interrelated with respect to image quality and radiation dose.

Guarantor
The scientific guarantor of this publication is Marcel van Straten

Statistics and biometry
One of the authors has significant statistical expertise.

Informed consent
Written informed consent was not required for this study because this concerns a narrative review paper.

Ethical approval
Institutional Review Board approval was not required because this concerns a narrative review paper.

Funding
The authors state that this work has not received any funding.

Acknowledgements
We would like to thank Dr. Wichor Bramer of the medical library  6. (a-d) CT images illustrating the influence of adjusting a reconstruction parameter (slice thickness) and its influence on contrast and spatial resolution. (a-b) CT images (axial view) of an adult brain (soft tissue) (a) 1.0 mm and b 5.0 mm. (c-d) CT images (multiplanar reconstruction, coronal view) of a wrist (bone). Contrast resolution is increased with thicker slices due to decrease of noise. (c) 0.75 mm and (d) 5.0 mm. Spatial resolution is increased with thinner slices due to decrease of the partial volume effect.
Erasmus MC for his support in building the search syntax.

Appendix A. Supplementary data
Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.ejrad.2020.109261.