Objective We investigated the variability in diagnostic information inherent in computed tomography (CT) images acquired at 68 different CT units, with the selected acquisition protocols aiming to answer the same clinical question. Methods An anthropomorphic abdominal phantom with two optional rings was scanned on 68 CTsystems from 62 centres using the local clinical acquisition parameters of the portal venous phase for the detection of focal liver lesions. Low-contrast detect- ability (LCD) was assessed objectively with channelised Hotelling observer (CHO) using the receiver operating characteristic (ROC) paradigm. For each lesion size, the area under the ROC curve (AUC) was calculated and considered as a figure of merit. The volume computed tomography dose index (CTDI ) was used to indicate radiation dose exposure. vol Results The median CTDI used was 5.8 mGy, 10.5 mGy and 16.3 mGy for the small, medium and large phantoms, respectively. vol The median AUC obtained from clinical CT protocols was 0.96, 0.90 and 0.83 for the small, medium and large phantoms, respectively. Conclusions Our study used a model observer to highlight the difference in image quality levels when dealing with the same clinical question. This difference was important and increased with growing phantom size, which generated large variations in patient exposure. In the end, a standardisation initiative may be launched to ensure comparable diagnostic information for well- defined clinical questions. The image quality requirements, related to the clinical question to be answered, should be the starting point of patient dose optimisation. Key Points � Model observers enable to assess image quality objectively based on clinical tasks. � Objective image quality assessment should always include several patient sizes. � Clinical diagnostic image quality should be the starting point for patient dose optimisation. � Dose optimisation by applying DRLs only is insufficient for ensuring clinical requirements. . . . . Keywords Abdominal computed tomography Image quality Model observer Standardisation Task-based assessment Abbreviations CTDI Volume computed tomography dose index vol ATCM Automatic tube current modulation DDoG Dense difference of Gaussian AUC Area under the ROC curve DLP Dose length product BMI Body mass index DRL Diagnostic reference level CHO Channelised Hotelling observer FBP Filtered back-projection CT Computed tomography IR Iterative reconstruction LCD Low-contrast detectability ROC Receiver operating characteristic Francis R. Verdun and Sabine Schmidt contributed equally to this work. * Sabine Schmidt Introduction firstname.lastname@example.org In diagnostic radiology, computed tomography (CT) contributes Institute of Radiation Physics, Lausanne University Hospital, Rue du Grand-Pré 1, 1007 Lausanne, Switzerland to a major part of the public radiation dose exposure, which leads to public concern over potential cancer induction risks Department of Diagnostic and Interventional Radiology, Lausanne University Hospital, Rue du Bugnon 46, 1011 Lausanne, Switzerland [1–4]. Many initiatives have been launched to avoid 5204 Eur Radiol (2018) 28:5203–5210 unnecessary or useless exposure, such as Image Gently and small phantom (anterior posterior (AP) x lateral diameter: 20 x Image Wisely. The introduction of diagnostic reference levels 30 cm) and represented a thin adult with a body mass index (DRLs) allowed, to a certain extent, a reduction in the heteroge- (BMI) of 20 kg/m or a patient weight of ~50 kg. To vary the neity of the delivered dose exposure from one institution to patient’s morphology, two additional rings, one medium-sized another . However, the DRLs provided for CT examinations (2.5 cm in thickness) and one large-sized (5 cm in thickness), are generally defined as a function of an anatomical region, were added to the phantom. With these two extra rings (AP x which is certainly a limitation, as a given anatomical region lateral diameter: 25 x 35 cm, and 30 x 40cm, respectively) the may not need the same image quality depending on the specific phantom represented patients with a BMI of 26 kg/m (patient clinical question (e.g., head trauma vs. ischaemic stroke). In weight ~90 kg) and 35 kg/m (patient weight ~120 kg), re- addition, technological developments, such as the automatic spectively. A cylindrical module containing spherical lesions tube current modulation (ATCM), using dynamic beam collima- 8, 6 and 5 mm in diameter with a contrast of 20 HU relative to tion with less over-ranging have been proposed to drastically the background at 120 kVp was inserted in the centre of the reduce patient exposure . Furthermore, in the last 10 years, phantom (four spheres per diameter). It is worth mentioning iterative reconstruction (IR) techniques have become increasing- the phantom is made of various plastics that do not contain ly popular as a mechanism to reduce CT dose exposure while any high atomic number materials. Thus, the variation of con- ensuring image quality. In general, IR techniques allow drastic trast with x-ray beam energy is negligible. noise reductions while maintaining a reasonable spatial resolu- The three abdomen phantom sizes were scanned on 68 CT tion compared to traditional filtered back-projection (FBP) tech- scanners installed in 62 centres (Table 1). The four major man- niques [7–12]. ufacturers, GE Healthcare, Philips Healthcare, Siemens With the large number of IR solutions now proposed by Healthineers and Canon Medical System, were represented. multiple CT vendors, it has become crucial to systematically GE and Philips accounted for 68% of the CT units involved in evaluate the dose reduction potential and subsequent image this study. Each data acquisition was performed according to the quality for each technique. These investigations have been local clinical acquisition and image reconstruction parameters of performed using both clinical images and phantom images the portal venous phase. To maximise the performance of the [13–15]. Despite the production of subjectively better looking automatic tube current modulation (ATCM) the phantoms were images, IR techniques do not allow full recovery of the detec- always positioned at the isocentre of the CT scanners [6, 19, 20]. tion of low-contrast structures when the applied dose reduc- To provide comparable spatial resolution, the scanned and recon- tions are too high [16, 17]. Thus, when dealing with dose structed field of view were set to 320 mm, 370 mm and 420 mm reductions by means of IR, the low-contrast detectability for the small, medium and large phantoms, respectively. To en- (LCD) should systematically be investigated using task- sure statistical robustness of the results, the phantom was scanned based image quality assessment methodologies. Given the ten times on each CT unit without changing its position between large number of IR solutions proposed by CT manufacturers, the different acquisitions. This resulted in 40 images containing a such image quality assessments should be performed using signal and 90 images that contained only noise [19, 20]. phantoms and objective quantitative methods. The US Food and Drug Administration recommends the use of mathemati- Image quality assessment cal model observers as surrogates to human observers . The outcomes provide image quality metrics measured on A channelised Hotelling observer (CHO) with dense differ- phantoms as image quality indicators, while the volume CT ence of Gaussian (DDoG) channels was used to assess the dose index (CTDI ) or the dose length product (DLP) are vol LCD. The chosen channels are known to represent the spatial used as patient exposure indicators. selectivity behaviour of the human primary visual cortex (V1). The aim of the present study was to investigate the variabil- With this channelisation process of the images (each image is ity of patient exposure and the CT image quality provided by a passed through the DDoG channels) the model observer is large number of centres when evaluating the presence or ab- considered anthropomorphic. In our study we used ten dense sence of focal abdominal lesions in phantoms of different sizes. differences of Gaussian channels, since this value has been known to be sufficient for mimicking well the human detec- tion in such a simple task [21–23]. Materials and methods Each channel is defined by Eq. 1: 2 2 Abdomen phantom and image acquisition 1 ρ 1 ρ − − 2 Qσ 2 σ j j C ðÞ ρ ¼ e −e ð1Þ An anthropomorphic abdomen phantom (QRM 401) simulat- ing three types of attenuations produced by an adult patient Where ρ was the spatial frequency, σ the standard deviation (Fig. 1) was used for this study. This phantom was called the of each channel and Q the filter bandwidth. Each standard Eur Radiol (2018) 28:5203–5210 5205 Fig. 1 CT images of the anthropomorphic abdomen phantom. From left to right: small, medium and large phantom th j-1 deviation of the j channel value was given by σ = σ α .As 40 signal-present images were used to compute 40 decision j 0 previously described , the following variable settings variables for the signal-present image category. In Eq. 2,T is were: σ =0.005, α =1.4 and Q = 1.67. the transpose operator; n represents the image category, As in other CHO models, the CHO model with DDoG chan- signal-absent or signal-present, s represents the lesion size, nels computes a decision variable, λ, from the dot product be- and i the image number. The template (w) takes into account tweenachannelisedimage vandatemplatew,asseeninEq. 2: the statistical knowledge of noise by computing the covari- ance matrix K from 90 channelised images containing no sig- λ ¼ w v ð2Þ n;s;i n;s;i nal. The template w also takes into account the signal by computing a theoretical signal that represents the different The decision variable can be seen as a grade given by the lesions sizes, as seen in Eq. 3: model to the images. The higher the decision variable, the higher the probability of the presence of a signal in the image. −1 w ¼ K v ð3Þ For this study, 90 signal-absent images were used to compute v=n Theo 90 decision variables for the signal-absent image category and Only the phantom sphere measuring 5 mm in diameter was Table 1 The 68 CT scanners involved in this study used in this study, as preliminary measurements showed that this size was the most critical parameter in terms of area under CT scanner Number the receiver operating characteristic (ROC) curve (AUC). To GE BrightSpeed S 1 avoid the need for acquiring too many images, we used a Discovery CT 750 HD 2 theoretical signal of 5 mm in diameter (v ) instead of a Theo LightSpeed VCT 7 mean signal (the mean signal is defined as the difference be- LightSpeed16 1 tween the mean of images containing a signal and the mean of Optima CT520 2 images that contained only noise) for the signal template. It Optima CT580 1 was created using a simulated 2D Gaussian curve with a full width at half maximum of 5 mm. In addition, this method Optima CT660 7 Revolution 1 avoids overfitting of the data. Using the decision variable distribution for the signal-absent category (90 decision vari- Revolution EVO 2 ables) and the signal-present category (40 decision variables), Philips Brilliance 40 1 a ROC study was then computed and characterised by its the Brilliance 64 6 AUC. The latter was a surrogate to assess image quality. The iCT 256 4 average and standard deviation of the AUC were estimated Ingenuity Core 128 4 using a bootstrap method . In practice, the model per- Ingenuity CT 4 formed 500 ROC experiments for each category. No internal Ingenuity Flex 2 noise was added to improve the match between the model IQon - Spectral CT 1 observer and the human observer performance . Siemens Perspective 1 The displayed CTDI was used as a figure of merit for the vol Sensation 64 2 patient dose exposure. We did not systematically measure SOMATOM Definition AS 2 these displayed CTDI values, since our national legislation vol SOMATOM Force 1 requires a conformity check every 3 months. This conformity Canon Medical System Activion16 4 check must be performed by the manufacturer and ensures that Aquilion 6 the displayed and measured CTDI values are equal with a vol Aquilion PRIME 6 permitted difference of ≤ 20%. 5206 Eur Radiol (2018) 28:5203–5210 As the four major manufacturers were represented in this varied from 2.3 to 18.7 mGy with a median of 5.8 mGy and study, two different types of ATCM were involved in the data third quartile of 7.5 mGy. For the medium-sized phantom, the acquisitions: one ATCM for which the user had to choose a CTDI varied from 5.5 to 34.2 mGy with a median of vol noise level (GE and Canon Medical System) and another 10.5 mGy and third quartile of 13.4 mGy. For the largest ATCM for which the user only introduced a reference image phantom, the CTDI varied from 8.6 to 34.2 mGy with a vol load (in mAs, Philips and Siemens). To study the impact of the median of 16.3 mGy and a third quartile of 20.9 mGy (Fig. 2). ATCM method on the image quality when the patient size varied, the correlation between the AUC of the different phan- Image quality for abdomen CT protocols tom sizes was calculated using the Pearson coefficient (r). Correlation is an effect size and so we can verbally describe Small phantom the strength of the correlation using the guide that Evans et al. suggest for the absolute value of r : 0.00–0.19 ‘very For the small phantom (Fig. 3), despite the use of relatively low weak’, 0.20–0.39 ‘weak’, 0.40–0.59 ‘moderate’, 0.60–0.79 CTDI values (median CTDI 5.8 mGy) compared to the vol vol ‘strong’ and 0.80–1.0 ‘very strong’. national DRLs provided for abdominal CT protocols (15 mGy) , excellent image quality (AUC ≥0.95) was obtained for most of the centres (Fig. 3). Indeed, the median AUC Results was 0.96 and the third quartile was 0.97, with several centres achieving an AUC >0.99. Only three centres had an AUC CTDI in terms of patient size <0.85. Two protocols (centre a, see Fig. 3) used too low a dose vol level for the CT and therefore onlyachievedanimagequality As expected, the CTDI increased with growing phantom with an AUC inferior to 0.85. The slice thickness used for vol size. Due to the selection of the locally implemented clinical centres b and c was equal to 5 mm and was thus not compatible CT protocol for each unit, the CTDI varied significantly for for accurately detecting a lesion 5 mm in diameter due to partial vol a given phantom size. For the small phantom, the CTDI volume effect, even if the dose was high. vol Fig. 2 CTDI obtained for the vol three phantom sizes as a function of the abdominal clinical CT protocol settings. The black line in the middle of the coloured rectangles represents the median. The bottom edge of the rectangle corresponds to the first quartile and the top edge to the third quartile. The bottom line represents the fifth percentile and the top line the 95th percentile. The red dots outside these two lines are outliers Eur Radiol (2018) 28:5203–5210 5207 Fig. 3 AUC as a function of CTDI for the small-sized vol phantom Medium-sized phantom significantly low in two protocols (centre a, see Fig. 5)due to partial volume effects associated with the use of one of the For the medium-sized phantom (Fig. 4), the median CTDI oldest CT units. vol was 10.2 mGy with a median AUC of 0.90 (third quartile 0.94). With this phantom size, almost a quarter of the centres had an AUC <0.85. The dose levels were relatively or significantly Image quality correlation high in three protocols (centres a and b, see Fig. 4)due to a suboptimal reconstruction slice thickness of 5 mm. These im- For the first type of ATCM for which the user had to choose a ages were obtained on the oldest CT scanners included in this noise level, the correlation (r) of AUC values between the study, which had been introduced on the market in 2007–2009. different phantom sizes varied from 0.33 between small and large to 0.49 between medium and large (Table 2). For the Large phantom second type of ATCM, for which the user only introduced a reference image load (in mAs), the correlation (r) of AUC For the large phantom (Fig. 5), the median CTDI was 16.1 values between the different phantom sizes varied from 0.40 vol mGy, with a median AUC of 0.83 (third quartile 0.87). With between small and large to 0.58 between medium and large this phantom size, almost a quarter of the centres had an AUC (Table 3). Thus, there was a weak to moderate positive corre- <0.80. As with the other phantom sizes, the image quality was lation between the level of CT image quality and phantom size. Fig. 4 AUC as a function of CTDI for the medium-sized vol phantom 5208 Eur Radiol (2018) 28:5203–5210 Fig. 5 AUC as a function of CTDI for the large-sized vol phantom Discussion image quality approach was relatively homogeneous among all centres with the small phantom; yet when scanning larger To justify a given CT examination, the imaging protocol phantom sizes, the image quality tended to vary between the should be adapted to answer one or several specific clinical different centres. However, for a given clinical question, there questions corresponding to the indication of the examination. is large variability in dose especially when dealing with large In this study, LCD was assessed using the portal venous phase patients, due to different ATCM settings. Therefore, it is im- for the detection of focal liver lesions. We thus created a task- portant to use different phantom sizes to assess the behaviour based approach by means of an anthropomorphic CHO model of image quality when varying the exposure and patient size. observer applied on three abdominal phantom sizes. Using All these variations in terms of dose and patient exposure this methodology, we showed that some clinical protocols provide a weak-to-moderate correlation between the outcomes applied by several CT units across the country allowed pro- of image quality when varying the phantom sizes (Tables 2 duction of a high image quality level in a low-dose range. and 3). Indeed, the local practice (especially the setting of the Nevertheless, some protocols applied on similar or different maximum mA during the image acquisition) and/or the limi- CT units produced lower AUC levels despite the use of a tation of the x-ray tube power influences the feasibility of comparable dose range. This was sometimes due to the use using an equivalent image quality level for each patient’smor- of sub-optimal ATCM settings or sub-optimal reconstructed phology. In such cases, the acquisition parameters had to be slice thicknesses (Figs. 4 and 5). Further investigations are adjusted accordingly, but the optimisation of acquisition pa- needed to fully understand if a better outcome could be ob- rameters and patient dose was not reached in most of the tained from the CT units that used high dose levels. centres included in our study. When the national/ Taylor et al. have recently shown that samples of approxi- international DRLs are used to optimise the clinical CT pro- mately 300 CT examinations of patients with a body weight in tocols, image quality is currently not considered. Our study the range of 67–73 kg are necessary to create reliable DRLs demonstrates that the DRL concept, when dealing with . For many years, broad application of the DRL concept patient-dose optimisation, has reached its limits since compa- has allowed national homogenisation of patient exposure in rable dose levels applied on different CT units provided highly practice for specific anatomical regions [30, 31]. In our study, variable diagnostic image quality levels. The optimisation of it appears that the image quality obtained using a task-based Table 3 Correlation matrix based on the area under the ROC curve Table 2 Correlation matrix based on the area under the ROC curve (AUC) obtained with automatic tube current modulation (ATCM) that (AUC) obtained with automatic tube current modulation (ATCM) that maintained a constant level of overall diagnostic quality for all patient maintained absolute noise levels close to target values sizes relating to a reference image Size S M L Size S M L S1 S1 M0.39 1 M0.49 1 L 0.33 0.49 1 L 0.40 0.58 1 Eur Radiol (2018) 28:5203–5210 5209 clinical CT protocols should rather ensure that a comparable images, which possibly impair diagnostic quality. The aim of diagnostic information, for well-defined clinical questions, is the previously introduced DRL concept was to reduce the vari- obtained at the lowest achievable dose on different CT units. ability of patients’ dose exposure, but as we have shown here, it Trying to only reach comparable patient-dose exposures on is insufficient to ensure comparable image quality on different different CT units has a limited potential in the framework of CT machines. We should now define a set of task-based image patient exposure and image quality optimisation. quality criteria related to well-defined clinical indications and Therefore, in the end, the challenge is to establish a link work towards the standardisation of image quality requirements. between the different clinical tasks and the surrogates used to To establish these requirements, it is important to define the assess image quality . These task-based image quality critical target to be detected and determine the AUC level to be criteria (for example the LCD requirements) should be initiat- used for standardised phantoms. ed for a few morphology types and a standardisation process Funding The authors state that this work has not received any funding. concerning image quality requirements as a function of the common clinical indications . Indeed, IR provide images Compliance with ethical standards that look satisfying in a larger dose range than with the stan- dard FBP reconstruction algorithm, since the amount of noise Guarantor The scientific guarantor of this publication is Francis R does not alert the radiologist. However, ‘blind’ dose reduc- Verdun. tions could be done that may impair the diagnostic image quality. Therefore one should ensure the presence of the nec- Conflict of interest The authors of this manuscript declare no relation- essary amount of information by objectively and quantitative- ships with any companies whose products or services may be related to the subject matter of the article. ly evaluating image quality to fully benefit from the potential of dose reductions provided by the use of IR . Statistics and biometry Damien Racine has significant statistical This study has limitations. First, we did not consider intrave- expertise. nous contrast medium administration, which was an important No complex statistical methods were necessary for this paper. part of the optimisation protocol and has an undeniable impact on contrast enhancement of organs and vessels and, thus, on the Informed consent Written informed consent was not required for this study because we performed a phantom study. diagnostic yield; yet this method enables up to 50% dose reduc- tion . Second, due to the use of a homogeneous phantom Ethical approval Institutional Review Board approval was not required that simplifies the objective assessment of image quality, it is far because this was a phantom study. from the current clinical situation with heterogeneous back- grounds. However, we still consider this as a good starting point Methodology for the optimisation process. Third, the largest phantom was � prospective � experimental made of muscle tissue only, which is not fully representative � multicentre study of obese patients. This increased the performance requirements for the investigated CT units. The largest rings added to the core Open Access This article is distributed under the terms of the Creative of the phantom should ideally have contained both muscle tis- Commons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, sue and fat. Four, the material composing the lesion inserted distribution, and reproduction in any medium, provided you give appro- inside the phantom would have been more clinically relevant priate credit to the original author(s) and the source, provide a link to the if it contained a high Z material, such as iodine, when we in- Creative Commons license, and indicate if changes were made. vestigated the portal venous phase in an abdominal CT protocol, a de facto injected phase. Finally, several other data acquisition and image reconstruction parameters, such as ATCM, tube volt- References age, slice thickness and IR level, influence the overall CT image quality. However, our primary aim was rather to highlight the 1. Perez A-F, Devic C, Colin C, Foray N (2015) The low doses of radi- wide spread of image quality and dose for a single clinical task. ation: towards a new reading of the risk assessment. Bull Cancer (Paris) 102:527–538. https://doi.org/10.1016/j.bulcan.2015.03.019 Assessing the confounding effects of all these parameters sepa- 2. Mathews JD, Forsythe AV, Brady Z et al (2013) Cancer risk in 680,000 rately would require further dedicated studies and was beyond people exposed to computed tomography scans in childhood or ado- the scope of the present investigation. lescence: data linkage study of 11 million Australians. BMJ 346:f2360 In summary, we developed an objective, quantitative and ro- 3. Hricak H, Brenner DJ, Adelstein SJ et al (2011) Managing radiation use in medical imaging: a multifaceted challenge. Radiology 258: bust method for benchmarking abdominal CT protocols using 889–905. https://doi.org/10.1148/radiol.10101157 three different phantom sizes and evaluating 68 different CT units 4. Brenner DJ, Shuryak I, Einstein AJ (2011) Impact of reduced pa- from 62 centres across the country. Our study demonstrates that tient life expectancy on potential cancer risks from radiologic im- radiologists, radiographers and medical physicists, must collabo- aging. Radiology 261:193–198. https://doi.org/10.1148/radiol. rate to ensure that dose reductions do not lead to sub-optimal 11102452 5210 Eur Radiol (2018) 28:5203–5210 5. Rosenstein M (2008) Diagnostic reference levels for medical expo- 19. Habibzadeh MA, Ay MR, Asl ARK et al (2012) Impact of miscentering on patient dose and image noise in x-ray CT imaging: sure of patients: ICRP guidance and related ICRU quantities. Health Phys 95:528–534. https://doi.org/10.1097/01.HP. phantom and clinical studies. Phys Med 28:191–199. https://doi. 0000326331.35187.63 org/10.1016/j.ejmp.2011.06.002 6. Schindera ST, Nelson RC, Yoshizumi T et al (2009) Effect of auto- 20. Racine D, Ba AH, Ott JG et al (2016) Objective assessment of low matic tube current modulation on radiation dose and image quality contrast detectability in computed tomography with Channelized for low tube voltage multidetector row CT angiography: phantom Hotelling Observer. Phys Med 32:76–83. https://doi.org/10.1016/ study. Acad Radiol 16:997–1002. https://doi.org/10.1016/j.acra. j.ejmp.2015.09.011 2009.02.021 21. Abbey CK, Barrett HH (2001) Human- and model-observer perfor- 7. Hara AK, Paden RG, Silva AC et al (2009) Iterative reconstruction mance in ramp-spectrum noise: effects of regularization and object technique for reducing body radiation dose at CT: feasibility study. variability. J Opt Soc Am A Opt Image Sci Vis 18:473–488. https:// AJR Am J Roentgenol 193:764–771. https://doi.org/10.2214/AJR. doi.org/10.1364/JOSAA.18.000473 09.2397 22. Tseng H-W, Fan J, Kupinski MA et al (2014) Assessing image 8. Omoumi P, Becce F, Ott JG et al (2015) Optimization of radiation quality and dose reduction of a new x-ray computed tomography dose and image Quality in musculoskeletal CT: emphasis on itera- iterative reconstruction algorithm using model observers. Med Phys tive reconstruction techniques (Part 1). Semin Musculoskelet 41:071910. https://doi.org/10.1118/1.4881143 Radiol 19:415–421. https://doi.org/10.1055/s-0035-1569255 23. Leng S, Yu L, Zhang Y et al (2013) Correlation between model 9. Silva AC, Lawder HJ, Hara A et al (2010) Innovations in CT dose observer and human observer performance in CT imaging when reduction strategy: application of the adaptive statistical iterative lesion location is uncertain. Med Phys 40:081908. https://doi.org/ reconstruction algorithm. AJR Am J Roentgenol 194:191–199. 10.1118/1.4812430 https://doi.org/10.2214/AJR.09.2953 24. Efron B, Tibshirani RJ (1994) An Introduction to the Bootstrap. 10. Sagara Y, Hara AK, Pavlicek W et al (2010) Abdominal CT: com- CRC Press, New York, Washington DC parison of low-dose CTwith adaptive statistical iterative reconstruc- 25. Brankov JG (2013) Evaluation of channelized hotelling observer tion and routine-dose CT with filtered back projection in 53 pa- with internal-noise model in a train-test paradigm for cardiac tients. AJR Am J Roentgenol 195:713–719. https://doi.org/10. SPECT defect detection. Phys Med Biol 58:7159–7182. https:// 2214/AJR.09.2989 doi.org/10.1088/0031-9155/58/20/7159 11. Pontana F, Duhamel A, Pagniez J et al (2011) Chest computed tomog- 26. Evans JD (1996) Straightforward statistics for the behavioral sci- raphy using iterative reconstruction vs filtered back projection (Part 2): ences. Brooks/Cole Pub. Co., Pacific Grove image quality of low-dose CT examinations in 80 patients. Eur Radiol 27. Office fédéral de la santé publique OFSP (2010) Notice R-06-06 : 21:636–643. https://doi.org/10.1007/s00330-010-1991-4 Niveaux de référence diagnostiques en tomodensitométrie. 12. Funama Y, Taguchi K, Utsunomiya D et al (2011) Combination of a Available via: https://www.bag.admin.ch/bag/fr/home/themen/ low-tube-voltage technique with hybrid iterative reconstruction mensch-gesundheit/strahlung-radioaktivitaet-schall/bewilligungen- (iDose) algorithm at coronary computed tomographic angiography. aufsicht-im-strahlenschutz/informationen-fuer-medizinische- J Comput Assist Tomogr 35:480–485. https://doi.org/10.1097/ betriebe/diagnostische-referenzwerte-im-strahlenschutz.html. RCT.0b013e31821fee94 Accessed 17 Mar 2018 13. Schindera ST, Odedra D, Raza SA et al (2013) Iterative reconstruc- 28. Youngstrom EA (2014) A primer on receiver operating character- tion algorithm for CT: can radiation dose be decreased while low- istic analysis and diagnostic efficiency statistics for pediatric psy- contrast detectability is preserved? Radiology 269:511–518. https:// chology: we are ready to ROC. J Pediatr Psychol 39:204–221. doi.org/10.1148/radiol.13122349 https://doi.org/10.1093/jpepsy/jst062 14. Flicek KT, Hara AK, Silva AC et al (2010) Reducing the radiation dose for CT colonography using adaptive statistical iterative recon- 29. Taylor S, Van Muylem A, Howarth N et al (2017) CT dose survey struction: A pilot study. AJR Am J Roentgenol 195:126–131. in adults: what sample size for what precision? Eur Radiol 27:365– https://doi.org/10.2214/AJR.09.3855 373. https://doi.org/10.1007/s00330-016-4333-3 15. Marin D, Nelson RC, Schindera ST et al (2010) Low-tube-voltage, 30. Schegerer AA, Nagel H-D, Stamm G et al (2017) Current CT practice high-tube-current multidetector abdominal CT: improved image in Germany: Results and implications of a nationwide survey. Eur J quality and decreased radiation dose with adaptive statistical itera- Radiol 90:114–128. https://doi.org/10.1016/j.ejrad.2017.02.021 tive reconstruction algorithm–initial clinical experience. Radiology 31. Journy NMY, Dreuil S, Boddaert N et al (2017) Individual radiation 254:145–153. https://doi.org/10.1148/radiol.09090094 exposure from computed tomography: a survey of paediatric prac- 16. Dobeli KL, Lewis SJ, Meikle SR et al (2013) Noise-reducing algo- tice in French university hospitals, 2010-2013. Eur Radiol. https:// rithms do not necessarily provide superior dose optimisation for doi.org/10.1007/s00330-017-5001-y hepatic lesion detection with multidetector CT. Br J Radiol 86: 32. Verdun FR, Racine D, Ott JG et al (2015) Image quality in CT: 20120500. https://doi.org/10.1259/bjr.20120500 From physical measurements to model observers. Phys Med 31: 17. Baker ME, Dong F, Primak A et al (2012) Contrast-to-noise ratio 823–843. https://doi.org/10.1016/j.ejmp.2015.08.007 and low-contrast object resolution on full- and low-dose MDCT: 33. Racine D, Viry A, Becce F et al (2017) Objective comparison of SAFIRE versus filtered back projection in a low-contrast object high-contrast spatial resolution and low-contrast detectability for phantom and in the liver. AJR Am J Roentgenol 199:8–18. various clinical protocols on multiple CT scanners. Med Phys 44: https://doi.org/10.2214/AJR.11.7421 e153–e163. https://doi.org/10.1002/mp.12224 18. Computed Tomography Image Quality (CTIQ): Low-Contrast 34. ICRP 2017 Diagnostic reference levels in medical imaging. ICRP Detectability (LCD) Assessment When Using Dose Reduction Publication 135. Ann. ICRP 46(1). Technology - NEMA. Available via: https://www.nema.org/ 35. Kalender WA, Deak P, Kellermeier M et al (2009) Application- and Standards/Pages/Computed-Tomography-Image-Quality-CTIQ- patient size-dependent optimization of x-ray spectra for CT. Med Low-Contrast-Detectability-LCD-Assessment-When-Using-Dose- Phys 36:993–1007. https://doi.org/10.1118/1.3075901 Reduction-Technology.aspx. Accessed 7 Mar 2018
European Radiology – Springer Journals
Published: Jun 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud