Toward Large-Scale Process Control to Enable Consistent CT Radiation Dose Optimization

Transcription

1 Pediatric Imaging Review Larson et al. CT Radiation Dose Optimization Pediatric Imaging Review FOCUS ON: David B. Larson 1 Keith J. Strauss 2 Daniel J. Podberesky 3 Larson DB, Strauss KJ, Podberesky DJ Keywords: CT radiation dose, optimization, patient safety, process control, quality improvement DOI:1.2214/AJR Received October 4, 214; accepted after revision November 18, 214. Based on a presentation at the Society for Pediatric Radiology 214 annual meeting, Washington, DC. D. B. Larson has developed intellectual property related to CT radiation dose optimization and process control. K. J. Strauss provides medical physics consultation services to Philips Healthcare on a requested basis. D. J. Podberesky is a member of the Professional Speaker Bureau of Toshiba of America Medical Systems, a consultant for Guerbet, an author for Amirsys, and receives reimbursement for travel from Philips Healthcare, GE Healthcare, and Siemens Healthcare. 1 Department of Radiology, Stanford University School of Medicine, 3 Pasteur Dr, Stanford, CA Address correspondence to D. B. Larson (david.larson@stanford.edu). 2 Department of Radiology, Cincinnati Children s Hospital Medical Center, Cincinnati, OH. 3 Department of Radiology, Nemours Children s Hospital, Orlando, FL. AJR 21; 24: X/1/24 99 American Roentgen Ray Society Toward Large-Scale Process Control to Enable Consistent CT Radiation Dose Optimization OBJECTIVE. This article reviews the concepts of CT radiation dose optimization and process control, discusses how to achieve optimization and how to verify that it is consistently accomplished, and proposes strategies to move toward large-scale application. CONCLUSION. CT dose optimization is achieved when the least amount of radiation necessary is used to achieve adequate image quality. The key to consistent optimization is minimization of unnecessary variation. This minimization is accomplished through local process control mechanisms. I n 21, two highly publicized articles in AJR revealed that radiology practices were not consistently optimizing CT radiation dose when scanning children [1] and that this lack of dose optimization was leading to an increased risk of radiation-induced cancer [2]. Since then, substantial amounts of time, effort, and money have been invested in published research regarding CT radiation dose management [3 ]. Despite these investments, the evidence suggests that widespread CT radiation dose optimization has not been achieved [6 8]. This lack of progress begs the questions: What barriers are preventing this from happening? Will widespread consistent CT dose optimization realistically occur in the future? We think that better local process control mechanisms are needed to achieve consistent optimization of CT radiation dose. Although professional societies and accreditation bodies may recommend or require documentation of estimated patient dose for a few standard examinations, these quality control programs typically do not ensure that a practice s CT radiation doses are well controlled for examinations of patients of all sizes. This article reviews the concepts of CT radiation dose optimization and process control, discusses how to achieve optimization and how to verify that it is consistently accomplished, and proposes strategies to move toward large-scale application. Ten principles central to the understanding of process control as applied to CT radiation dose manage- ment at the local level are presented; these principles are focused on the individuals who manage CT dose on a daily basis. Four proposals for regional, national, and international strategies to support local efforts are also presented; these proposals illustrate how certain important aspects of the approach by individuals in oversight positions can help support how dose is managed at the local level. CT of the abdomen and pelvis is used as a prototypical example of these principles. Concepts of Optimization and Process Control The Oxford Dictionaries website [9] defines optimize as to make the best or most effective use of. The engineering definition of the term optimization refers to a field of applied mathematics whose principles and methods are used to solve quantitative problems in disciplines including physics, biology, engineering, and economics [1]. Optimization mathematically frames a problem to reach a goal. For example, a driver may wish to optimize speed based on a vehicle s fuel economy. At low speeds, fuel economy is low because of fuel consumption required to power accessories and operate the engine near idle. Fuel economy improves with increasing speed and then tapers off at high speeds because of increased air resistance and other opposing forces. The speed at which the fuel economy peaks is the optimum speed; it can be predicted mathematically. However, other factors can also play a role in determining AJR:24, May 21 99

2 Larson et al. Fig. 1 CT radiation dose optimization problem. Operator has direct control over only scanner settings (tube voltage [kv] and tube current exposure time product [mas]) and must adjust those settings, accounting for uncontrollable variables, to achieve target image quality and patient dose. Middle row represents target image quality and dose. These targets are consistently achieved as long as operator can accurately predict outcomes and adjust settings accordingly. Feedback mechanism allows model to be adjusted to more accurately achieve targets. If predictive model is not available, operator must rely on experience, which is generally based on trial and error. (Illustrations by Larson D) the appropriate speed. If the optimal speed is higher than a legal speed limit, for example, the problem is constrained by the speed limit instead of the fuel economy. The major elements of an optimization problem are input variables, outcome variables, and constraints. Some of the input variables, such as the accelerator pedal, can be controlled, whereas other input variables, such as the terrain, cannot. The operator must respond to uncontrollable variables by adjusting input variables. Multiple outcome variables (e.g., speed and fuel economy) may exist with varying degrees of relevance. Some constraints (e.g., engine power) may impose physical limitations on the system, whereas other constraints (e.g., the speed limit) may constitute artificial limitations on outcomes. The optimization problem determines the target outcome and the inputs needed to achieve that outcome. Process control is a system for achieving optimal outcomes on a consistent basis in the practical setting [11]. Process control systems provide methods to monitor outcome variables and adjust the input variables accordingly. Consistently driving at an optimal speed over varying terrain is an example. The driver continuously varies the position of the accelerator pedal to maintain constant speed. Because the vehicle s response to the terrain is predictable and the driver has real-time visual and quantitative feedback, the driver has control of the vehicle s speed. However, if the vehicle s response were unpredictable or if the driver had no visual reference or speed indicators, the driver would not really be in control of the vehicle s speed despite having direct control of the accelerator pedal. These closedloop control systems are commonplace in other industries such as manufacturing. However, closed-loop systems are largely absent in clinical practice in CT. Although operators have direct control of the CT settings, they generally do not have access to the other essential elements of a process control system such as feedback mechanisms. The absence of closed-loop systems in the management of CT radiation dose explains the lack of consistent dose optimization. The optimization problem of CT radiation dose management can be summarized as follows: CT radiation dose is optimized when the scanner controls are set to expose a patient of a given size to the lowest possible radiation dose while providing adequate diagnostic image quality [12]. Controllable input variables are scanner settings. Uncontrollable input variables include the patient size and examination type. Outcome variables include the image quality and patient radiation dose (Fig. 1). Optimization is achieved when radiation dose is minimized subject to the constraint of achieving adequate image quality on a given examination. Process control is accomplished when optimization is achieved on a consistent basis in practice [13]. Principles of CT Radiation Dose Process Control at the Local Level We highlight 1 principles essential to achieving consistent CT radiation dose optimization using process controls at the local level. Some elements of this approach may not be intuitive because comprehensive process control methods typically are not part of CT dose management at the current time. Principle 1: The Operator Does Not Have Direct Control Over Patient Dose The CT operator indirectly controls the target outcome variables of image quality and patient dose by manipulating the scanner settings (i.e., the controllable variables) (Fig. 1). The operator s level of control of patient dose depends on the operator s ability to predict patient dose on the basis of scanner inputs. This prediction requires a feedback mechanism that links the target outcomes to the controllable variables. The feedback loop is a critical element of process control. If the monitored outcome variable provides immediate feedback to the controllable input variables, the feedback loop alone enables control. For example, a coffee dispenser could have a detector that sends a real-time feedback signal to stop the flow of coffee when the cup is full regardless of cup size (i.e., real-time closed-loop feedback). If this type of detector is not feasible in a coffee dispenser, the machine could be programmed to dispense a fixed volume of coffee based on user input regarding the size of the cup. Because no real-time closed-loop feedback mechanism exists in this case, filling the cup correctly depends on accurate information from the operator and accurate prediction by the machine. After entering the cup size, the operator has no further control; the process runs its course to completion. 96 AJR:24, May 21

3 CT Radiation Dose Optimization 2 SD of Image Noise (HU) A B Fig. 2 Illustrations show image noise and size-specific dose estimate (SSDE) as function of patient size for CT of abdomen and pelvis. A and B, Illustrations show image noise (A) and SSDE (B) as function of patient size for CT of abdomen and pelvis. Solid black lines represent target image quality (A) and target SSDE (B) as function of patient size. Dashed lines represent effect of keeping tube output constant for all examinations regardless of patient size, and gray lines represent effect of keeping image noise constant for all examinations regardless of patient size. Similarly, once a CT operator starts scanning, scanning continues until the examination is completed. Although the automatic tube current modulation (ATCM) system of the scanner may use internal real-time feedback loops, this logic is embedded in the scanner just as the logic for coffee cup filling is embedded in the coffee dispenser. The operator can manipulate the ATCM settings, but the operator must predict the final image quality on the basis of the scanner settings and enter the correct settings before the examination starts. If the operator is not able to mathematically predict the outcomes using patient characteristics and scanner settings, the settings must be chosen on the basis of experience, which generally involves trial and error (Fig. 1). If certain settings produce excellent image quality, the dose is lowered for a group of patients. If the image quality is poor, the dose is raised. Although these results serve as a feedback mechanism, this type of feedback mechanism is a relatively crude one. This approach is not optimization from an engineering perspective because there is no mathematic understanding of the relationship of input variables to outcome variables. Principle 2: Consistent Optimization Requires a Target Outcome Process control requires a so-called set point, which is the value of the desired outcome variable. This set point is realized by adjusting the available input variables. In the vehicle speed example, the chosen speed is the desired outcome; the driver adjusts the accelerator pedal to maintain the chosen speed. The establishment of a target outcome is a sine qua non for consistently optimizing a process and for verifying that the process is, in fact, optimized. One cannot reach a target on a consistent basis if no target exists. Principle 3: CT Image Quality, Not Patient Dose, Constrains the Optimization Problem and Determines the Target Outcome The objective in CT radiation dose optimization is to manage radiation dose while maintaining adequate diagnostic image quality. Reaching that objective is a two-step process. The first step is to determine what is meant by adequate diagnostic image quality. Image quality is difficult to objectively measure, and those measures are difficult to link to a diagnostic outcome. Fortunately, subjective radiologist assessments can reasonably determine the point at which image quality would become inadequate with further dose reduction by viewing a series of images of the same anatomy at different dose levels. This so-called floor for image quality becomes the primary constraint of the optimization problem. The floor for image quality must be established for the process control system to function. Subjective image quality assessment is an admittedly imperfect mechanism for determining optimal image quality. Although some individuals would prefer a more robust evaluation of image quality to establish the primary constraint, one should not abandon the entire process control effort because of the limitations of subjective image quality assessment. In the second step, scanner settings that result in achieving image quality targets are determined. There are a variety of scanner settings that can be used to obtain images of specified quality. These different settings may deliver different doses to the patient [14]. The optimization problem is solved when the settings are determined that will result in images of target image quality at the lowest possible dose. Principle 4: The Process Control Model Must Be Framed to Recognize Image Quality as the Primary Target Outcome Variable The primary focus of process control in CT is the reduction of unnecessary variation in image quality from patient to patient using the lowest dose methods available (i.e., optimization). Verifying that the process is in control requires an image quality measure. Unfortunately, a measure of image quality is not readily available. Image noise is one possibility, but it cannot be reliably measured on clinical images because of patient variation and varying image-processing techniques. One can monitor the variation in image quality by monitoring the variation in patient dose, although this strategy is effective only if the dose estimate is strongly correlated with image quality. In other words, an increase in the dose measure must directly correspond to an increase in image quality. If an increase in the measure could represent an increased dose without an increase in image quality, such as an increase in scan length, then the measure is not suitable for this purpose. A measure of scanner output such as the volume CT dose index (CTDI vol ) could be used, but the CTDI vol does not accurately AJR:24, May

4 Larson et al reflect patient dose because it is not adjusted for patient size. The size-specific dose estimate (SSDE) [1] is preferred because it estimates the dose to the patient regardless of the patient s size; thus, SSDE is a reasonable estimate of both patient dose and image quality. Process control in CT is focused on minimizing unnecessary variation of image quality from patient to patient and on using the lowest-dose methods to achieve the desired image quality. This challenge must not be confused with that of seeking a more accurate estimate of the patient s organ doses from CT to provide a better estimate of associated cancer risks. Although more accurate and rigorous methods for assessing organ dose may assist in assessing risk, these methods do not help eliminate unnecessary variation in image quality or patient dose (i.e., optimization) from one CT examination to the next. It is perhaps counterintuitive that more A C accurate estimates of radiation dose generally do not contribute to improved CT radiation dose optimization as defined here. Principle : Consistently Reaching an Image Quality or Dose Target Is Difficult to Achieve in Practice A number of factors make consistent image quality and CT radiation dose optimization an especially difficult challenge in clinical practice. For CT of the abdomen and pelvis, the optimal image noise level differs over the range of patient sizes. Specifically, a radiologist can tolerate greater image noise in larger patients than in smaller patients [12] (Fig. 2A, solid black line). The SSDE delivered to the patient associated with the correct image quality also varies as a function of patient size: SSDE is lower in smaller patients and higher in larger patients [13] (Fig. 2B, solid black line) B Fig. 3 Illustrations show examples of size-specific dose estimates (SSDEs) as function of patient size for CT of abdomen and pelvis in different hypothetical radiology practices. A, Illustration represents process with relatively low mean patient dose (mean SSDE = 9. mgy, line) but with high degree of variation in outcome variable; this example shows process that is not in control. B, Illustration represents process that is in control with relatively high mean patient dose (mean SSDE = 13.8 mgy, line). C, Illustration represents process that is in control with relatively low mean patient dose (mean SSDE = 9. mgy, line). For SSDE to be useful for process control, a target SSDE curve must be determined quantitatively. If tube output is not adjusted on the basis of patient size, then image noise will be lower than necessary for small patients and excessive for large patients (Fig. 2A, dashed line). This corresponds to SSDE values that are excessive for small patients and insufficient for large patients (Fig. 2B, dashed line). Once known, the target values of SSDE as a function of patient size are difficult to achieve because of large differences in attenuation in patients of different sizes. This requires the operator to use a large range of tube currents, rotation times, and tube voltages as a function of patient size and to know precisely how those settings will affect image quality and dose. ATCM algorithms constitute a significant step forward in enabling more accurate optimization of patient dose, but ATCM algo- 962 AJR:24, May 21

5 CT Radiation Dose Optimization rithms do not serve as comprehensive process control systems because they lack the combination of externally reported target dose values that vary with patient size, monitoring systems to report quantitative outcomes, and feedback mechanisms to adjust input variables to consistently achieve target outcomes. Furthermore, ATCM algorithms are not necessarily designed to match target doses for a large range of patient sizes. For example, the ATCM algorithm for two of the major manufacturers is based on maintaining constant image noise. This strategy results in image noise that is excessive for small patients and lower than necessary for large patients (Fig. 2A, gray line) relative to the target image noise curve (Fig. 2A, solid black line). This corresponds to insufficient doses to smaller patients and excessive doses to larger patients (Fig. 2B, gray line) relative to the target SSDE curve (Fig. 2B, solid black line). Manufacturers typically do not publish their ATCM algorithms. This lack of information prevents most radiologists from predicting dose outcomes on the basis of scanner parameters. The number of variables involved in predicting dose outcomes results in a complex optimization problem. For example, input variables include scanner settings (e.g., tube current or ATCM setting, tube voltage, filter, focal spot size, gantry rotation speed, and pitch), patient variables (e.g., cross-sectional diameter, density, and morphology), and examination type Principle 6: Without Control Mechanisms, Processes Naturally Produce a High Degree of Variation In the absence of tightly coupled feedback mechanisms, process control does not happen naturally; deliberate development and implementation are required [11]. Consistent achievement of a target outcome variable requires a process control system. Conversely, if target outcomes are consistently obtained, adequate process controls can be assumed to be in place; if they are not consistently obtained, process controls can be assumed to be inadequate or absent. Principle 7: Minimizing Unnecessary Variation Is More Important Than Minimizing Average Dose Figure 3 illustrates three hypothetical radiology practices with identical patient populations, using identical scanners, seeking to optimize patient dose. In Figure 3A, the average SSDE is relatively low, 9. mgy, but the process is not in control. Half of the patients doses are unnecessarily high, whereas the other half are too low, which compromises image quality. In Figure 3B, the average SSDE is relatively high, 13.8 mgy, but variation is low, indicating the process is in control. In Figure 3C, both the average patient dose and variation are relatively low, with an average SSDE of 9. mgy. If further reduction of dose would produce inadequate image quality, the process represented by Figure 3C is both optimized and in control. The process represented by Figure 3B is closer to consistent optimization than the one represented by Figure 3A. Although the average dose is lower in Figure 3A, the doses in most of the examinations are either inappropriately high or inappropriately low. The practice represented by Figure 3A must go through all the necessary steps of establishing a process control system to match the outcomes of the practice represented in Figure 3C, including establishing a target outcome. The low variation achieved by the practice represented by Figure 3B implies that the practice has already established an effective process control system. This practice can simply adjust the target doses and corresponding scanner settings to achieve the outcomes in Figure 3C. Process control occurs when target outcomes are consistently achieved. Radiology practices should focus their efforts on minimizing variation in doses first and then on decreasing doses second. A decreased average dose should not be celebrated unless it is accompanied by minimal variation in dose outcomes. The focus on decreased average dose rather than on decreased variation is a major distraction in dose optimization efforts. Principle 8: Verification of Process Control Is Greatly Enhanced by Proper Framing and Graphical Display The goal of process control is to minimize unnecessary variation. To determine whether process control has been achieved, data must be displayed in a manner that allows unneces- CT of Abdomen and Pelvis A B C Fig. 4 Illustrations show different representations of same size-specific dose estimates (SSDEs) for CT examinations of abdomen and pelvis. A, Illustration presents same data shown in Figure 3A displayed as box-and-whisker plot. Upper and lower whiskers represent maximum and minimum SSDEs, respectively, and upper and lower ends of box represent 7th and 2th percentiles, respectively. Horizontal line within box represents median SSDE. Mean SSDE is 9. mgy. B, Illustration presents same range of SSDE values shown in A but corresponding to different patient sizes than in Figure 3A. Note that this process is tightly controlled, illustrating that box-and-whisker-plot cannot be used to determine whether process is well controlled because it does not illustrate effect of patient size on dose. Line shows that mean SSDE is 9. mgy. C, Illustration presents same data shown in A and B as run chart, which plots data over time. Because target SSDE varies with patient size, A and B are incapable of displaying whether process is in control because they do not include patient size. Because run chart does not illustrate effect of patient size on dose, one cannot use run chart to determine whether process is well controlled. Line shows that mean SSDE is 9. mgy AJR:24, May

6 Larson et al. sary variation to be readily identified. For example, the same data of the poorly controlled system displayed in Figure 3A are represented by a box-and-whisker plot in Figure 4A. However, the same box-and-whisker plot might correspond to doses in very tight control, as displayed in Figure 4B. Thus, the box-and-whisker plot does not allow the viewer to determine whether the process is well controlled. Figure 4C, a run chart, which plots the same SSDE values over time, is another example of poor data presentation. Because run chart does not illustrate the effect of patient size on dose, the run chart cannot be used to determine whether the process is well controlled. Framing the problem in a way that is not confounded by variables that contain appro- DLP (mgy cm) Weight (kg) A C E Diameter (cm) Age (y) Fig. Illustrations show different representations of data from same hypothetical CT examinations of abdomen and pelvis using different measures of patient size and dose estimates. A, Illustration uses water-equivalent diameter as measure of patient size; graph illustrates that process is reasonably well controlled. Line shows that mean sizespecific dose estimate (SSDE) is 9. mgy. B D, Illustrations use mean patient diameter (B), patient weight (C), and patient age (D) as measures of patient size. Because these measures do not directly correlate with patient water-equivalent diameter, graphs are misleading in that they imply greater degree of unnecessary variation in dose than is actually present. Lines show that mean SSDE is 9. mgy. E, Illustration uses dose-length product (DLP) as dose estimate. Because DLP incorporates both tube output and scanning length, graph does not differentiate appropriate variation due to differences in scanning length from unnecessary variation in dose. Line shows that mean DLP is 742 mgy cm. B D 964 AJR:24, May 21

7 CT Radiation Dose Optimization priate variation allows confirmation that unnecessary variation has been minimized. An example of confounding is an imprecise measure of patient size for CT of the abdomen and pelvis. The water-equivalent diameter is an excellent measure of patient size because it reflects the path length of the x-ray as well as tissue density, thereby providing an estimate of average beam attenuation [16]. Patient crosssectional diameter (i.e., x-ray path length) is a less accurate proxy for average attenuation because it does not account for tissue density, which is especially problematic in the thorax. Patient weight is also less accurate measure of average attenuation, and patient age is an even less accurate measure [17]. If these less accurate variables are used on the x-axis of the graph, unnecessary variation is injected into the graph. The observer cannot determine whether the process is in control. For example, a viewer can quickly see that Figure A shows a process that is relatively well controlled because water-equivalent diameter is used on the x-axis and because there is little deviation from the target SSDE curve. Figure B represents the same data but uses the less desirable mean patient diameter rather than water-equivalent diameter. The process appears slightly less well controlled than that shown in Figure A even though the graphs present data from the same examinations. Figures C and D represent the same examinations using the less desirable variables of patient weight and age, respectively. These graphs show relatively high variation, misleading the observer to conclude that the process is not well controlled. Data also can be confounded when a dose measure that is not highly correlated with image quality, such as the dose-length product (DLP), is used on the y-axis of the graph instead of CTDI vol or SSDE. For example, Figure E represents the same examinations using DLP, which incorporates both CTDI vol and scan length, instead of SSDE. Because patients of similar girths vary in torso length, the graph does not distinguish between unnecessary variation in dose and image quality versus appropriate variation in dose and image quality due to differences in scan lengths for patients of different torso lengths. Principle 9: Complexity Should Be Reduced to the Minimum and Then Increased Only as Needed Achieving process control generally requires initially reducing the number of variables to a manageable level. Although a reduction in variables may temporarily decrease the Fig. 6 Size-specific dose estimate (SSDE) as function of patient size for CT of abdomen and pelvis. Black line indicates example target SSDE curve. Dashed line represents effect of keeping tube output same for all examinations regardless of patient size on SSDE. Keeping tube current constant results in doses that are excessive for smaller patients and insufficient for larger patients. If operator adjusts settings to maintain adequate image quality for all images on basis of negative feedback from images of larger patients that contain excessive noise, radiation doses will be excessive for almost all patients (gray line). precision and flexibility in the model, it prevents the model from becoming overwhelmingly complex. Once the process is in control, desired flexibility and precision can be recovered by adding back necessary elements. For example, the control variables tube current, tube voltage, filter, focal spot size, gantry rotation speed, and pitch require an equation with six dimensions. If the uncontrollable variables patient size and examination type are added, along with the outcome variable SSDE, the problem has nine dimensions. A nine-dimensional optimization problem is both prohibitively expensive to construct and computationally difficult to solve. On the other hand, holding all of the controllable variables except tube current (or ATCM parameter setting) constant reduces the problem to four overall variables; the problem is still complex, but it is solvable. As another example, it is not practical to conduct an image quality assessment to establish a unique dose target curve for every protocol used on every scanner in a given practice. A handful of target curves will initially suffice for all protocols. As the specificity of a practice s protocols increases, the number of target curves can be added as needed. Overall, the goal is to simplify the model to include only the most relevant variables. Principle 1: Optimization and Process Control Methods Are the Same for Children as for Adults The attenuation of the x-ray beam which is a function of patient size, morphology, and density rather than age drives image quality and radiation dose. A wide variety of patient sizes exist in both children and adults. A process control system as previously outlined is needed to control doses of all patients regardless of patient age, size, or any other characteristic For example, Figure 6 shows an SSDE target curve for CT of the abdomen and pelvis relative to patient size (solid black line). If the tube current is not adjusted for patient size, the radiation delivered by the CT scanner results in too low a dose to the adult patient and too high a dose to the small child (Fig. 6, dashed line). This one-size-fits-all compromise setting provides poor image quality in larger patients. In the absence of a quantitative feedback and control system, feedback tends to be dominated by examples of poor image quality in larger patients. This feedback may encourage operators to change the settings to increase dose. This change in the settings results in excessive doses used for all patients except the largest patients (Fig. 6, gray line). In this situation, the resulting dose for the smallest children may exceed the target dose by up to a factor of 1 or more. Process control systems are needed to achieve consistent dose optimization in every practice that scans children regardless of the number of children scanned. Practices that only rarely scan pediatric patients generally have an even greater need for process control mechanisms because they have fewer small patients to learn from in a trial-anderror approach. Proposals for Regional, National, and International Efforts to Support Local CT Radiation Dose Process Control The achievement of consistent dose optimization is primarily the responsibility of those who control the CT scanner settings the operators at the local level. However, regional, national, and international strategies by professional organizations or accreditation entities along with those of CT scanner manufacturers play a AJR:24, May 21 96

8 role in advocating for consistent dose optimization. We offer a few proposals that we think would promote consistent dose optimization on a large scale. Proposal 1: The Problem Must Be Framed Properly As shown in Figure 3, review of an appropriate graph can quickly illustrate how well a practice is managing dose. Therefore, we propose the widespread adoption of this graph. Specifically, we have found that SSDE and water-equivalent diameter are the best parameters because they are based on accurate attenuation measures. Reporting data in terms of age, weight, DLP, or CTDI vol is insufficient. A practice that cannot provide these data almost certainly does not have a tightly controlled process. Conversely, a practice that has achieved consistent optimization in CT of the abdomen and pelvis as illustrated by Figure 3C almost certainly has implemented effective process control mechanisms. Proposal 2: Focus More on Consistency and Less on Dose Moving from tight control at a higher dose to tight control at a lower dose involves changing the target outcome values and adjusting CT protocols as necessary. On the other hand, moving from poor control at any dose to tight control at a low dose generally requires establishment of a comprehensive process control system, of which target outcome values are only one element. Journal editors and reviewers as well as professional societies should encourage and facilitate the publication of articles about studies that focus on decreasing unnecessary variation rather than those that focus on simply decreasing dose. Proposal 3: Practices Should Report Radiation Dose Outcomes Data in a Way That Reveals Unnecessary Variation Reporting of data in the format of Figure 3 would provide more transparency concerning how well CT doses are optimized. Identifying the need to decrease variation or to decrease target doses sets the stage for implementing the appropriate steps to improve outcomes. Just as monitoring and feedback mechanisms are needed for practices to optimize dose for individual CT examinations, similar mechanisms are also needed to ensure that practices doses are optimized. Development and implementation of process control systems require expertise and resources that may not be available to many radiology practices at the current time. If radiology practices want to tightly control dose increases, the marketplace will likely develop solutions to help them do so. Without such motivation, companies have little incentive to Larson et al. invest the resources to develop and to market solutions to these problems. In the meantime, individual radiology practices are encouraged to contact qualified medical physicists for assistance in developing and implementing process control systems within their practices. Proposal 4: CT Radiation Control Algorithms, Especially ATCM, Should Be Shared by CT Manufacturers A major factor contributing to the difficulty of optimizing CT dose is that manufacturers currently do not share their dose modulation algorithms. This lack of knowledge about the algorithms prevents operators from mathematically predicting outcome variables on the basis of scanner settings, which, in turn, prevents operators from managing dose on the basis of image quality. A network of children s hospitals working together to improve patient safety provides an admirable model [18]. They state as one of their core principles that network hospitals will NOT compete on safety [18]. Rather, they openly share performance data and experiences with each other despite the fact that many of them are competitors. In a similar way, we think that the open sharing of control algorithms of radiation-emitting medical equipment would enable better widespread radiation dose optimization and process control. Conclusion It has been more than 1 decade since the landmark articles highlighting the lack of optimization of CT radiation dose in examinations of children were published, but the problem remains unsolved. We have outlined how the use of process control systems can lead to more consistent and verifiable dose optimization at the local level. Regional, national, and international professional organizations should promote the use of these types of process control systems. We think that the adoption of these strategies will enable the problem of consistent CT radiation dose optimization to be solved on a large scale. Acknowledgment We gratefully acknowledge Donald Frush for his editorial guidance. References 1. Paterson A, Frush DP, Donnelly LF. Helical CT of the body: are settings adjusted for pediatric patients? AJR 21; 176: Brenner D, Elliston C, Hall E, Berdon W. Estimated risks of radiation-induced fatal cancer from pediatric CT. AJR 21; 176: Donnelly LF, Emery KH, Brody AS, et al. Minimizing radiation dose for pediatric body applications of single-detector helical CT: strategies at a large children s hospital. AJR 21; 176: Kalra MK, Maher MM, Toth TL, et al. Strategies for CT radiation dose optimization. Radiology 24; 23: Goske MJ, Applegate KE, Boylan J, et al. The Image Gently campaign: increasing CT radiation dose awareness through a national education and awareness program. Pediatr Radiol 28; 38: Smith-Bindman R, Lipson J, Marcus R. Radiation dose associated with common computed tomography examinations and the associated lifetime attributable risk of cancer. Arch Intern Med 29; 169: Kanal KM, Vavilala MS, Raelson C, et al. Variation in pediatric head CT imaging protocols in Washington state. J Am Coll Radiol 211; 8: Thickman D, Wang G, White J, Cutter G. National variation of technical factors in computed tomography of the head. J Comput Assist Tomogr 213; 37: Oxford Dictionaries website. Optimize. oxforddictionaries.com/us/definition/american_english/ optimize. Accessed September 29, Merriam-Webster website. Optimization. www. merriam-webster.com/dictionary/optimization. Accessed September 29, Romagnoli JA, Palazoglu A. Why process control? In: Romagnoli JA, Palazoglu A. Introduction to process control, 2nd ed. Boca Raton, FL: CRC Press, 212: Larson DB, Wang LL, Podberesky DJ, Goske MJ. System for verifiable CT radiation dose optimization based on image quality. Part I. Optimization model. Radiology 213; 269: Larson DB, Malarik RJ, Hall SM, Podberesky DJ. System for verifiable CT radiation dose optimization based on image quality. Part II. Process control system. Radiology 213; 269: Yu L, Bruesewitz MR, Thomas KB, Fletcher JG, Kofler JM, McCollough CH. Optimal tube potential for radiation dose reduction in pediatric CT: principles, clinical implementations, and pitfalls. Radio- Graphics 211; 31: American Association of Physicists in Medicine website. Boone JM, Strauss KJ, Cody DD, et al. Size-specific dose estimates (SSDE) in pediatric and adult body CT examinations. reports/rpt_24.pdf. Published February 211. Accessed September 29, Menke J. Comparison of different body size parameters for individual dose adaptation in body CT of adults. Radiology 2; 236: Kleinman PL, Strauss KJ, Zurakowski D, Buckley KS, Taylor GA. Patient size measured on CT images as a function of age at a tertiary care children s hospital. AJR 21; 194: Children s Hospitals Solutions for Patient Safety website. Children s Hospitals Solutions for Patient Safety: about us how we work. Accessed September 29, AJR:24, May 21