Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A simple method for low-contrast detectability, image quality and dose optimisation with CT iterative reconstruction algorithms and model observers

A simple method for low-contrast detectability, image quality and dose optimisation with CT... Background: The aim of this work was to evaluate detection of low-contrast objects and image quality in computed tomography (CT) phantom images acquired at different tube loadings (i.e. mAs) and reconstructed with different algorithms, in order to find appropriate settings to reduce the dose to the patient without any image detriment. Methods: Images of supraslice low-contrast objects of a CT phantom were acquired using different mAs values. Images were reconstructed using filtered back projection (FBP), hybrid and iterative model-based methods. Image quality parameters were evaluated in terms of modulation transfer function; noise, and uniformity using two software resources. For the definition of low-contrast detectability, studies based on both human (i.e. four- alternative forced-choice test) and model observers were performed across the various images. Results: Compared to FBP, image quality parameters were improved by using iterative reconstruction (IR) algorithms. In particular, IR model-based methods provided a 60% noise reduction and a 70% dose reduction, preserving image quality and low-contrast detectability for human radiological evaluation. According to the model observer, the diameters of the minimum detectable detail were around 2 mm (up to 100 mAs). Below 100 mAs, the model observer was unable to provide a result. Conclusion: IR methods improve CT protocol quality, providing a potential dose reduction while maintaining a good image detectability. Model observer can in principle be useful to assist human performance in CT low- contrast detection tasks and in dose optimisation. Keywords: Low-contrast object detection, Computed tomography, Image quality, Model-based iterative reconstruction, Model observer * Correspondence: Luca.Bellesi@eoc.ch Medical Physics Unit, Ente Ospedaliero Cantonale, Ospedale San Giovanni, Bellinzona 6500, Switzerland Full list of author information is available at the end of the article © The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Bellesi et al. European Radiology Experimental (2017) 1:18 Page 2 of 10 Key points tasks such as patient classification or estimation of vol- ume and/or other characteristics of tumours. However, Detection of low-contrast objects and image quality studies based on human observers are resource- in CT phantom images were evaluated demanding and involve a significant variability of intra- Different tube loadings and image reconstruction observer and interobserver performance. Being able to methods were tested extract as much statistical information as possible from Iterative reconstruction in CT provided significant the available images, computational model observers can mAs reduction without image detriment be used as convenient and objective surrogates of hu- Model observers are useful for parameter man beings to predict and/or define their expected per- optimisation in CT dose reduction tasks formance [6, 8]. In medical imaging, model observers were developed Background to study how system parameters affect signal detection The overall per caput mean effective dose per year to [9], taking into account physical factors that degrade the population in European countries, due to X-ray pro- image quality. They are also useful to evaluate and opti- cedures, is about 1.05 mSv. Computed tomography mise software systems, such as image reconstruction or (CT), which is a key medical imaging modality within processing methods, both to study and predict their ef- clinical diagnostic applications, contributes, on average, fects on human-observer performance [10–12]. to 57% of this dose (range 5.31–83.1%) [1], with a mean The purpose of this work was to evaluate image qual- value of 7.44 mSv [2]. In Switzerland, through 2013 the ity and low-contrast object detectability in CT phantom number of CT exams was 117 per 1000 inhabitants, with images acquired at different tube loadings (i.e. mAs) and an average dose per exam of 8.54 mSv. CT alone con- reconstructed with different algorithms in order to re- tributed to about 70% of the collective dose, with an duce mAs and consequently CT dose, with respect to a average annual effective dose of 1 mSv per inhabitant standard reference value, without detriment of the im- [3]. In light of these data, reduction of radiation dose ages. Model observer performance in terms of minimum from CT has become an essential field of study. In the diameter of the detectable low-contrast details was also last years, the advent of faster microprocessors, CT evaluated. iterative reconstruction (IR) methods were launched to integrate already existing algorithms such as filtered back projection (FBP) as a way to reduce patient Methods radiation exposure while maintaining high-contrast CT phantom image acquisition spatial resolution. In this study, a Catphan 504 phantom (The Phantom iDose4 (Philips iDose4™ system, Philips Healthcare, Laboratory, Salem, NY, USA) was used to perform all Cleveland, OH, USA) belongs to the first generation of image quality tests. It is a cylindrical phantom of 20-cm iterative hybrid reconstruction algorithms, which com- length and 20-cm diameter, containing several test mod- bine FBP and IR algorithms [4]. On the contrary, Itera- ules: a solid image uniformity module (CTP486), a 21- tive Model Reconstruction (IMR; Philips Healthcare, line pair and point source high resolution module Cleveland, OH, USA) is an advanced knowledge-based (CTP528), a module for slice width, sensitometry and algorithm that models the process of physical data ac- pixel size evaluation (CTP401), and a low-contrast mod- quisition through the iterative minimisation of the differ- ule (CTP515). In particular, the low contrast CTP515 ences between image raw data and the estimated image module contains two sub-regions: the supraslice region [5]. IMR differs from FBP methods in that the recon- with three groups of low-contrast objects, consisting of struction becomes an optimisation process that takes nine circular objects with diameters in the range of 2– into account data statistics, image statistics, and system 15 mm and contrast of 0.3%, 0.5% and 1.0%, respectively, models. IMR levels differ by number of processing cycles and a subslice region with three groups of four circular which increase concurrently with increasing levels. The objects each (diameters in the range of 3–9 mm, con- main difficulty is to preserve an adequate diagnostic trast of 1.0%). image quality reducing exposure mAs values and there- CT scans of the Catphan phantom were acquired with fore reducing the dose to the patient [6, 7]. a Philips 256 iCT multi-slice CT unit (Brilliance iCT, Medical image quality assessment involves both a sci- Philips, Best, The Netherlands), aligning the phantom entific and philosophical approach to define how ‘well’ main axis with the axis of rotation of the scanner (z- specific information of interest is obtained from images. axis). Acquisitions were performed selecting a beam col- One method to define medical image quality is called limation of 128 × 0.625 mm, scanning field of view of ‘statistical task-based assessment approach’ and consists 214 mm, scan length of 213 mm, helical acquisition with of the evaluation of the observer performance during 0.976 pitch factor, tube voltage of 100 kVp, scan time Bellesi et al. European Radiology Experimental (2017) 1:18 Page 3 of 10 2.5 s, rotation time 0.5 s, slice thickness 3 mm and zoom Catphan QA were used and the obtained results were 100%. compared. The product of tube current and exposure time per ro- tation (i.e. tube load) was in the range of 15–300 mAs. Noise Image reconstruction was performed using the following Noise was characterised on the images of the Catphan reconstruction algorithms: FBP; iDose with levels in the CTP 485 uniform module as the standard deviation of range of 1–6; and IMR with levels in the range of 1–3. pixel values within a square region of interest (ROI) lo- cated at the centre of the phantom module. Software for image analysis Uniformity For image quality analysis, one software product was Uniformity was calculated in the homogeneous region of used and compared, in terms of numerical results, with the CP486 module as the deviation in CT numbers of an advanced automated quality assurance software ser- the mean value of upper, right, lower, and left circular vice available on the web. For the definition of low- off-centre ROIs from the mean value of a ROI placed at contrast detectability, however, studies based on both the centre of the image of the phantom. Position and di- human and model observers were performed. mension of the ROIs could change between the two soft- CT image quality parameters were evaluated with two ware products. In any case, the closer to unity was the different software resources, CTQA_cp and Catphan result, the more uniform was the image. QA (Image Owl, Inc., Greenwich, NY, USA), in order to cross-check the obtained results and validate CTQA_cp High-contrast spatial resolution results with a reference. CTQA_cp (version 0.3.1) is a MTF was calculated as the Fourier transform of the freeware software package developed to aid CT quality point spread function of a region of interest centred on assurance programs and able to automatically produce the lower bead point object of the Catphan CTP 528- image quality reports. In particular, the following param- point source module. eters are analysed with CTQA_cp: slice thickness, pixel size, CT number linearity, uniformity, homogeneity, Low-contrast spatial resolution image noise across detector rows, and modulation trans- As described below, empirical and computational fer function (MTF). A low-contrast resolution analysis methods were evaluated in this study to quantify low- tool of the Catphan CP515 module based on a model contrast spatial resolution. observer is also available. Catphan QA executes an automatic analysis of CT Evaluation with the four-alternative forced test Catphan images and produces an image quality report. Four-alternative forced-choice (4-AFC) [13] test was ex- The following CT imaging performance parameters are ecuted to evaluate low-contrast spatial resolution by five evaluated: sensitometry; MTF (i.e. from beads and wires radiologists with at least 15 years of experience in clin- analysis); critical frequency; CT linearity; phantom pos- ical CT and four experienced radiology technicians [14]. ition; rotation and yaw; slice width; and contrast Observers were trained on all technical aspects and ob- detectability. jectives of the study and frontal training was performed Catphan QA also includes a contrast diameter detail through examples before the test. function that returns dimensions of the smallest detect- The 4-AFC test was performed in a darkened room able target for each of the three contrast values and was with a constant level of low ambient lighting and images used in order to obtain image quality low-contrast were presented on a DICOM-calibrated megapixel information. colour LCD screen (Radiforce RX320 LCD, EIZO Cor- Image quality parameters were evaluated with poration) with a native resolution of 1536 × 2048. Initial CTQA_cp and Catphan QA and on the phantom images window and level values of 100 and 1090 were sug- acquired with the different CT mAs values and recon- gested, respectively, but observers were free to modify structed with different reconstruction algorithms. them if necessary. No limitations on viewing distance and time were set and no reference image was provided Physical metrics quantification with CTQA_cp and before the start. Each human observer analysed 543 Catphan QA stacks of four images containing either just background For each adopted scanning protocol (i.e. different mAs) or the 6-mm and 7-mm diameter objects (1% contrast) and reconstruction algorithms, noise, uniformity, and of the low-contrast supraslice region of the Catphan high-contrast spatial resolution were evaluated in order phantom. To create the stack of images for the 4-AFC to quantify how the different CT acquisition parameters test, dedicated macros were created using the freeware impact on the physical metrics. Both CTQA_cp and software ImageJ (National Institute of Health Image, Bellesi et al. European Radiology Experimental (2017) 1:18 Page 4 of 10 Bethesda, MD, USA) that automatically executes the fol- lowing steps: (1) extracts samples of the low-contrast objects (diameters 6–7 mm, 1% contrast) or of the back- ground from low-contrast Catphan module (Fig. 1); (2) generates a series of images each containing four quad- rants with low-contrast circular objects or background, randomly chosen from Catphan images acquired at dif- ferent experimental conditions (i.e. mAs in the range of 30–300) and reconstructed by means of FBP, iDOSE (i.e. levels 1–6) and IMR (i.e. levels 1–3). Sixteen images for each CT protocol modality were overall selected and randomly arranged over the stack of 543 images. One example of images is provided in Fig. 2, each quadrant possibly representing the particular of the Catphan CTP515 low-contrast module shown in Fig. 1; (3) cre- ates a stack of 543 images in a single DICOM image se- quence that was loaded to a picture archiving and communication system PACS (Philips IntelliSpace PACS Enterprise 4.4.532.1, Philips Healthcare Informatics, Inc., Foster City, CA, USA) for further evaluation by the observers. Fig. 2 Example of image for the 4-AFC test. In this case, low-contrast In this study, observers had to identify the presence of objects were in quadrants a, b and c while quadrant d is empty one or more quadrants with low-contrast lesions and to indicate their position within the image. In principle, in each one of the 543 images, low-contrast objects were in Computational evaluation none, one, two, three, or any quadrant. The percentage The computer model observer provided with CTQA_cp of correct answers given by each observer subjected to was used to define low-contrast detectability on the Cat- the 4-AFC experiment was analysed and evaluated. phan low-contrast supraslice images acquired at differ- Inter-CT protocol modality (i.e. each combination of mAs ent experimental conditions (i.e. mAs in the range of and reconstruction algorithms) analysis was performed. 15–300) and reconstructed by means of FBP, iDOSE (i.e. levels 1–6) and IMR (i.e. levels 1–3). According to the method, which is exhaustively described by Hernandez- Giron et al. [6], output of the software system is the smallest ‘visible’ object size at 1%, 0.5%, and 0.1% con- trast. Only objects with 1% contrast were evaluated be- cause 0.5% and 0.1% objects were often not visible during first visual evaluations after phantom CT acquisi- tions. Catphan QA also includes a function for low- contrast diameter detail evaluation, which returns dimensions of the smallest detectable target for each of the three contrast values. This function is not based on a model observer-based statistical approach, but it is re- lated on the use of an algorithm for image analysis. Results Physical metrics quantification with CTQA_cp and Catphan QA Noise and uniformity evaluations are provided in Figs. 3 and 4, respectively; the high-contrast spatial resolutions for 50% and 10% MTF are given in Table 1. Numerical results of both software systems resulted to be compar- able in terms of noise analysis, whereas a difference arose for uniformity. High-contrast spatial resolutions Fig. 1 Image of the low-contrast module of the Catphan phantom evaluated with Catphan QA resulted furthermore Bellesi et al. European Radiology Experimental (2017) 1:18 Page 5 of 10 Fig. 3 Noise quantification with Catphan QA and CTQA_cp for the different CT protocols and reconstruction algorithms systematically higher than those evaluated with the low-contrast object detectability [15]. Introducing CTQA_cp, although the difference was limited and al- IMR (levels 1–3), the average of the percentage of cor- ways below 1. At mAs values less than or equal to 30, rect answers increases significantly and remains above Catphan QA was unable to quantify MTF. Uniformity 90% while lowering the mAs values up to 40. results, shown in Fig. 4, showed small deviations vari- Table 2 shows Catphan QA low-contrast results. De- ability, especially below 80 mAs, for CTQA_cp. tectability of objects with 1% contrast is incremented from 3 mm to 2 mm details with the introduction of Low-contrast spatial resolution evaluation IMR, indicating that the use of iterative algorithms Figure 5 shows the average and standard deviation of the slightly improves the detection of low-contrast objects percentage of correct answers provided by the human [16]. observers at changing CT protocol. For mAs in the Table 3 shows model observer results in terms of range from 240 to 160, using FBP or iDOSE (levels 1–6), minimum diameter of the detectable low-contrast de- the average of correct answers is suboptimal, indicating tails. Results from 300 to 200 mAs showed a high vari- a net degradation of the perceived image quality and of ability, whereas from 180 to 100 mAs they were almost Fig. 4 Uniformity quantification with Catphan QA and CTQA_cp for the different CT protocols and reconstruction algorithms Bellesi et al. European Radiology Experimental (2017) 1:18 Page 6 of 10 Table 1 High-contrast resolution results for 50% and 10% MTF, obtained with Catphan QA and CTQA_cp MTF results Catphan QA CTQA_cp CT acquisition parameter MTF (ll/cm) 50% MTF (ll/cm) 10% MTF (ll/cm) 50% MTF (ll/cm) 10% 300 mAs FBP 3.7 6.4 3.0 6.1 280 mAs FBP 3.7 6.3 3.0 6.5 260 mAs FBP 3.7 6.4 3.5 6.1 240 mAs FBP 3.7 6.3 3.0 6.1 220 mAs FBP 3.8 6.5 3.5 6.5 200 mAs FBP 3.9 6.6 3.1 6.5 180 mAs IDOSE 1 3.8 6.5 2.8 6.0 180 mAs IDOSE 2 3.7 6.5 3.1 6.0 180 mAs IDOSE 3 3.8 6.5 3.5 6.5 180 mAs IDOSE 4 3.9 6.6 3.4 6.0 170 mAs IDOSE 4 3.8 6.5 3.0 6.0 170 mAs IDOSE 5 3.8 6.5 3.0 6.1 170 mAs IDOSE 6 3.9 6.7 3.0 6.1 160 mAs IDOSE 6 3.7 6.5 2.9 5.6 160 mAs IMR1 4.2 7.1 3.6 6.5 150 mAs IMR1 4.1 7.1 3.6 6.6 140 mAs IMR1 4.1 6.9 3.5 6.5 140 mAs IMR2 4.0 6.8 3.7 6.4 130 mAs IMR2 4.2 7.1 3.8 6.5 130 mAs IMR3 4.0 6.8 3.5 6.3 120 mAs IMR3 3.9 6.7 3.5 6.0 110 mAs IMR3 4.1 7.0 3.9 6.5 100 mAs IMR3 3.9 6.7 3.5 6.1 90 mAs IMR3 3.9 6.7 3.6 6.1 80 mAs IMR3 4.0 6.8 3.6 6.4 70 mAs IMR3 3.6 6.2 3.5 6.0 60 mAs IMR3 3.8 6.5 3.5 5.9 50 mAs IMR3 3.4 6.0 3.6 5.7 40 mAs IMR3 3.4 6.0 3.2 5.6 30 mAs IMR3 NE NE 4.4 6.9 25 mAs IMR3 NE NE 3.5 6.0 20 mAs IMR3 NE NE 3.7 7.0 15 mAs IMR3 NE NE 4.0 7.0 NE not evaluated by the software constant, with the diameters of the minimum detectable also evaluated. In general, results obtained by means of detail all being around 2 mm. Below 100 mAs the soft- CTQA_cp and Catphan QA in terms of image quality ware system was unable to detect objects probably due were approximately in agreement. The resolution of a to intrinsic algorithm limitations. CT imaging system is well characterised with the MTF, which indicates its ability to reproduce various levels of Discussion detail from a region of the patient to its image. Small The purpose of this work was to use IR algorithms for differences obtained for uniformity and MTF are likely obtaining a percentage threshold value of mAs in order due to small differences between the applied calculation to reduce CT dose while maintaining image quality. Hu- algorithms. In particular, for uniformity analysis, position man and computational detection performances were and dimension of the ROIs may change between Bellesi et al. European Radiology Experimental (2017) 1:18 Page 7 of 10 Fig. 5 Percentage of mean correct answers (histogram) and of their variability (error bars representing the standard deviation, k = 1) at changing CT protocols and reconstruction algorithms CTQA_cp and Catphan QA. It is only specified that in Evaluation of the low-contrast performance in CT im- CTQA_cp the area of the ROIs correspond to the area aging is a difficult task. It is related to the ability of an of a circle with diameter 10% of the diameter of the operator to distinguish between two objects or regions homogeneous region in the CP486 module. Whereas for with similar CT number and it depends on statistical Catphan QA, it is indicated that the outer edge of each noise levels, contrast and size of the signal. ROI is located 1 cm from module border. For MTF Referring to Fig. 5, on the one hand the percentage of evaluation, the small changes might be due to differ- correct answers is a proper quantification of the effi- ences in the Fourier transform analysis of the images. ciency of the application of the various reconstruction Image quality analysis anyway confirmed data already algorithms for low-contrast details identification, on the reported in literature, supporting the efficiency of the other hand the standard deviation is a good descriptor novel IR methods if compared to standard reconstruc- of inter-observer’s variability of image quality evaluation. tion algorithms such as FBP [17, 18]. Regarding noise, it The comparison between the different acquisition and initially increased while mAs values were lowered using image reconstruction modalities confirmed the highest FBP reconstruction. The application of iDose con- efficiency for IMR, level 3 [22]. In fact, in all tested con- stantly reduced it, even at decreasing mAs, and IMR ditions low-contrast detection rates were greater for kept it low while tube loading reduced to 50 mAs. IMR than for FBP or iDOSE; low-contrast detectability Below this value, noise increased with a consequent was preserved with IMR up to a tube loading reduction degradation of the image quality due to IR limits at to 40 mAs. In accordance to Katsura et al. [23], a conse- very low mAs values [19]. It was observed that the quent decrease of dose to the patient by a factor up to 7 uniformity values were within the advised limit (i.e. 80% dose reduction) seems, therefore, to be possible (ΔHU ≤4)[20].iDOSE provided,therefore,a similar without producing any significant detriment to the image resolution to that obtained with FBP at signifi- images. cantly higher mAs values. Our results represent, Interestingly, the variability of the percentage of cor- therefore, a valuable confirmation that the use of IR rect answers was high for iDose in the range of 220–160 algorithms preserves the spatial resolution while redu- mAs, whereas it was much lower with IMR, even at re- cing mAs [21], except at very low tube load (i.e. < 50 duced tube loading. This was due to possible reconstruc- mAs) where a very small decrease in spatial reso- tion limitations of iDOSE, which might stress the lution was found [19]. perception variability among observers. As previously Bellesi et al. European Radiology Experimental (2017) 1:18 Page 8 of 10 Table 2 1%, 05% and 0.3% low-contrast detectability obtained with Catphan QA CT acquisition parameter Detail at 1% contrast (mm) Detail at 0.5% contrast (mm) Detail at 0.3% contrast (mm) 300 mAs Standard 2 5 7 280 mAs Standard 3 5 7 260 mAs Standard 3 5 7 240 mAs Standard 3 5 7 220 mAs Standard 3 5 8 200 mAs Standard 3 6 8 180 mAs ISODOSE 1 3 6 9 180 mAs ISODOSE 2 3 6 9 180 mAs ISODOSE 3 3 5 9 180 mAs ISODOSE 4 3 5 8 170 mAs ISODOSE 4 3 6 9 170 mAs ISODOSE 5 3 5 8 170 mAs ISODOSE 6 2 5 9 160 mAs ISODOSE 6 3 5 9 160 mAs IMR1 2 5 6 150 mAs IMR1 2 3 6 140 mAs IMR1 2 3 6 140 mAs IMR2 2 3 5 130 mAs IMR2 2 3 6 130 mAs IMR3 2 2 5 120 mAs IMR3 2 3 6 110 mAs IMR3 2 3 6 100 mAs IMR3 2 3 6 90 mAs IMR3 2 2 5 80 mAs IMR3 2 3 7 70 mAs IMR3 2 3 6 60 mAs IMR3 2 3 6 50 mAs IMR3 2 3 8 40 mAs IMR3 2 4 15 30 mAs IMR3 NE NE NE 25 mAs IMR3 NE NE NE 20 mAs IMR3 NE NE NE 15 mAs IMR3 NE NE NE NE not evaluated described, psychological factors could, in fact, affect the for specific clinical applications (e.g. morphological test results [10]. As observers have performed the test in evaluation of tumours) in specific anatomical regions different moments of the day, diagnostic accuracy, visual and diseases. We did not focus on specific anatomical accommodation, reading time, subjective ratings of fa- regions and diseases because the approach described in tigue and visual strain, before and after a day of clinical this study could be adopted in many different clinical reading, may all have contributed as confounding factors applications, including the low-contrast regions/tissues in terms of image quality evaluations. Image texture, ar- detection task, such as liver lesion identification in tefacts and over-smoothing of images with higher abdominal CT [24, 25]. strengths of IR may have affected diagnostic results [24]. The performance of model observer software was, in One could argue that, different from our work, a large general, good in terms of low-contrast detectability up amount of papers compared iterative with exact methods to 100 mAs. Below this value, the model observer did Bellesi et al. European Radiology Experimental (2017) 1:18 Page 9 of 10 Table 3 1% low-contrast detectability obtained with the model The model observer given in CTQA_cp showed to be observer in CTQA_cp a valid tool for a first evaluation of the analysed data, CT acquisition parameter Detail at 1% contrast (mm) but presented the following limitations that would re- quire upgrades and improvements: (1) no optimisation/ 300 mAs Standard 5.8 adaptation is possible to ‘instruct’ the system for specific 280 mAs Standard 7.6 study conditions; (2) no univocal and absolute detect- 260 mAs Standard 15 ability scoring is provided as output. In particular, a de- 240 mAs Standard 0 tectability scoring could be important to better quantify 220 mAs Standard 2 the right mAs reduction percentage, optimised in terms 200 mAs Standard 2 of human-perceived image quality. In general, a better model observer software, with sophisticated interfaces 180 mAs ISODOSE 1 0 and specific setup possibilities, should probably be 180 mAs ISODOSE 2 2 adopted in future to assist better and predict human 180 mAs ISODOSE 3 2.2 observer’s performance. Implementation of a more ad- 180 mAs ISODOSE 4 2 vanced software, which is beyond the aim of this study, 170 mAs ISODOSE 4 2.4 is, however, very complex, as it requires a thorough 170 mAs ISODOSE 5 2.3 knowledge of model observer theory, statistics and in- formatics [26]. 170 mAs ISODOSE 6 2 In conclusion, this study demonstrated that the appli- 160 mAs ISODOSE 6 2.1 cation of the IR algorithm IMR to phantom images pre- 160 mAs IMR1 2 serves a good image quality and object detectability for 150 mAs IMR1 2 human radiological evaluation of CT exams, with a po- 140 mAs IMR1 2.5 tential noise reduction up to 60% and, in particular, an 140 mAs IMR2 2 85% dose reduction to the patient. With respect to other studies, the method presented in this work can be easily 130 mAs IMR2 2 implemented and contains a thorough analysis for the 130 mAs IMR3 2 evaluation and optimisation of mAs according to the 120 mAs IMR3 2 adopted reconstruction algorithms. The model observer 110 mAs IMR3 2 can, in principle, be useful to assist human performance 100 mAs IMR3 2 in CT low-contrast detection tasks and in dose optimisa- 90 mAs IMR3 ND tion, but needs to be optimised in order to extract useful information to support and predict human observer 80 mAs IMR3 ND evaluations on CT images. Further studies are required 70 mAs IMR3 ND to confirm the reported findings. 60 mAs IMR3 ND 50 mAs IMR3 ND Abbreviations 40 mAs IMR3 ND 4-AFC: Four-alternative forced-choice; CT: Computed tomography; FBP: Filtered back projection; IMR: Iterative model reconstruction; 30 mAs IMR3 ND MTF: Modulation transfer function; RA: Reconstruction algorithm; ROI: Region 25 mAs IMR3 ND of interest 20 mAs IMR3 ND Availability of data and materials 15 mAs IMR3 ND Data will not be shared because of dimension and number of files involved. ND not detected by the software Authors’ contributions Luca Bellesi; substantial conception and design of the work; substantial data collection, analysis and interpretation; drafting and revising the work critically not work well probably due to software intrinsic limita- for important intellectual content; approval of the final version to be tions. Different from human observers, model observer published. Stefano Presilla; substantial conception and design of the work; software recognised objects of 2 mm in diameter as a data collection; revising the work for important intellectual content, approval of the final version to be published. Mauro Carrara; drafting and reviewing of prediction of human observer performance also between the manuscript. Paolo Colleoni, Diego Gaudino and Francesco Pupillo; 220 and 160 mAs. This difference was particularly evi- contribution to data analysis. Rolf Wyttenbach, Antonio Braghetti, Carla dent on the 220-mAs images, where the mean value of Puligheddu; 4-AFC test, data discussion and revision of the work. correct answers by human observers was 9%, whereas according to the model observer a 2-mm diameter ob- Competing interests ject is detectable. The authors declare that they have no competing interests. Bellesi et al. European Radiology Experimental (2017) 1:18 Page 10 of 10 Publisher’sNote 18. Joemai RM, Veldkamp WJ, Kroft LJ, Hernandez-Giron I, Geleijns J (2013) Springer Nature remains neutral with regard to jurisdictional claims in Adaptive iterative dose reduction 3D versus filtered back projection in CT: published maps and institutional affiliations. evaluation of image quality. AJR Am J Roentgenol 201:1291–1297 19. McCollough CH, Yu L, Kofler JM et al (2015) Degradation of CT low-contrast Author details spatial resolution due to the use of iterative reconstruction and reduced Medical Physics Unit, Ente Ospedaliero Cantonale, Ospedale San Giovanni, dose levels. Radiology 276:499–506 Bellinzona 6500, Switzerland. Department of Radiology, Ente Ospedaliero 20. International Electrotechnical Commission (2004) Evaluation and routine Cantonale, Ospedale San Giovanni, Bellinzona, Switzerland. University of testing in medical imaging departments. Part 3-5: Acceptancy tests – Bern, Bern, Switzerland. Imaging performance of computed tomography X-ray equipment. IEC 61223-3-5, 1st ed. Geneva. Received: 2 June 2017 Accepted: 21 August 2017 21. Aurumskjöld ML, Ydström K, Tingberg A, Söderberg M (2016) Model-based iterative reconstruction enables the evaluation of thin-slice computed tomography images without degrading image quality or increasing radiation dose. Radiat Prot Dosimetry 169:100–106 22. Yoon JH, Lee JM, Yu MH et al (2014) Comparison of iterative model-based References reconstruction versus conventional filtered back projection and hybrid 1. Contract ENER/2010/NUCL/SI2.581237. Study on European population doses iterative reconstruction techniques: lesion conspicuity and influence of from medical exposure. (Dose Datamed 2) DDM2 project report on body size in anthropomorphic liver phantoms. J Comput Assist Tomogr 38: European population dose estimation. http://ddmed.eu/_media/news: 859–868 ddm2_project_report_supplement_drls_final_draft_on_web_page_28_jan_ 23. Katsura M, Matsuda I, Akahane M et al (2012) Model-based iterative 2013.pdf. Accessed 22 Feb 2017. reconstruction technique for radiation dose reduction in chest CT: 2. European Commission (EC) (2014) Medical radiation exposure of the comparison with the adaptive statistical iterative reconstruction technique. European population. Radiation Protection 180. https://ec.europa.eu/energy/ Eur Radiol 22:1613–1623 sites/ener/files/documents/RP180.pdf. Accessed 15 Mar 2017. 24. Padole A, Ali Khawaja RD, Kalra MK, Singh S (2015) CT radiation dose and 3. Le Coultre R, Bize J, Champendal M et al (2016) Exposure of the Swiss iterative reconstruction techniques. AJR Am J Roentgenol 204:W384–W392 population by radiodiagnostics: 2013 review. Radiat Prot Dosimetry 169: 25. Lu L (2014) Model-based iterative reconstruction: a promising algorithm for 221–224 today’s computed tomography imaging. JMIRS 45:131–136 4. Willemink MJ, de Jong PA, Leiner T et al (2013) Iterative reconstruction 26. Ott JG, Ba A, Racine D et al (2016) Patient exposure optimisation through techniques for computed tomography Part 1: technical principles. Eur task-based assessment of a new model-based iterative reconstruction Radiol 23:1623–1631 technique. Radiat Prot Dosimetry 169:68–72 5. Eck BL, Fahmi R, Brown KM et al (2015) Computational and human observer image quality evaluation of low dose, knowledge-based CT iterative reconstruction. Med Phys 42:6098–6111 6. Hernandez-Giron I, Geleijns J, Calzado A, Veldkamp WJ (2011) Automated assessment of low contrast sensitivity for CT systems using a model observer. Med Phys 38:S25–S35 7. Greffier J, Macri F, Larbi A et al (2015) Dose reduction with iterative reconstruction: Optimization of CT protocols in clinical practice. Diagn Interv Imaging 96:477–486 8. Verdun FR, Racine D, Ott JG et al (2015) Image quality in CT: From physical measurements to model observers. Phys Med 31:823–843 9. Barrett HH, Yao J, Rolland JP, Myers KJ (1993) Model observers for assessment of image quality. Proc Natl Acad Sci 90:9758–9765 10. He X, Park S (2013) Model observers in medical imaging research. Theranostics 3:774–786 11. Vaishnav JY, Jung WC, Popescu LM, Zeng R, Myers KJ (2014) Objective assessment of image quality and dose reduction in CT iterative reconstruction. Med Phys 41:071904 12. Thompson JD, Manning DJ, Hogg P (2013) The value of observer performance studies in dose optimization: a focus on free-response receiver operating characteristic methods. J Nucl Med Technol 41:57–64 13. Miéville FA, Gudinchet F, Brunelle F, Bochud FO, Verdun FR (2013) Iterative reconstruction methods in two different MDCT scanners: physical metrics and 4-alternative forced-choice detectability experiments–a phantom approach. Phys Med 29:99–110 14. Hernandez-Giron I, Calzado A, Geleijns J, Joemai RM, Veldkamp WJ (2014) Comparison between human and model observer performance in low- contrast detection tasks in CT images: application to images reconstructed with filtered back projection and iterative algorithms. Br J Radiol 87: 15. Goenka AH, Herts BR, Obuchowski NA et al (2014) Effect of reduced radiation exposure and iterative reconstruction on detection of low-contrast low-attenuation lesions in an anthropomorphic liver phantom: an 18-reader study. Radiology 272:154–163 16. Vlassenbroek A, Millon D, Michoux N, Coche EE (2014) Low contrast detectability improvements with iterative reconstructions of multi detector computed tomography images: a phantom study. ECR Poster C-1782. doi:10.1594/ecr2014/C-1782. 17. Mehta D, Thompson R, Morton T, Dhanantwari A, Shefer E (2013) Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality. Int J Med Phys 1:147–155 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Radiology Experimental Springer Journals

A simple method for low-contrast detectability, image quality and dose optimisation with CT iterative reconstruction algorithms and model observers

Download PDF
10 pages

Loading next page...
 
/lp/springer-journals/a-simple-method-for-low-contrast-detectability-image-quality-and-dose-tIv80ooMwD

References (31)

Publisher
Springer Journals
Copyright
Copyright © 2017 by The Author(s)
Subject
Medicine & Public Health; Imaging / Radiology; Diagnostic Radiology; Interventional Radiology; Neuroradiology; Ultrasound; Internal Medicine
eISSN
2509-9280
DOI
10.1186/s41747-017-0023-4
Publisher site
See Article on Publisher Site

Abstract

Background: The aim of this work was to evaluate detection of low-contrast objects and image quality in computed tomography (CT) phantom images acquired at different tube loadings (i.e. mAs) and reconstructed with different algorithms, in order to find appropriate settings to reduce the dose to the patient without any image detriment. Methods: Images of supraslice low-contrast objects of a CT phantom were acquired using different mAs values. Images were reconstructed using filtered back projection (FBP), hybrid and iterative model-based methods. Image quality parameters were evaluated in terms of modulation transfer function; noise, and uniformity using two software resources. For the definition of low-contrast detectability, studies based on both human (i.e. four- alternative forced-choice test) and model observers were performed across the various images. Results: Compared to FBP, image quality parameters were improved by using iterative reconstruction (IR) algorithms. In particular, IR model-based methods provided a 60% noise reduction and a 70% dose reduction, preserving image quality and low-contrast detectability for human radiological evaluation. According to the model observer, the diameters of the minimum detectable detail were around 2 mm (up to 100 mAs). Below 100 mAs, the model observer was unable to provide a result. Conclusion: IR methods improve CT protocol quality, providing a potential dose reduction while maintaining a good image detectability. Model observer can in principle be useful to assist human performance in CT low- contrast detection tasks and in dose optimisation. Keywords: Low-contrast object detection, Computed tomography, Image quality, Model-based iterative reconstruction, Model observer * Correspondence: Luca.Bellesi@eoc.ch Medical Physics Unit, Ente Ospedaliero Cantonale, Ospedale San Giovanni, Bellinzona 6500, Switzerland Full list of author information is available at the end of the article © The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Bellesi et al. European Radiology Experimental (2017) 1:18 Page 2 of 10 Key points tasks such as patient classification or estimation of vol- ume and/or other characteristics of tumours. However, Detection of low-contrast objects and image quality studies based on human observers are resource- in CT phantom images were evaluated demanding and involve a significant variability of intra- Different tube loadings and image reconstruction observer and interobserver performance. Being able to methods were tested extract as much statistical information as possible from Iterative reconstruction in CT provided significant the available images, computational model observers can mAs reduction without image detriment be used as convenient and objective surrogates of hu- Model observers are useful for parameter man beings to predict and/or define their expected per- optimisation in CT dose reduction tasks formance [6, 8]. In medical imaging, model observers were developed Background to study how system parameters affect signal detection The overall per caput mean effective dose per year to [9], taking into account physical factors that degrade the population in European countries, due to X-ray pro- image quality. They are also useful to evaluate and opti- cedures, is about 1.05 mSv. Computed tomography mise software systems, such as image reconstruction or (CT), which is a key medical imaging modality within processing methods, both to study and predict their ef- clinical diagnostic applications, contributes, on average, fects on human-observer performance [10–12]. to 57% of this dose (range 5.31–83.1%) [1], with a mean The purpose of this work was to evaluate image qual- value of 7.44 mSv [2]. In Switzerland, through 2013 the ity and low-contrast object detectability in CT phantom number of CT exams was 117 per 1000 inhabitants, with images acquired at different tube loadings (i.e. mAs) and an average dose per exam of 8.54 mSv. CT alone con- reconstructed with different algorithms in order to re- tributed to about 70% of the collective dose, with an duce mAs and consequently CT dose, with respect to a average annual effective dose of 1 mSv per inhabitant standard reference value, without detriment of the im- [3]. In light of these data, reduction of radiation dose ages. Model observer performance in terms of minimum from CT has become an essential field of study. In the diameter of the detectable low-contrast details was also last years, the advent of faster microprocessors, CT evaluated. iterative reconstruction (IR) methods were launched to integrate already existing algorithms such as filtered back projection (FBP) as a way to reduce patient Methods radiation exposure while maintaining high-contrast CT phantom image acquisition spatial resolution. In this study, a Catphan 504 phantom (The Phantom iDose4 (Philips iDose4™ system, Philips Healthcare, Laboratory, Salem, NY, USA) was used to perform all Cleveland, OH, USA) belongs to the first generation of image quality tests. It is a cylindrical phantom of 20-cm iterative hybrid reconstruction algorithms, which com- length and 20-cm diameter, containing several test mod- bine FBP and IR algorithms [4]. On the contrary, Itera- ules: a solid image uniformity module (CTP486), a 21- tive Model Reconstruction (IMR; Philips Healthcare, line pair and point source high resolution module Cleveland, OH, USA) is an advanced knowledge-based (CTP528), a module for slice width, sensitometry and algorithm that models the process of physical data ac- pixel size evaluation (CTP401), and a low-contrast mod- quisition through the iterative minimisation of the differ- ule (CTP515). In particular, the low contrast CTP515 ences between image raw data and the estimated image module contains two sub-regions: the supraslice region [5]. IMR differs from FBP methods in that the recon- with three groups of low-contrast objects, consisting of struction becomes an optimisation process that takes nine circular objects with diameters in the range of 2– into account data statistics, image statistics, and system 15 mm and contrast of 0.3%, 0.5% and 1.0%, respectively, models. IMR levels differ by number of processing cycles and a subslice region with three groups of four circular which increase concurrently with increasing levels. The objects each (diameters in the range of 3–9 mm, con- main difficulty is to preserve an adequate diagnostic trast of 1.0%). image quality reducing exposure mAs values and there- CT scans of the Catphan phantom were acquired with fore reducing the dose to the patient [6, 7]. a Philips 256 iCT multi-slice CT unit (Brilliance iCT, Medical image quality assessment involves both a sci- Philips, Best, The Netherlands), aligning the phantom entific and philosophical approach to define how ‘well’ main axis with the axis of rotation of the scanner (z- specific information of interest is obtained from images. axis). Acquisitions were performed selecting a beam col- One method to define medical image quality is called limation of 128 × 0.625 mm, scanning field of view of ‘statistical task-based assessment approach’ and consists 214 mm, scan length of 213 mm, helical acquisition with of the evaluation of the observer performance during 0.976 pitch factor, tube voltage of 100 kVp, scan time Bellesi et al. European Radiology Experimental (2017) 1:18 Page 3 of 10 2.5 s, rotation time 0.5 s, slice thickness 3 mm and zoom Catphan QA were used and the obtained results were 100%. compared. The product of tube current and exposure time per ro- tation (i.e. tube load) was in the range of 15–300 mAs. Noise Image reconstruction was performed using the following Noise was characterised on the images of the Catphan reconstruction algorithms: FBP; iDose with levels in the CTP 485 uniform module as the standard deviation of range of 1–6; and IMR with levels in the range of 1–3. pixel values within a square region of interest (ROI) lo- cated at the centre of the phantom module. Software for image analysis Uniformity For image quality analysis, one software product was Uniformity was calculated in the homogeneous region of used and compared, in terms of numerical results, with the CP486 module as the deviation in CT numbers of an advanced automated quality assurance software ser- the mean value of upper, right, lower, and left circular vice available on the web. For the definition of low- off-centre ROIs from the mean value of a ROI placed at contrast detectability, however, studies based on both the centre of the image of the phantom. Position and di- human and model observers were performed. mension of the ROIs could change between the two soft- CT image quality parameters were evaluated with two ware products. In any case, the closer to unity was the different software resources, CTQA_cp and Catphan result, the more uniform was the image. QA (Image Owl, Inc., Greenwich, NY, USA), in order to cross-check the obtained results and validate CTQA_cp High-contrast spatial resolution results with a reference. CTQA_cp (version 0.3.1) is a MTF was calculated as the Fourier transform of the freeware software package developed to aid CT quality point spread function of a region of interest centred on assurance programs and able to automatically produce the lower bead point object of the Catphan CTP 528- image quality reports. In particular, the following param- point source module. eters are analysed with CTQA_cp: slice thickness, pixel size, CT number linearity, uniformity, homogeneity, Low-contrast spatial resolution image noise across detector rows, and modulation trans- As described below, empirical and computational fer function (MTF). A low-contrast resolution analysis methods were evaluated in this study to quantify low- tool of the Catphan CP515 module based on a model contrast spatial resolution. observer is also available. Catphan QA executes an automatic analysis of CT Evaluation with the four-alternative forced test Catphan images and produces an image quality report. Four-alternative forced-choice (4-AFC) [13] test was ex- The following CT imaging performance parameters are ecuted to evaluate low-contrast spatial resolution by five evaluated: sensitometry; MTF (i.e. from beads and wires radiologists with at least 15 years of experience in clin- analysis); critical frequency; CT linearity; phantom pos- ical CT and four experienced radiology technicians [14]. ition; rotation and yaw; slice width; and contrast Observers were trained on all technical aspects and ob- detectability. jectives of the study and frontal training was performed Catphan QA also includes a contrast diameter detail through examples before the test. function that returns dimensions of the smallest detect- The 4-AFC test was performed in a darkened room able target for each of the three contrast values and was with a constant level of low ambient lighting and images used in order to obtain image quality low-contrast were presented on a DICOM-calibrated megapixel information. colour LCD screen (Radiforce RX320 LCD, EIZO Cor- Image quality parameters were evaluated with poration) with a native resolution of 1536 × 2048. Initial CTQA_cp and Catphan QA and on the phantom images window and level values of 100 and 1090 were sug- acquired with the different CT mAs values and recon- gested, respectively, but observers were free to modify structed with different reconstruction algorithms. them if necessary. No limitations on viewing distance and time were set and no reference image was provided Physical metrics quantification with CTQA_cp and before the start. Each human observer analysed 543 Catphan QA stacks of four images containing either just background For each adopted scanning protocol (i.e. different mAs) or the 6-mm and 7-mm diameter objects (1% contrast) and reconstruction algorithms, noise, uniformity, and of the low-contrast supraslice region of the Catphan high-contrast spatial resolution were evaluated in order phantom. To create the stack of images for the 4-AFC to quantify how the different CT acquisition parameters test, dedicated macros were created using the freeware impact on the physical metrics. Both CTQA_cp and software ImageJ (National Institute of Health Image, Bellesi et al. European Radiology Experimental (2017) 1:18 Page 4 of 10 Bethesda, MD, USA) that automatically executes the fol- lowing steps: (1) extracts samples of the low-contrast objects (diameters 6–7 mm, 1% contrast) or of the back- ground from low-contrast Catphan module (Fig. 1); (2) generates a series of images each containing four quad- rants with low-contrast circular objects or background, randomly chosen from Catphan images acquired at dif- ferent experimental conditions (i.e. mAs in the range of 30–300) and reconstructed by means of FBP, iDOSE (i.e. levels 1–6) and IMR (i.e. levels 1–3). Sixteen images for each CT protocol modality were overall selected and randomly arranged over the stack of 543 images. One example of images is provided in Fig. 2, each quadrant possibly representing the particular of the Catphan CTP515 low-contrast module shown in Fig. 1; (3) cre- ates a stack of 543 images in a single DICOM image se- quence that was loaded to a picture archiving and communication system PACS (Philips IntelliSpace PACS Enterprise 4.4.532.1, Philips Healthcare Informatics, Inc., Foster City, CA, USA) for further evaluation by the observers. Fig. 2 Example of image for the 4-AFC test. In this case, low-contrast In this study, observers had to identify the presence of objects were in quadrants a, b and c while quadrant d is empty one or more quadrants with low-contrast lesions and to indicate their position within the image. In principle, in each one of the 543 images, low-contrast objects were in Computational evaluation none, one, two, three, or any quadrant. The percentage The computer model observer provided with CTQA_cp of correct answers given by each observer subjected to was used to define low-contrast detectability on the Cat- the 4-AFC experiment was analysed and evaluated. phan low-contrast supraslice images acquired at differ- Inter-CT protocol modality (i.e. each combination of mAs ent experimental conditions (i.e. mAs in the range of and reconstruction algorithms) analysis was performed. 15–300) and reconstructed by means of FBP, iDOSE (i.e. levels 1–6) and IMR (i.e. levels 1–3). According to the method, which is exhaustively described by Hernandez- Giron et al. [6], output of the software system is the smallest ‘visible’ object size at 1%, 0.5%, and 0.1% con- trast. Only objects with 1% contrast were evaluated be- cause 0.5% and 0.1% objects were often not visible during first visual evaluations after phantom CT acquisi- tions. Catphan QA also includes a function for low- contrast diameter detail evaluation, which returns dimensions of the smallest detectable target for each of the three contrast values. This function is not based on a model observer-based statistical approach, but it is re- lated on the use of an algorithm for image analysis. Results Physical metrics quantification with CTQA_cp and Catphan QA Noise and uniformity evaluations are provided in Figs. 3 and 4, respectively; the high-contrast spatial resolutions for 50% and 10% MTF are given in Table 1. Numerical results of both software systems resulted to be compar- able in terms of noise analysis, whereas a difference arose for uniformity. High-contrast spatial resolutions Fig. 1 Image of the low-contrast module of the Catphan phantom evaluated with Catphan QA resulted furthermore Bellesi et al. European Radiology Experimental (2017) 1:18 Page 5 of 10 Fig. 3 Noise quantification with Catphan QA and CTQA_cp for the different CT protocols and reconstruction algorithms systematically higher than those evaluated with the low-contrast object detectability [15]. Introducing CTQA_cp, although the difference was limited and al- IMR (levels 1–3), the average of the percentage of cor- ways below 1. At mAs values less than or equal to 30, rect answers increases significantly and remains above Catphan QA was unable to quantify MTF. Uniformity 90% while lowering the mAs values up to 40. results, shown in Fig. 4, showed small deviations vari- Table 2 shows Catphan QA low-contrast results. De- ability, especially below 80 mAs, for CTQA_cp. tectability of objects with 1% contrast is incremented from 3 mm to 2 mm details with the introduction of Low-contrast spatial resolution evaluation IMR, indicating that the use of iterative algorithms Figure 5 shows the average and standard deviation of the slightly improves the detection of low-contrast objects percentage of correct answers provided by the human [16]. observers at changing CT protocol. For mAs in the Table 3 shows model observer results in terms of range from 240 to 160, using FBP or iDOSE (levels 1–6), minimum diameter of the detectable low-contrast de- the average of correct answers is suboptimal, indicating tails. Results from 300 to 200 mAs showed a high vari- a net degradation of the perceived image quality and of ability, whereas from 180 to 100 mAs they were almost Fig. 4 Uniformity quantification with Catphan QA and CTQA_cp for the different CT protocols and reconstruction algorithms Bellesi et al. European Radiology Experimental (2017) 1:18 Page 6 of 10 Table 1 High-contrast resolution results for 50% and 10% MTF, obtained with Catphan QA and CTQA_cp MTF results Catphan QA CTQA_cp CT acquisition parameter MTF (ll/cm) 50% MTF (ll/cm) 10% MTF (ll/cm) 50% MTF (ll/cm) 10% 300 mAs FBP 3.7 6.4 3.0 6.1 280 mAs FBP 3.7 6.3 3.0 6.5 260 mAs FBP 3.7 6.4 3.5 6.1 240 mAs FBP 3.7 6.3 3.0 6.1 220 mAs FBP 3.8 6.5 3.5 6.5 200 mAs FBP 3.9 6.6 3.1 6.5 180 mAs IDOSE 1 3.8 6.5 2.8 6.0 180 mAs IDOSE 2 3.7 6.5 3.1 6.0 180 mAs IDOSE 3 3.8 6.5 3.5 6.5 180 mAs IDOSE 4 3.9 6.6 3.4 6.0 170 mAs IDOSE 4 3.8 6.5 3.0 6.0 170 mAs IDOSE 5 3.8 6.5 3.0 6.1 170 mAs IDOSE 6 3.9 6.7 3.0 6.1 160 mAs IDOSE 6 3.7 6.5 2.9 5.6 160 mAs IMR1 4.2 7.1 3.6 6.5 150 mAs IMR1 4.1 7.1 3.6 6.6 140 mAs IMR1 4.1 6.9 3.5 6.5 140 mAs IMR2 4.0 6.8 3.7 6.4 130 mAs IMR2 4.2 7.1 3.8 6.5 130 mAs IMR3 4.0 6.8 3.5 6.3 120 mAs IMR3 3.9 6.7 3.5 6.0 110 mAs IMR3 4.1 7.0 3.9 6.5 100 mAs IMR3 3.9 6.7 3.5 6.1 90 mAs IMR3 3.9 6.7 3.6 6.1 80 mAs IMR3 4.0 6.8 3.6 6.4 70 mAs IMR3 3.6 6.2 3.5 6.0 60 mAs IMR3 3.8 6.5 3.5 5.9 50 mAs IMR3 3.4 6.0 3.6 5.7 40 mAs IMR3 3.4 6.0 3.2 5.6 30 mAs IMR3 NE NE 4.4 6.9 25 mAs IMR3 NE NE 3.5 6.0 20 mAs IMR3 NE NE 3.7 7.0 15 mAs IMR3 NE NE 4.0 7.0 NE not evaluated by the software constant, with the diameters of the minimum detectable also evaluated. In general, results obtained by means of detail all being around 2 mm. Below 100 mAs the soft- CTQA_cp and Catphan QA in terms of image quality ware system was unable to detect objects probably due were approximately in agreement. The resolution of a to intrinsic algorithm limitations. CT imaging system is well characterised with the MTF, which indicates its ability to reproduce various levels of Discussion detail from a region of the patient to its image. Small The purpose of this work was to use IR algorithms for differences obtained for uniformity and MTF are likely obtaining a percentage threshold value of mAs in order due to small differences between the applied calculation to reduce CT dose while maintaining image quality. Hu- algorithms. In particular, for uniformity analysis, position man and computational detection performances were and dimension of the ROIs may change between Bellesi et al. European Radiology Experimental (2017) 1:18 Page 7 of 10 Fig. 5 Percentage of mean correct answers (histogram) and of their variability (error bars representing the standard deviation, k = 1) at changing CT protocols and reconstruction algorithms CTQA_cp and Catphan QA. It is only specified that in Evaluation of the low-contrast performance in CT im- CTQA_cp the area of the ROIs correspond to the area aging is a difficult task. It is related to the ability of an of a circle with diameter 10% of the diameter of the operator to distinguish between two objects or regions homogeneous region in the CP486 module. Whereas for with similar CT number and it depends on statistical Catphan QA, it is indicated that the outer edge of each noise levels, contrast and size of the signal. ROI is located 1 cm from module border. For MTF Referring to Fig. 5, on the one hand the percentage of evaluation, the small changes might be due to differ- correct answers is a proper quantification of the effi- ences in the Fourier transform analysis of the images. ciency of the application of the various reconstruction Image quality analysis anyway confirmed data already algorithms for low-contrast details identification, on the reported in literature, supporting the efficiency of the other hand the standard deviation is a good descriptor novel IR methods if compared to standard reconstruc- of inter-observer’s variability of image quality evaluation. tion algorithms such as FBP [17, 18]. Regarding noise, it The comparison between the different acquisition and initially increased while mAs values were lowered using image reconstruction modalities confirmed the highest FBP reconstruction. The application of iDose con- efficiency for IMR, level 3 [22]. In fact, in all tested con- stantly reduced it, even at decreasing mAs, and IMR ditions low-contrast detection rates were greater for kept it low while tube loading reduced to 50 mAs. IMR than for FBP or iDOSE; low-contrast detectability Below this value, noise increased with a consequent was preserved with IMR up to a tube loading reduction degradation of the image quality due to IR limits at to 40 mAs. In accordance to Katsura et al. [23], a conse- very low mAs values [19]. It was observed that the quent decrease of dose to the patient by a factor up to 7 uniformity values were within the advised limit (i.e. 80% dose reduction) seems, therefore, to be possible (ΔHU ≤4)[20].iDOSE provided,therefore,a similar without producing any significant detriment to the image resolution to that obtained with FBP at signifi- images. cantly higher mAs values. Our results represent, Interestingly, the variability of the percentage of cor- therefore, a valuable confirmation that the use of IR rect answers was high for iDose in the range of 220–160 algorithms preserves the spatial resolution while redu- mAs, whereas it was much lower with IMR, even at re- cing mAs [21], except at very low tube load (i.e. < 50 duced tube loading. This was due to possible reconstruc- mAs) where a very small decrease in spatial reso- tion limitations of iDOSE, which might stress the lution was found [19]. perception variability among observers. As previously Bellesi et al. European Radiology Experimental (2017) 1:18 Page 8 of 10 Table 2 1%, 05% and 0.3% low-contrast detectability obtained with Catphan QA CT acquisition parameter Detail at 1% contrast (mm) Detail at 0.5% contrast (mm) Detail at 0.3% contrast (mm) 300 mAs Standard 2 5 7 280 mAs Standard 3 5 7 260 mAs Standard 3 5 7 240 mAs Standard 3 5 7 220 mAs Standard 3 5 8 200 mAs Standard 3 6 8 180 mAs ISODOSE 1 3 6 9 180 mAs ISODOSE 2 3 6 9 180 mAs ISODOSE 3 3 5 9 180 mAs ISODOSE 4 3 5 8 170 mAs ISODOSE 4 3 6 9 170 mAs ISODOSE 5 3 5 8 170 mAs ISODOSE 6 2 5 9 160 mAs ISODOSE 6 3 5 9 160 mAs IMR1 2 5 6 150 mAs IMR1 2 3 6 140 mAs IMR1 2 3 6 140 mAs IMR2 2 3 5 130 mAs IMR2 2 3 6 130 mAs IMR3 2 2 5 120 mAs IMR3 2 3 6 110 mAs IMR3 2 3 6 100 mAs IMR3 2 3 6 90 mAs IMR3 2 2 5 80 mAs IMR3 2 3 7 70 mAs IMR3 2 3 6 60 mAs IMR3 2 3 6 50 mAs IMR3 2 3 8 40 mAs IMR3 2 4 15 30 mAs IMR3 NE NE NE 25 mAs IMR3 NE NE NE 20 mAs IMR3 NE NE NE 15 mAs IMR3 NE NE NE NE not evaluated described, psychological factors could, in fact, affect the for specific clinical applications (e.g. morphological test results [10]. As observers have performed the test in evaluation of tumours) in specific anatomical regions different moments of the day, diagnostic accuracy, visual and diseases. We did not focus on specific anatomical accommodation, reading time, subjective ratings of fa- regions and diseases because the approach described in tigue and visual strain, before and after a day of clinical this study could be adopted in many different clinical reading, may all have contributed as confounding factors applications, including the low-contrast regions/tissues in terms of image quality evaluations. Image texture, ar- detection task, such as liver lesion identification in tefacts and over-smoothing of images with higher abdominal CT [24, 25]. strengths of IR may have affected diagnostic results [24]. The performance of model observer software was, in One could argue that, different from our work, a large general, good in terms of low-contrast detectability up amount of papers compared iterative with exact methods to 100 mAs. Below this value, the model observer did Bellesi et al. European Radiology Experimental (2017) 1:18 Page 9 of 10 Table 3 1% low-contrast detectability obtained with the model The model observer given in CTQA_cp showed to be observer in CTQA_cp a valid tool for a first evaluation of the analysed data, CT acquisition parameter Detail at 1% contrast (mm) but presented the following limitations that would re- quire upgrades and improvements: (1) no optimisation/ 300 mAs Standard 5.8 adaptation is possible to ‘instruct’ the system for specific 280 mAs Standard 7.6 study conditions; (2) no univocal and absolute detect- 260 mAs Standard 15 ability scoring is provided as output. In particular, a de- 240 mAs Standard 0 tectability scoring could be important to better quantify 220 mAs Standard 2 the right mAs reduction percentage, optimised in terms 200 mAs Standard 2 of human-perceived image quality. In general, a better model observer software, with sophisticated interfaces 180 mAs ISODOSE 1 0 and specific setup possibilities, should probably be 180 mAs ISODOSE 2 2 adopted in future to assist better and predict human 180 mAs ISODOSE 3 2.2 observer’s performance. Implementation of a more ad- 180 mAs ISODOSE 4 2 vanced software, which is beyond the aim of this study, 170 mAs ISODOSE 4 2.4 is, however, very complex, as it requires a thorough 170 mAs ISODOSE 5 2.3 knowledge of model observer theory, statistics and in- formatics [26]. 170 mAs ISODOSE 6 2 In conclusion, this study demonstrated that the appli- 160 mAs ISODOSE 6 2.1 cation of the IR algorithm IMR to phantom images pre- 160 mAs IMR1 2 serves a good image quality and object detectability for 150 mAs IMR1 2 human radiological evaluation of CT exams, with a po- 140 mAs IMR1 2.5 tential noise reduction up to 60% and, in particular, an 140 mAs IMR2 2 85% dose reduction to the patient. With respect to other studies, the method presented in this work can be easily 130 mAs IMR2 2 implemented and contains a thorough analysis for the 130 mAs IMR3 2 evaluation and optimisation of mAs according to the 120 mAs IMR3 2 adopted reconstruction algorithms. The model observer 110 mAs IMR3 2 can, in principle, be useful to assist human performance 100 mAs IMR3 2 in CT low-contrast detection tasks and in dose optimisa- 90 mAs IMR3 ND tion, but needs to be optimised in order to extract useful information to support and predict human observer 80 mAs IMR3 ND evaluations on CT images. Further studies are required 70 mAs IMR3 ND to confirm the reported findings. 60 mAs IMR3 ND 50 mAs IMR3 ND Abbreviations 40 mAs IMR3 ND 4-AFC: Four-alternative forced-choice; CT: Computed tomography; FBP: Filtered back projection; IMR: Iterative model reconstruction; 30 mAs IMR3 ND MTF: Modulation transfer function; RA: Reconstruction algorithm; ROI: Region 25 mAs IMR3 ND of interest 20 mAs IMR3 ND Availability of data and materials 15 mAs IMR3 ND Data will not be shared because of dimension and number of files involved. ND not detected by the software Authors’ contributions Luca Bellesi; substantial conception and design of the work; substantial data collection, analysis and interpretation; drafting and revising the work critically not work well probably due to software intrinsic limita- for important intellectual content; approval of the final version to be tions. Different from human observers, model observer published. Stefano Presilla; substantial conception and design of the work; software recognised objects of 2 mm in diameter as a data collection; revising the work for important intellectual content, approval of the final version to be published. Mauro Carrara; drafting and reviewing of prediction of human observer performance also between the manuscript. Paolo Colleoni, Diego Gaudino and Francesco Pupillo; 220 and 160 mAs. This difference was particularly evi- contribution to data analysis. Rolf Wyttenbach, Antonio Braghetti, Carla dent on the 220-mAs images, where the mean value of Puligheddu; 4-AFC test, data discussion and revision of the work. correct answers by human observers was 9%, whereas according to the model observer a 2-mm diameter ob- Competing interests ject is detectable. The authors declare that they have no competing interests. Bellesi et al. European Radiology Experimental (2017) 1:18 Page 10 of 10 Publisher’sNote 18. Joemai RM, Veldkamp WJ, Kroft LJ, Hernandez-Giron I, Geleijns J (2013) Springer Nature remains neutral with regard to jurisdictional claims in Adaptive iterative dose reduction 3D versus filtered back projection in CT: published maps and institutional affiliations. evaluation of image quality. AJR Am J Roentgenol 201:1291–1297 19. McCollough CH, Yu L, Kofler JM et al (2015) Degradation of CT low-contrast Author details spatial resolution due to the use of iterative reconstruction and reduced Medical Physics Unit, Ente Ospedaliero Cantonale, Ospedale San Giovanni, dose levels. Radiology 276:499–506 Bellinzona 6500, Switzerland. Department of Radiology, Ente Ospedaliero 20. International Electrotechnical Commission (2004) Evaluation and routine Cantonale, Ospedale San Giovanni, Bellinzona, Switzerland. University of testing in medical imaging departments. Part 3-5: Acceptancy tests – Bern, Bern, Switzerland. Imaging performance of computed tomography X-ray equipment. IEC 61223-3-5, 1st ed. Geneva. Received: 2 June 2017 Accepted: 21 August 2017 21. Aurumskjöld ML, Ydström K, Tingberg A, Söderberg M (2016) Model-based iterative reconstruction enables the evaluation of thin-slice computed tomography images without degrading image quality or increasing radiation dose. Radiat Prot Dosimetry 169:100–106 22. Yoon JH, Lee JM, Yu MH et al (2014) Comparison of iterative model-based References reconstruction versus conventional filtered back projection and hybrid 1. Contract ENER/2010/NUCL/SI2.581237. Study on European population doses iterative reconstruction techniques: lesion conspicuity and influence of from medical exposure. (Dose Datamed 2) DDM2 project report on body size in anthropomorphic liver phantoms. J Comput Assist Tomogr 38: European population dose estimation. http://ddmed.eu/_media/news: 859–868 ddm2_project_report_supplement_drls_final_draft_on_web_page_28_jan_ 23. Katsura M, Matsuda I, Akahane M et al (2012) Model-based iterative 2013.pdf. Accessed 22 Feb 2017. reconstruction technique for radiation dose reduction in chest CT: 2. European Commission (EC) (2014) Medical radiation exposure of the comparison with the adaptive statistical iterative reconstruction technique. European population. Radiation Protection 180. https://ec.europa.eu/energy/ Eur Radiol 22:1613–1623 sites/ener/files/documents/RP180.pdf. Accessed 15 Mar 2017. 24. Padole A, Ali Khawaja RD, Kalra MK, Singh S (2015) CT radiation dose and 3. Le Coultre R, Bize J, Champendal M et al (2016) Exposure of the Swiss iterative reconstruction techniques. AJR Am J Roentgenol 204:W384–W392 population by radiodiagnostics: 2013 review. Radiat Prot Dosimetry 169: 25. Lu L (2014) Model-based iterative reconstruction: a promising algorithm for 221–224 today’s computed tomography imaging. JMIRS 45:131–136 4. Willemink MJ, de Jong PA, Leiner T et al (2013) Iterative reconstruction 26. Ott JG, Ba A, Racine D et al (2016) Patient exposure optimisation through techniques for computed tomography Part 1: technical principles. Eur task-based assessment of a new model-based iterative reconstruction Radiol 23:1623–1631 technique. Radiat Prot Dosimetry 169:68–72 5. Eck BL, Fahmi R, Brown KM et al (2015) Computational and human observer image quality evaluation of low dose, knowledge-based CT iterative reconstruction. Med Phys 42:6098–6111 6. Hernandez-Giron I, Geleijns J, Calzado A, Veldkamp WJ (2011) Automated assessment of low contrast sensitivity for CT systems using a model observer. Med Phys 38:S25–S35 7. Greffier J, Macri F, Larbi A et al (2015) Dose reduction with iterative reconstruction: Optimization of CT protocols in clinical practice. Diagn Interv Imaging 96:477–486 8. Verdun FR, Racine D, Ott JG et al (2015) Image quality in CT: From physical measurements to model observers. Phys Med 31:823–843 9. Barrett HH, Yao J, Rolland JP, Myers KJ (1993) Model observers for assessment of image quality. Proc Natl Acad Sci 90:9758–9765 10. He X, Park S (2013) Model observers in medical imaging research. Theranostics 3:774–786 11. Vaishnav JY, Jung WC, Popescu LM, Zeng R, Myers KJ (2014) Objective assessment of image quality and dose reduction in CT iterative reconstruction. Med Phys 41:071904 12. Thompson JD, Manning DJ, Hogg P (2013) The value of observer performance studies in dose optimization: a focus on free-response receiver operating characteristic methods. J Nucl Med Technol 41:57–64 13. Miéville FA, Gudinchet F, Brunelle F, Bochud FO, Verdun FR (2013) Iterative reconstruction methods in two different MDCT scanners: physical metrics and 4-alternative forced-choice detectability experiments–a phantom approach. Phys Med 29:99–110 14. Hernandez-Giron I, Calzado A, Geleijns J, Joemai RM, Veldkamp WJ (2014) Comparison between human and model observer performance in low- contrast detection tasks in CT images: application to images reconstructed with filtered back projection and iterative algorithms. Br J Radiol 87: 15. Goenka AH, Herts BR, Obuchowski NA et al (2014) Effect of reduced radiation exposure and iterative reconstruction on detection of low-contrast low-attenuation lesions in an anthropomorphic liver phantom: an 18-reader study. Radiology 272:154–163 16. Vlassenbroek A, Millon D, Michoux N, Coche EE (2014) Low contrast detectability improvements with iterative reconstructions of multi detector computed tomography images: a phantom study. ECR Poster C-1782. doi:10.1594/ecr2014/C-1782. 17. Mehta D, Thompson R, Morton T, Dhanantwari A, Shefer E (2013) Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality. Int J Med Phys 1:147–155

Journal

European Radiology Experimental – Springer Journals

Published: Oct 23, 2017

There are no references for this article.