- Email: [email protected]
Reducing Radiation Dose at Chest CT
ARTICLE IN PRESS
Original Investigation
Reducing Radiation Dose at Chest CT: Comparison Among Model-based Type Iterative Reconstruction, Hybrid Iterative Reconstruction, and Filtered Back Projection Constance de Margerie-Mellon, MD, Cédric de Bazelaire, MD, PhD, Claire Montlahuc, MD, Jérôme Lambert, MD, PhD, Antoine Martineau, PhD, Philippe Coulon, PhD, Eric de Kerviler, MD, Catherine Beigelman, MD Rationale and Objectives: The study aimed to evaluate the performances of two iterative reconstruction (IR) algorithms and of filtered back projection (FBP) when using reduced-dose chest computed tomography (RDCT) compared to standard-of-care CT. Materials and Methods: An institutional review board approval was obtained. Thirty-six patients with hematologic malignancies referred for a control chest CT of a known lung disease were prospectively enrolled. Patients underwent standard-of-care scan reconstructed with hybrid IR, followed by an RDCT reconstructed with FBP, hybrid IR, and iterative model reconstruction. Objective and subjective quality measurements, lesion detectability, and evolution assessment on RDCT were recorded. Results: For RDCT, the CTDIvol (volumetric computed tomography dose index) was 0.43 mGy·cm for all patients, and the median [interquartile range] effective dose was 0.22 mSv [0.22–0.24]; corresponding measurements for standard-of-care scan were 3.4 mGy [3.1–3.9] and 1.8 mSv [1.6–2.0]. Noise significantly decreased from FBP to hybrid IR and from hybrid IR to iterative model reconstruction on RDCT, whereas lesion conspicuity and diagnostic confidence increased. Accurate evolution assessment was obtained in all cases with IR. Emphysema identification was higher with iterative model reconstruction. Conclusion: Although iterative model reconstruction offered better diagnostic confidence and emphysema detection, both IR algorithms allowed an accurate evolution assessment with an effective dose of 0.22 mSv. Key Words: Computed tomography; lung; radiation dose reduction; iterative reconstruction; image noise. © 2016 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
Acad Radiol 2016; ■:■■–■■ From the Department of Radiology, INSERM UMR_S1165, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, (C. de M.-M., C. de B., E. de K.); Department of Informatics and Biostatistics, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, (C.M., J.L.); INSERM, ECSTRA team, CRESS-UMR_S 1153, 75010 Paris, (C.M., J.L.); Department of Nuclear Medicine, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, France (A.M.); Philips Healthcare, 33 rue de Verdun, 92156 Suresnes Cedex, France (P.C.); Department of Radiology, Vaudois University Hospital, rue du Bugnon 46, 1011 Lausanne, Switzerland (C.B.). Received January 8, 2016; revised May 30, 2016; accepted May 31, 2016. Address correspondence to: C. de M.-M. e-mail: [email protected] © 2016 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.acra.2016.05.019
O
ver the past decades, the average annual radiation dose delivered has raised significantly worldwide (1). Reducing computed tomography (CT) radiation dose has become a major concern because of the potential risk of radiation-induced cancer (2). Various strategies have been developed to reduce radiation exposure while maintaining image quality, including tube potential selection and tube current modulation. Unfortunately, radiation dose decrease is limited when using filtered back projection (FBP) reconstruction. FBP assumes that each pixel value perfectly represents the attenuation of the object at this location, setting aside system hardware details and photon noise statistic information (3). Because of those idealized assumptions, a lowered radiation dose is accompanied by an increased noise and often increased artifacts. 1
Academic Radiology, Vol ■, No ■, ■■ 2016
DE MARGERIE-MELLON ET AL
To overcome those difficulties, iterative reconstruction (IR) has been developed by CT manufacturers, enabling reduction in radiation dose without compromising spatial resolution by decreasing image noise and artifacts. Several algorithms have been developed by different companies, working either in the raw data space, image space, or both. Among them, hybrid generation IR called iDose4 (Philips Healthcare, Best, The Netherlands) is an algorithm that iterates in the raw data and image domains. More recently, a model-based type reconstruction called iterative model reconstruction (IMR, Philips Healthcare) has been developed. Compared to the previous algorithm, it takes into account data and image statistics but also considers system models (4). To date, few data are published about its potential applications in chest CT (5–7). We aimed to evaluate iDose4 and IMR performances in clinical practice. We hypothesized that reduced-dose chest computed tomography (RDCT) reconstructed with IR (iDose4 and IMR) for lung disease follow-up would enable obtaining the same final conclusion compared to a standard-of-care chest CT in a population of patients with hematologic malignancies. This population of patients was chosen because they frequently undergo successive chest CT examinations for their disease assessment, including lung tumoral spread (8), intercurrent complications, especially pulmonary infectious disorders (9,10) due to their immunocompromised status, or drug-induced lung disease (11,12). Moreover, compared to healthy subjects, such patients have high morbidity and mortality. Chest CT scan is here the best imaging modality for diagnostic orientation, biological sampling guidance, and follow-up under treatment. Although the potential risk of radiation-induced cancer seems low compared to the consequences of the hematologic disease and its treatments (chemotherapy and radiotherapy), such knowledge might help modify our routine practice even in the general population each time a follow-up is required, especially in young patients. We also wanted to evaluate FBP, iDose4, and IMR technique regarding objective and subjective image quality, and their detection potential when using RDCT.
MATERIALS AND METHODS Population
This prospective single-center study was approved by the ethics committee of our institution. According to local legislation, potential subjects were free to refuse to participate, but no written consent was required if they accepted. From February 2014 to June 2014, patients with hematologic malignancy and lung disease coming for a control unenhanced chest CT were screened. Inclusion criteria were as follows: age >18 years, available previous chest CT performed within the last 2 months (used as baseline), ability to understand the study protocol and its implications, and capacity to perform breath-holding. Patients who were under 18, who were unable to understand information for the participation to the study or to perform breath-holding, or who underwent previous chest CT more than 2 months before were 2
TABLE 1. Characteristics of Included Patients Characteristics
Patients (n = 36)
Age (y)* 59 [28.5–66] Sex ratio (M/F) 21/15 21.9 [19.8–24.2] Body mass index (kg/m2)* Effective diameter† (cm)* 27 [25.6–29.0] 28 [14–50] Delay since the initial chest computed tomography (d)* Initial pulmonary diagnosis Invasive aspergillosis (proven or probable) 7 (19%) Bacterial pneumonia (proven or probable) 7 (19%) Nonspecific lesions, probabilistic antifungal 11 (31%) or antibiotic therapy Nonspecific lesions, no therapy 7 (19%) All trans-retinoic acid syndrome 1 (3%) Organizing pneumonia 1 (3%) Lymphoma tumoral extension 1 (3%) Alveolar proteinosis 1 (3%) * Median [interquartile range] n. Effective diameter = AP × LAT , where AP represents the anterior-posterior dimension and LAT the lateral dimension of the patient. †
excluded. Each selected subject was given oral and written information in simple language about the objective, methods, and risks of study participation. Thirty-eight patients were selected to participate in the study. No patient opposed to its participation in the study. Readers used data sets of two patients for training. As a result, 36 patients were included in the study. Demographic and morphologic information is summarized in Table 1. Final diagnosis of the lung disease was based on the combination of clinical and radiological data, including the control chest CT. Imaging Protocol
All CT examinations were performed using a 64-row singlesource CT (Brilliance CT, Philips Healthcare). Each patient underwent two unenhanced chest acquisitions with the same length of helical run: first a standard-of-care CT, immediately followed by a RDCT. Acquisitions were performed in the supine position, with arms elevated, during a single breath-hold at full inspiration. Standard-of-care chest CT was performed according to our institution protocol for follow-up with a 140 kVp tube voltage, automatic exposure control (Z-DOM), and a 1.015 pitch factor. A combination of high tube voltage and limited currenttime product is usually employed at our center for this protocol to ensure the best image quality (especially to reduce the streak artifacts due to the high attenuation of shoulders in the apex) while keeping a reasonable CTDIvol (volumetric computed tomography dose index) (13). RDCT was performed with a fixed kVp (100 kV), a fixed current-time product (11 mAs), and a 1.172 pitch factor. Other scanning parameters were held constant for both standard-of-care CT and RDCT: 350 mm field of view, 64× 0.625 mm detector collimation, and 0.4-second rotation time. Reconstruction section thickness was 0.9 mm.
Academic Radiology, Vol ■, No ■, ■■ 2016
Reconstruction Process
Standard-of-care CT raw data were reconstructed according to manufacturer-advised protocol, using two different kernels (filter YA for lung parenchyma, filter B for soft tissue) both with iterative reconstruction iDose4 level 5. RDCT raw data were reconstructed using FBP (B kernel), iDose4 level 7 (B kernel), and IMR level 2 (body soft kernel). Given the fact that at reduced dose levels image noise is too high with sharp kernels usually employed for lung parenchyma, a soft tissue kernel was used for both soft tissue and parenchyma analyses (14,15). Furthermore, the reconstruction kernels available with IMR are different from the ones used with FBP and iDose4, and the body soft kernel was chosen for IMR images, being the closest to B kernel according to manufacturer information. Reconstructions using IMR were performed using a prototype in Philips factory (Best, The Netherlands), as there was no commercially available software at the time of the study. As a result, there were five data sets (two standard-of-care and three RDCT reconstructions) for each included patient. Images were anonymized and coded for reading. Radiation Dose Assessment
For each patient, the volumetric CT dose index (CTDIvol) and the dose length product (DLP) were recorded. Effective dose was estimated by multiplying the DLP with the chestspecific conversion coefficient (0.014 mSv/mGy·cm) (16). Sizespecific dose estimation was also calculated by multiplying CTDIvol with a conversion factor f depending on the effective diameter (17).
REDUCING RADIATION DOSE AT CHEST CT
the lesion assessment. Those data sets were subsequently withdrawn from statistical analysis. Standard-of-care chest CT was assessed jointly by the two readers (consensus reading). The three reconstructions of RDCT (FBP, iDose4, and IMR) were independently assessed by the two readers. Sessions were separated by a minimum of 2 weeks to minimize recall bias. In case of disagreement, a consensus reading between these two readers was made afterwards during a new session. Images were displayed with a lung window setting (level = −600 HU, width = 1600 HU) and a soft window setting (level = 40 HU, width = 400 HU), but the radiologists were allowed to modify the window settings. Five parameters were assessed with a 4or 5-point scale: subjective noise, normal anatomic structure detectability (subpleural vessels, main fissures), artifacts (helical artifacts, streak artifacts, beam-hardening artifacts, and blotchy pixilated appearance), lesion conspicuity, and diagnostic confidence. In case of multiple parenchymal abnormalities, grading for lesion conspicuity was attributed considering the less visible lesion. Diagnostic confidence was rated considering the evolution assessment. Grading scales are detailed in Table 2. For lesion detection study, standard-of-care CT was considered as the reference. Observers were asked to record any lesion in each pulmonary lobe, the left upper lobe being divided into culmen (upper part) and lingula (lower part). Lung lesions were defined according to the glossary of terms for thoracic imaging from the Fleischner Society (21). The exact number of micronodules (<3 mm) was recorded if fewer or equal to 5, conversely to cases with more than 5 micronodules, for which the mention >5 was noted. Finally, evolution assessment (lesion stability, progression, or regression) was evaluated compared to the initial evaluation.
Objective Measurements
The mean Hounsfield unit (HU) and its standard deviation representing objective noise were measured for all the CT image series by one radiologist on a clinical workstation (IntelliSpace Portal, Philips Healthcare). Circular regions of interest were drawn inside the trachea immediately over the carina and in the descending thoracic aorta at the carina level. The ROIs were duplicated in the same location for each image reconstruction. For standard-of-care acquisition, aorta noise was measured using soft tissue kernel and trachea noise using lung parenchyma kernel. Measurements were repeated three times in three adjacent slices and the mean value was considered. Subjective Measurements
The image quality criteria chosen in the study have previously been totally or partially used in the literature (18–20). Subjective image quality of all data sets was independently assessed by two radiologists (C. de M.-M., C. de B) with 6 and 15 years of experience in general radiology. They were first trained together on two image data sets to get themselves familiar with the grading of subjective measurements and with
Statistical Analysis
The results were expressed as median [interquartile ranges] for continuous and semiquantitative variables, and described as numbers and percentages for qualitative and ordinal variables. For subjective quality criteria and lesion detection, analysis was performed with the consensus ratings. Objective noise was compared between each reconstruction using pairwise Wilcoxon test, and subjective image quality scores were compared using chi-square trend tests. For lesion detection, each pulmonary lobe was considered as an independent observation; 216 lobes (36 patients × 6 lobes) were then analyzed. Sensitivities and specificities for each RDCT were assessed using standard-of-care CT as the gold standard. Comparisons between sensitivities were performed using adjusted Wald confidence intervals (CIs) for a difference of proportions with matched pairs (22). Interobserver agreement between the two radiologists for the subjective quality rating and the lesion detection was calculated using percentage of agreement. Statistical analysis was performed using R software (R 2.14.0, R Foundation for Statistical Computing, Vienna, Austria). P < 0.05 (two-sided) was the criterion for statistical significance in all analyses. 3
Academic Radiology, Vol ■, No ■, ■■ 2016
DE MARGERIE-MELLON ET AL
TABLE 2. Grading for Subjective Image Quality 1 Subjective noise Normal anatomic structure detectability* Artifacts†
Lesion conspicuity Diagnostic confidence
2
3
4
5
Above average noise Unacceptable visualization of small structures Artifacts affecting diagnostic information Probable artifact mimicking a lesion Poor confidence
Unacceptable image noise Not applicable
Minimal image noise Excellent visualization
Less than average noise
Average image noise
Above average visibility
Acceptable visibility
No artifacts
Minor artifacts not interfering with diagnosis decision making Well-seen lesion with poorly demarcated margins
Major artifacts but possible diagnosis
Probably confident
Confident only for limited clinical situations
Well-seen lesion with sharp margins Completely confident
Subtle lesion
Not applicable
Definite artifact mimicking a lesion Not applicable
* Subpleural vessels and main fissures. † Helical artifacts, streak artifacts, beam-hardening artifacts, and blotchy pixilated appearance.
Figure 1. Boxplots representing the trachea and aorta noise of the standardof-care protocol and of the three reconstructions of reduced-dose chest computed tomography. Boxes indicate interquartile range; whiskers extend the data extremes, and horizontal lines the median value. All pairwise comparisons were statistically significant.
RESULTS Technical Parameters
FBP and iDose4 reconstructions took a few seconds. IMR reconstruction time was 126 seconds for 777 images. The median [interquartile range] for CTDIvol, DLP, sizespecific dose estimation, and estimated effective radiation dose were 3.4 mGy [3.1–3.9], 129.0 mGy·cm [112.8–144.0], 4.5 mGy [4.2–5.2], and 1.8 mSv [1.6–2.0], respectively, for the standard-of-care protocol, and 0.43 mGy for all patients, 16.0 mGy·cm [15.5–16.9], 0.59 mGy [0.55–0.62], and 0.22 mSv [0.22–0.24], respectively, for RDCT (P < 0.0001). 4
Objective Measurements
Noise measurements are presented in Figure 1. The aorta and trachea noise were, respectively, 13.2 [12.0–15.4] HU and 48.8 [45.2–53.0] HU for the standard-of-care protocol, 61.5 [51.6–83.6] HU and 42.8[38.3–50.8] HU with FBP, 26.7 [23.7–33.0] HU and 28.0 [21.3–32.3] HU with iDose4, and 14.0 [11.8–16.7] HU and 13.3 [9.3–16.3] HU with IMR for RDCT. Measurements were significantly different between all pairs with RDCT (P < 0.0001 for all pairwise comparison), showing a 77% median decrease in aorta noise and a 70% median decrease in trachea noise between FBP and IMR. Concerning standard-of-care CT and IMR reconstruction of
Academic Radiology, Vol ■, No ■, ■■ 2016
REDUCING RADIATION DOSE AT CHEST CT
RDCT, aorta noise was significantly lower (P = 0.03) and trachea noise was significantly higher (P < 0.0001) with standard-of-care CT. Subjective Measurements
Subjective scores are detailed in Table 3. For RDCT, normal anatomic structure detectability score was identical with the three reconstructions. Other subjective criteria (subjective noise, artifacts, lesion conspicuity, and diagnostic confidence) were significantly different between all pairs, lower with iDose4 (corresponding to higher performances) compared to FBP, and lower with IMR compared to iDose4 (P < 0.0001 for all pairwise comparison). Concerning standard-of-care CT and IMR reconstruction of RDCT, all measurements were also significantly different: standard-of-care CT scores were lower than IMR scores, except for subjective noise, which was lower with IMR (P < 0.001). Examples are provided in Figures 2–4. For all RDCT reconstructions, the percentages of agreement between the two readers range from 69% to 94% for artifacts and subjective noise, and from 50% to 94% for diagnostic confidence, and from 31% to 56% for lesion conspicuity. The percentage of agreement for normal structure detectability was 100% for all RDCT reconstructions. Sixty-one countable nodules were identified on standardof-care CT. FBP and iDose4 reconstructions of RDCT identified, respectively, 53 and 55 countable nodules without
any false-positive. IMR reconstruction identified 56 countable nodules, including one false-positive. Thus, detectability rates (the number of true-positive nodules divided by the number of nodules detected on standard-of-care CT) were 87% (95% CI: 76–94) for FBP and 90% (95% CI: 80–96) for iDose4 and IMR. These differences were not statistically significant. The percentage of agreement for the nodule number per lobe was higher than 90% in all reconstructions. Standard-of-care examination identified 83 countable micronodules among all patients. Forty micronodules were identified with FBP reconstruction of RDCT, including one false-positive, 57 with iDose4 without any false-positive, and 65 with IMR including one false-positive. Countable micronodule detectability rates were 47% (95% CI: 36–58), 69% (95% CI: 57–78), and 77% (95% CI: 66–86), respectively. Countable micronodules detection was significantly higher with IMR and iDose4 than with FBP. The percentage of agreement for the micronodule number per lobe was higher than 80% in all reconstructions. Table 4 summarizes the performances in lesion detection for the three reconstructions of RDCT compared to standardof-care CT (reference) considering 216 lobes. For groundglass opacities, detection sensitivities of IMR and iDose4 were significantly higher than FBP. For emphysema detection, sensitivity of IMR was significantly higher than FBP and iDose4. Other differences in sensitivities among the three reconstructions were found to be nonsignificant. The percentage of
TABLE 3. Subjective Quality Scores (Consensus Reading) Ultra Reduced-dose CT
Subjective noise (1–5) 2 3 4 5 Normal anatomic structure detectability (1–4) 2 3 4 Artifacts (1–4) 1 2 3 4 Lesion conspicuity (1–5) 1 2 3 Diagnostic confidence (1–4) 1 2 3
Standard-of-care CT
FBP
iDose4
IMR
7 (19) 29 (81) 0 (0) 0 (0)
0 (0) 0 (0) 18 (50) 18 (50)
0 (0) 10 (28) 26 (72) 0 (0)
32 (89) 4 (11) 0 (0) 0 (0)
10 (28) 26 (72) 0 (0)
0 (0) 0 (0) 36 (100)
0 (0) 0 (0) 36 (100)
0 (0) 0 (0) 36 (100)
21 (58) 15 (42) 0 (0) 0 (0)
0 (0) 2 (6) 32 (88) 2 (6)
0 (0) 25 (69) 11 (31) 0 (0)
0 (0) 36 (100) 0 (0) 0 (0)
35 (97) 1 (3) 0 (0)
0 (0) 14 (39) 22 (61)
0 (0) 31 (86) 5 (14)
25 (69) 11 (31) 0 (0)
36 (100) 0 (0) 0 (0)
0 (0) 20 (56) 16 (44)
0 (0) 35 (97) 1 (3)
15 (42) 21 (58) 0 (0)
CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction. Data are expressed as count (percentages).
5
DE MARGERIE-MELLON ET AL
Figure 2. Example of normal anatomic lung structure detectability scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show pulmonary fissures, which are well seen as a complete thin line in (a), but which are ill-defined in (b), (c), and (d). Detectability of normal anatomic lung structures was ranked 3 in (a), and ranked 4 in (b), (c), and (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
agreement between the two readers was higher than 94% for all lesions in the three reconstructions, except for groundglass opacities (87%–93%). There were 12 pleural effusions on standard-of-care CT, which were all detected with iDose4 and IMR RDCT reconstructions. One small pleural effusion was missed with FBP. Over the four pericardial effusions diagnosed on standardof-care CT, two were detected on FBP reconstruction and three were detected on iDose4 and IMR. Evolution assessment (stability, progression, and regression) was made on the standard-of-care CT by comparison with the previous examination. FBP reconstruction of RDCT provided the correct evolution assessment in 34 cases over 36. With IMR and iDose4 reconstructions, the evolution assessment was correct in all cases. DISCUSSION Lowering radiation dose induced by CT scan examinations without compromising image quality is a critical issue today, in order to 6
Academic Radiology, Vol ■, No ■, ■■ 2016
respect the “as low as reasonably achievable” (ALARA) principle. The current prospective study suggests that chest RDCT (CTDI = 0.43 mGy, effective dose of 0.22 mSv) reconstructed with IR allows for an accurate assessment of the evolution of lung affections in a selected population. As expected, objective noise was lower in iDose4 compared to FBP and higher in iDose4 compared to IMR, as well as the artifacts and the subjective noise ratings. Because of the low dose level, artifacts were mainly due to beam hardening. IMR is the latest algorithm that aims to improve noise and artifact suppression by modeling not only noise statistics but also system geometry. Noise and artifacts are important factors that alter image quality, and in parallel with their decrease we observed an improvement of diagnostic confidence when using iDose4 and IMR. Diagnostic confidence was greater with IMR than with iDose4 (“completely confident” in 42% and 0% of cases, respectively); however, readers were mostly “completely confident” or “probably confident” with both reconstructions (100% or 97%, respectively, for IMR and iDose4), which can be considered as clinically acceptable. On the opposite, diagnostic confidence of FBP reconstruction (“confident only for limited clinical situations” in 44% of the cases) was insufficient. However, the loss of small anatomic elements like lung fissures was observed in all reconstruction methods. IR algorithms did not allow subjective improvement of spatial resolution at this very low dose level. Concerning lesion detection, no significant difference was observed among the three reconstruction methods for nodule detectability, which was close to 90% in all cases. Nodules are high contrast lesions, and their sizes (>3 mm) made them visible even with high background noise. Conversely, for countable micronodule detectability, IMR and iDose4 showed better performances than FBP but remained below 80%. Likewise, ground-glass opacity detectability was close to 100% with iDose4 and IMR and much higher than with FBP. It is likely that high background noise in FBP significantly hampered micronodules and ground-glass opacity identification. Finally, emphysema detectability was performing well with IMR but insufficient with iDose4 and FBP. This appears mainly related to the reduction of streak artifacts at the level of the shoulders, which may preclude visualization of emphysema. Although emphysema is usually not relevant data in case of immunocompromised lung disease, other low density abnormalities (mosaic perfusion, thin-walled cysts) can be encountered in such pathologies. At the clinical level, accurate evolution assessment was obtained in all cases with both iDose4 and IMR. On the contrary, RDCT images obtained with FBP led to misdiagnosis in two cases. The estimated effective dose delivered by RDCT protocol (0.22 mSv) in our study was much lower than the standard low-dose CT scans (0.8–1.5 mSv) (23), coming close to the level of a posteroanterior and a lateral chest x-ray (0.05– 0.24 mSv) (1). Moreover, IMR reconstruction time was low (126 seconds for 777 images), close to those reported in the literature (from 80 seconds for CT colonography [24] to 217 seconds for a chest CT with 777 images [4]).
Academic Radiology, Vol ■, No ■, ■■ 2016
REDUCING RADIATION DOSE AT CHEST CT
Figure 3. Example 1 of lesion conspicuity scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show a ground-glass opacity in the left lung. Definition of this opacity grows from (b) to (d). Lesion conspicuity was ranked 1 in (a), 3 in (b), 2 in (c), and 1 in (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Figure 4. Example 2 of lesion conspicuity scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show emphysema of the right apex. Definition of emphysema grows from (b) to (d). Lesion conspicuity was ranked 1 in (a), 3 in (b), 2 in (c), and 1 in (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Similar dose reductions were achieved in several studies. Some of them (25–27) used MBIR algorithm (model-based iterative reconstruction), a fully iterative algorithm from GE Healthcare (Waukesha, WI, USA). Like IMR, it incorporates a system geometry modeling in addition to statistics model. However, MBIR requires high computational power, resulting in long processing time of up to 1 hour (25,28), which may limit routine applicability. Concerning lesion detection, several publications (26,27) only focused on nodule detection in the context of lung cancer screening. They showed the great potential of MBIR with RDCT images (effective dose: 0.17–0.20 mSv). Neroladaki et al. (25) confirmed MBIR high diagnostic performances for nodules, but demonstrated its limitations at this dose level (mean effective dose: 0.16 mSv) for the detection of ground-glass opacities and emphysema, contrary to the results obtained in the current study with
IMR. Some studies (29,30) with a similar dose level used SAFIRE algorithm (sinogram-affirmed iterative reconstruction, Siemens Healthcare, Forchheim, Germany). SAFIRE is an IR method that processes both in raw data and in image domains. Lee et al. (30) demonstrated that reduced-dose (mean effective dose: 0.29 mSv) chest CT with SAFIRE generates sufficient image diagnostic quality in 90% of the cases. Like with MBIR, performances were also lower in case of lesions of decreased attenuation, like emphysema and ground-glass opacities. Finally, Nagatani et al. (31) used AIDR 3D hybrid algorithm (adaptative iterative dose reduction using threedimensional processing (Toshiba Medical Systems Corporation, Otawara, Japan) with reduced dose (0.29 mSv) to assess lung nodule detection. They showed comparable lung nodule detection to reference CT, except for smaller ground-glass nodules. 7
Academic Radiology, Vol ■, No ■, ■■ 2016
8
0.95 (0.87–0.99) 0.96 (0.91–0.98) 0.9 (0.77–0.97) 0.99 (0.96–1) 1 (0.89–1) 1 (0.98–1) 0.86 (0.64–0.97) 1 (0.98–1) 0.6 (0.32–0.84) 1 (0.98–1) 0.57 (0.18–0.9) 1 (0.98–1) 77 45 32 18 9 4 0.96 (0.89–0.99) 0.98 (0.94–1) 0.67 (0.52–0.8) 0.99 (0.97–1) 0.94 (0.79–0.99) 1 (0.98–1) 1 (0.84–1) 1 (0.98–1) 0.73 (0.45–0.92) 1 (0.98–1) 0.57 (0.18–0.9) 1 (0.98–1) 75 33 30 21 11 4 0.67 (0.55–0.77) 0.98 (0.94–0.97) 0.65 (0.50–0.78) 0.99 (0.96–1) 0.94 (0.79–0.99) 1 (0.98–1) 0.95 (0.76–1) 0.99 (0.96–1) 0.53 (0.27–0.79) 1 (0.98–1) 0.43 (0.01–0.82) 1 (0.98–1) 53 33 30 22 8 3 75 48 32 21 15 7
CI, confidence interval; CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Lesion
Ground-glass opacity Emphysema Consolidation Linear atelectasis Bronchiectasis Septal thickening
Sensitivity (95% CI) Sensitivity (95% CI) Sensitivity (95% CI) Number of Lesion with Number of Standard-of- Lesion with care CT RDCT
TABLE 4. Lesion Detection
FBP
Specificity (95% CI)
Number of Lesion with RDCT
iDose4
Specificity (95% CI)
Number of Lesion with RDCT
IMR
Specificity (95% CI)
DE MARGERIE-MELLON ET AL
To our knowledge, few publications (6,7) studied IMR clinical application for the chest. Both studies demonstrated great potential for IMR in image quality. Khawaja et al. (7) also assessed lesion detection in a group of 10 patients. All lesions could be detected with IMR. Dose level was higher for the low-dose protocol (0.9 mSv), but the clinical context was different. Indeed, to give clinical relevance to our study, we focused on a specific group of patients who underwent a fulldose chest CT examination in the previous 2 months. The main question of the control CT scan was only to assess evolution of a known lung disease. We showed that RDCT reconstructed with iDose4 and IMR was able to answer this question. Further studies are needed to evaluate IMR performances at this reduced dose level for initial diagnosis, owing to the fact that subtle details could be suppressed. This study has some limitations. First, the iDose4 levels for standard-of-care CT and RDCT were subjectively selected, as our preferred compromise for noise reduction and usual image texture preservation for the selected dose level and according to the settings used in our department, taking into account that no consensus has been reported in the literature regarding the optimal level of this hybrid IR (7,15,32,33). Second, blinding to the reconstruction mode was not applicable. Indeed, the visual aspects of FBP, iDose4, and IMR images were easily recognizable. It may have induced some bias by favoring the most recent technology. Third, quality image scoring and lesion detection are intrinsically subjective and difficult to transpose to other studies, which is why we associated objective measurements. Fourth, the sample size was small because of the strict inclusion criteria and the limited duration of the study. Therefore, most of the comparisons can suffer from a lack of power. Those inclusion criteria were chosen to obtain a homogeneous group of patients in order to give immediate clinical relevance to our results. Finally, as expected in a population of patients treated for hematologic malignancy, the median body mass index (BMI) of the cohort was low (22 kg/m2), with only four patients with a BMI > 25 kg/m2. In the study by Lee et al. (30) using RDCT with similar dose (mean effective dose: 0.29 mSv) reconstructed with SAFIRE, images of diagnostic quality were obtained in more than 95% of cases for patients with a BMI of less than 25 kg/m2, whereas this rate decreased to 70% in patients with BMI > 25 kg/m2. It is probable that our RDCT protocol parameters should be adapted for patients with higher BMI. Nevertheless, these parameters could also benefit a pediatric population. To conclude, our study shows that chest CT may be achievable with a dose close to 1 posteroanterior and lateral chest x-ray using the last generation of IR for lung affection monitoring in a specific population usually requiring repeated followup studies. Similar evolution assessment can be obtained for all patients either with iDose4 and IMR. Nevertheless, IMR appears to improve diagnostic confidence and lesion conspicuity, as well as detection of reduced-contrast lesions like emphysema. Contrary to other recent IR algorithms, reconstruction time is compatible with clinical exercise. However, it has to
Academic Radiology, Vol ■, No ■, ■■ 2016
be demonstrated that relevant information is not missing with such dose and reconstructions for initial diagnosis of lung disorders. ACKNOWLEDGMENTS This work was supported by grants from Société Française de Radiologie and from Fonds d’Etude et de Recherche du Corps Médical. REFERENCES 1. Mettler FA, Jr, Bhargavan M, Faulkner K, et al. Radiologic and nuclear medicine studies in the United States and worldwide: frequency, radiation dose, and comparison with other radiation sources–1950-2007. Radiology 2009; 253:520–531. doi:10.1148/radiol.2532082010. 2. Brenner DJ, Hall EJ. Computed tomography—an increasing source of radiation exposure. N Engl J Med 2007; 357:2277–2284. doi:10.1056/ NEJMra072149. 3. Brooks RA, Di Chiro G. Theory of image reconstruction in computed tomography. Radiology 1975; 117:561–572. doi:10.1148/117.3.561. 4. Mehta D, Thompson R, Morton T, et al. Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality. Med Phys Int J 2013; 1:147–155. 5. Kim H, Park CM, Song YS, et al. Influence of radiation dose and iterative reconstruction algorithms for measurement accuracy and reproducibility of pulmonary nodule volumetry: a phantom study. Eur J Radiol 2014; 83:848–857. doi:10.1016/j.ejrad.2014.01.025. 6. Yuki H, Oda S, Utsunomiya D, et al. Clinical impact of model-based type iterative reconstruction with fast reconstruction time on image quality of low-dose screening chest CT. Acta Radiol 2015; 57:295–302. doi:10.1177/ 0284185115575537. 7. Khawaja RDA, Singh S, Gilman M, et al. Computed tomography (CT) of the chest at less than 1 mSv: an ongoing prospective clinical trial of chest CT at submillisievert radiation doses with iterative model image reconstruction and iDose4 technique. J Comput Assist Tomogr 2014; 38:613– 619. doi:10.1097/RCT.0000000000000087. 8. Koh TT, Colby TV, Müller NL. Myeloid leukemias and lung involvement. Semin Respir Crit Care Med 2005; 26:514–519. doi:10.1055/s-2005922034. 9. Georgiadou SP, Sipsas NV, Marom EM, et al. The diagnostic value of halo and reversed halo signs for invasive mold infections in compromised hosts. Clin Infect Dis 2011; 52:1144–1155. doi:10.1093/cid/cir122. 10. Bissinger AL, Einsele H, Hamprecht K, et al. Infectious pulmonary complications after stem cell transplantation or chemotherapy: diagnostic yield of bronchoalveolar lavage. Diagn Microbiol Infect Dis 2005; 52:275– 280. doi:10.1016/j.diagmicrobio.2005.03.005. 11. Vahid B, Marik PE. Infiltrative lung diseases: complications of novel antineoplastic agents in patients with hematological malignancies. Can Respir J 2008; 15:211. 12. Nakaseko C, Ozawa S, Sakaida E, et al. Incidence, risk factors and outcomes of bronchiolitis obliterans after allogeneic stem cell transplantation. Int J Hematol 2011; 93:375–382. doi:10.1007/s12185-011-0809-8. 13. Tack D, Kalra MK, Genevois PA. ALARA concept for MDCT optimization: what is reasonable, what is achievable? In: Radiation dose multidetector CT. Berlin Heidelberg: Springer, 2012; 231–258. Editors: Tack D, Kalra MK, Gevenois PA. 14. Wang Y, de Bock GH, van Klaveren RJ, et al. Volumetric measurement of pulmonary nodules at low-dose chest CT: effect of reconstruction setting on measurement variability. Eur Radiol 2010; 20:1180–1187. doi:10.1007/s00330-009-1634-9. 15. Nakajo C, Heinzer S, Montandon S, et al. Chest CT at a dose below 0.3 mSv: impact of iterative reconstruction on image quality and lung analysis. Acta Radiol 2015; 57:311–317. doi:10.1177/0284185115578469.
REDUCING RADIATION DOSE AT CHEST CT
16. Christner JA, Kofler JM, McCollough CH. Estimating effective dose for CT using dose-length product compared with using organ doses: consequences of adopting International Commission on Radiological Protection publication 103 or dual-energy scanning. AJR Am J Roentgenol 2010; 194:881–889. doi:10.2214/AJR.09.3462. 17. American Association of Physicist in Medicine. Size-specific dose estimates (SSDE) in pediatric and adult body CT examinations. Task Group 204, 2011. 18. Singh S, Kalra MK, Gilman MD, et al. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study. Radiology 2011; 259:565–573. doi:10.1148/radiol.11101450. 19. Singh S, Kalra MK, Moore MA, et al. Dose reduction and compliance with pediatric CT protocols adapted to patient size, clinical indication, and number of prior studies. Radiology 2009; 252:200–208. doi:10.1148/radiol.2521081554. 20. Yamada Y, Jinzaki M, Hosokawa T, et al. Dose reduction in chest CT: comparison of the adaptive iterative dose reduction 3D, adaptive iterative dose reduction, and filtered back projection reconstruction techniques. Eur J Radiol 2012; 81:4185–4195. doi:10.1016/j.ejrad .2012.07.013. 21. Hansell DM, Bankier AA, MacMahon H, et al. Fleischner Society: glossary of terms for thoracic imaging. Radiology 2008; 246:697–722. doi:10.1148/radiol.2462070712. 22. Agresti A, Min Y. Simple improved confidence intervals for comparing matched proportions. Stat Med 2005; 24:729–740. doi:10.1002/sim.1781. 23. Larke FJ, Kruger RL, Cagnon CH, et al. Estimated radiation dose associated with low-dose chest CT of average-size participants in the national lung screening trial. Am J Roentgenol 2011; 197:1165–1169. doi:10.2214/AJR.11.6533. 24. Lambert L, Ourednicek P, Briza J, et al. Sub-milliSievert ultralow-dose CT colonography with iterative model reconstruction technique. PeerJ 2016; 4:e1883. http://dx.doi.org/10.7717/peerj.1883. 25. Neroladaki A, Botsikas D, Boudabbous S, et al. Computed tomography of the chest with model-based iterative reconstruction using a radiation exposure similar to chest x-ray examination: preliminary observations. Eur Radiol 2012; 23:360–366. doi:10.1007/s00330-012-2627-7. 26. Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for ultralow-dose chest CT. Invest Radiol 2013; 48:206– 212. doi:10.1097/RLI.0b013e31827efc3a. 27. Yamada Y, Jinzaki M, Tanami Y, et al. Model-based iterative reconstruction technique for ultralow-dose computed tomography of the lung: a pilot study. Invest Radiol 2012; 47:482–489. doi:10.1097/ RLI.0b013e3182562a89. 28. Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. Eur Radiol 2012; 22:1613–1623. doi:10.1007/s00330-012-2452-z. 29. Kim Y, Kim YK, Lee BE, et al. Ultra-low-dose CT of the thorax using iterative reconstruction: evaluation of image quality and radiation dose reduction. AJR Am J Roentgenol 2015; 204:1197–1202. doi:10.2214/ AJR.14.13629. 30. Lee SW, Kim Y, Shim SS, et al. Image quality assessment of ultra lowdose chest CT using sinogram-affirmed iterative reconstruction. Eur Radiol 2014; 24:817–826. doi:10.1007/s00330-013-3090-9. 31. Nagatani Y, Takahashi M, Murata K, et al. Lung nodule detection performance in five observers on computed tomography (CT) with adaptive iterative dose reduction using three-dimensional processing (AIDR 3D) in a Japanese multicenter study: comparison between ultra-low-dose CT and low-dose CT by receiver-operating characteristic analysis. Eur J Radiol 2015; 84:1401–1412. doi:10.1016/j.ejrad.2015.03.012. 32. Laqmani A, Buhk JH, Henes FO, et al. Impact of a 4th generation iterative reconstruction technique on image quality in low-dose computed tomography of the chest in immunocompromised patients. Rofo 2013; 185:749–757. doi:10.1055/s-0033-1335577. 33. Willemink MJ, Borstlap J, Takx RAP, et al. The effects of computed tomography with iterative reconstruction on solid pulmonary nodule volume quantification. PLoS ONE 2013; 8:e58053. doi:10.1371/journal. pone.0058053.
9
Original Investigation
Reducing Radiation Dose at Chest CT: Comparison Among Model-based Type Iterative Reconstruction, Hybrid Iterative Reconstruction, and Filtered Back Projection Constance de Margerie-Mellon, MD, Cédric de Bazelaire, MD, PhD, Claire Montlahuc, MD, Jérôme Lambert, MD, PhD, Antoine Martineau, PhD, Philippe Coulon, PhD, Eric de Kerviler, MD, Catherine Beigelman, MD Rationale and Objectives: The study aimed to evaluate the performances of two iterative reconstruction (IR) algorithms and of filtered back projection (FBP) when using reduced-dose chest computed tomography (RDCT) compared to standard-of-care CT. Materials and Methods: An institutional review board approval was obtained. Thirty-six patients with hematologic malignancies referred for a control chest CT of a known lung disease were prospectively enrolled. Patients underwent standard-of-care scan reconstructed with hybrid IR, followed by an RDCT reconstructed with FBP, hybrid IR, and iterative model reconstruction. Objective and subjective quality measurements, lesion detectability, and evolution assessment on RDCT were recorded. Results: For RDCT, the CTDIvol (volumetric computed tomography dose index) was 0.43 mGy·cm for all patients, and the median [interquartile range] effective dose was 0.22 mSv [0.22–0.24]; corresponding measurements for standard-of-care scan were 3.4 mGy [3.1–3.9] and 1.8 mSv [1.6–2.0]. Noise significantly decreased from FBP to hybrid IR and from hybrid IR to iterative model reconstruction on RDCT, whereas lesion conspicuity and diagnostic confidence increased. Accurate evolution assessment was obtained in all cases with IR. Emphysema identification was higher with iterative model reconstruction. Conclusion: Although iterative model reconstruction offered better diagnostic confidence and emphysema detection, both IR algorithms allowed an accurate evolution assessment with an effective dose of 0.22 mSv. Key Words: Computed tomography; lung; radiation dose reduction; iterative reconstruction; image noise. © 2016 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
Acad Radiol 2016; ■:■■–■■ From the Department of Radiology, INSERM UMR_S1165, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, (C. de M.-M., C. de B., E. de K.); Department of Informatics and Biostatistics, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, (C.M., J.L.); INSERM, ECSTRA team, CRESS-UMR_S 1153, 75010 Paris, (C.M., J.L.); Department of Nuclear Medicine, University Paris Diderot, Sorbonne Paris-Cité, AP-HP, Saint Louis Hospital, 1 avenue Claude Vellefaux, 75 010 Paris, France (A.M.); Philips Healthcare, 33 rue de Verdun, 92156 Suresnes Cedex, France (P.C.); Department of Radiology, Vaudois University Hospital, rue du Bugnon 46, 1011 Lausanne, Switzerland (C.B.). Received January 8, 2016; revised May 30, 2016; accepted May 31, 2016. Address correspondence to: C. de M.-M. e-mail: [email protected] © 2016 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.acra.2016.05.019
O
ver the past decades, the average annual radiation dose delivered has raised significantly worldwide (1). Reducing computed tomography (CT) radiation dose has become a major concern because of the potential risk of radiation-induced cancer (2). Various strategies have been developed to reduce radiation exposure while maintaining image quality, including tube potential selection and tube current modulation. Unfortunately, radiation dose decrease is limited when using filtered back projection (FBP) reconstruction. FBP assumes that each pixel value perfectly represents the attenuation of the object at this location, setting aside system hardware details and photon noise statistic information (3). Because of those idealized assumptions, a lowered radiation dose is accompanied by an increased noise and often increased artifacts. 1
Academic Radiology, Vol ■, No ■, ■■ 2016
DE MARGERIE-MELLON ET AL
To overcome those difficulties, iterative reconstruction (IR) has been developed by CT manufacturers, enabling reduction in radiation dose without compromising spatial resolution by decreasing image noise and artifacts. Several algorithms have been developed by different companies, working either in the raw data space, image space, or both. Among them, hybrid generation IR called iDose4 (Philips Healthcare, Best, The Netherlands) is an algorithm that iterates in the raw data and image domains. More recently, a model-based type reconstruction called iterative model reconstruction (IMR, Philips Healthcare) has been developed. Compared to the previous algorithm, it takes into account data and image statistics but also considers system models (4). To date, few data are published about its potential applications in chest CT (5–7). We aimed to evaluate iDose4 and IMR performances in clinical practice. We hypothesized that reduced-dose chest computed tomography (RDCT) reconstructed with IR (iDose4 and IMR) for lung disease follow-up would enable obtaining the same final conclusion compared to a standard-of-care chest CT in a population of patients with hematologic malignancies. This population of patients was chosen because they frequently undergo successive chest CT examinations for their disease assessment, including lung tumoral spread (8), intercurrent complications, especially pulmonary infectious disorders (9,10) due to their immunocompromised status, or drug-induced lung disease (11,12). Moreover, compared to healthy subjects, such patients have high morbidity and mortality. Chest CT scan is here the best imaging modality for diagnostic orientation, biological sampling guidance, and follow-up under treatment. Although the potential risk of radiation-induced cancer seems low compared to the consequences of the hematologic disease and its treatments (chemotherapy and radiotherapy), such knowledge might help modify our routine practice even in the general population each time a follow-up is required, especially in young patients. We also wanted to evaluate FBP, iDose4, and IMR technique regarding objective and subjective image quality, and their detection potential when using RDCT.
MATERIALS AND METHODS Population
This prospective single-center study was approved by the ethics committee of our institution. According to local legislation, potential subjects were free to refuse to participate, but no written consent was required if they accepted. From February 2014 to June 2014, patients with hematologic malignancy and lung disease coming for a control unenhanced chest CT were screened. Inclusion criteria were as follows: age >18 years, available previous chest CT performed within the last 2 months (used as baseline), ability to understand the study protocol and its implications, and capacity to perform breath-holding. Patients who were under 18, who were unable to understand information for the participation to the study or to perform breath-holding, or who underwent previous chest CT more than 2 months before were 2
TABLE 1. Characteristics of Included Patients Characteristics
Patients (n = 36)
Age (y)* 59 [28.5–66] Sex ratio (M/F) 21/15 21.9 [19.8–24.2] Body mass index (kg/m2)* Effective diameter† (cm)* 27 [25.6–29.0] 28 [14–50] Delay since the initial chest computed tomography (d)* Initial pulmonary diagnosis Invasive aspergillosis (proven or probable) 7 (19%) Bacterial pneumonia (proven or probable) 7 (19%) Nonspecific lesions, probabilistic antifungal 11 (31%) or antibiotic therapy Nonspecific lesions, no therapy 7 (19%) All trans-retinoic acid syndrome 1 (3%) Organizing pneumonia 1 (3%) Lymphoma tumoral extension 1 (3%) Alveolar proteinosis 1 (3%) * Median [interquartile range] n. Effective diameter = AP × LAT , where AP represents the anterior-posterior dimension and LAT the lateral dimension of the patient. †
excluded. Each selected subject was given oral and written information in simple language about the objective, methods, and risks of study participation. Thirty-eight patients were selected to participate in the study. No patient opposed to its participation in the study. Readers used data sets of two patients for training. As a result, 36 patients were included in the study. Demographic and morphologic information is summarized in Table 1. Final diagnosis of the lung disease was based on the combination of clinical and radiological data, including the control chest CT. Imaging Protocol
All CT examinations were performed using a 64-row singlesource CT (Brilliance CT, Philips Healthcare). Each patient underwent two unenhanced chest acquisitions with the same length of helical run: first a standard-of-care CT, immediately followed by a RDCT. Acquisitions were performed in the supine position, with arms elevated, during a single breath-hold at full inspiration. Standard-of-care chest CT was performed according to our institution protocol for follow-up with a 140 kVp tube voltage, automatic exposure control (Z-DOM), and a 1.015 pitch factor. A combination of high tube voltage and limited currenttime product is usually employed at our center for this protocol to ensure the best image quality (especially to reduce the streak artifacts due to the high attenuation of shoulders in the apex) while keeping a reasonable CTDIvol (volumetric computed tomography dose index) (13). RDCT was performed with a fixed kVp (100 kV), a fixed current-time product (11 mAs), and a 1.172 pitch factor. Other scanning parameters were held constant for both standard-of-care CT and RDCT: 350 mm field of view, 64× 0.625 mm detector collimation, and 0.4-second rotation time. Reconstruction section thickness was 0.9 mm.
Academic Radiology, Vol ■, No ■, ■■ 2016
Reconstruction Process
Standard-of-care CT raw data were reconstructed according to manufacturer-advised protocol, using two different kernels (filter YA for lung parenchyma, filter B for soft tissue) both with iterative reconstruction iDose4 level 5. RDCT raw data were reconstructed using FBP (B kernel), iDose4 level 7 (B kernel), and IMR level 2 (body soft kernel). Given the fact that at reduced dose levels image noise is too high with sharp kernels usually employed for lung parenchyma, a soft tissue kernel was used for both soft tissue and parenchyma analyses (14,15). Furthermore, the reconstruction kernels available with IMR are different from the ones used with FBP and iDose4, and the body soft kernel was chosen for IMR images, being the closest to B kernel according to manufacturer information. Reconstructions using IMR were performed using a prototype in Philips factory (Best, The Netherlands), as there was no commercially available software at the time of the study. As a result, there were five data sets (two standard-of-care and three RDCT reconstructions) for each included patient. Images were anonymized and coded for reading. Radiation Dose Assessment
For each patient, the volumetric CT dose index (CTDIvol) and the dose length product (DLP) were recorded. Effective dose was estimated by multiplying the DLP with the chestspecific conversion coefficient (0.014 mSv/mGy·cm) (16). Sizespecific dose estimation was also calculated by multiplying CTDIvol with a conversion factor f depending on the effective diameter (17).
REDUCING RADIATION DOSE AT CHEST CT
the lesion assessment. Those data sets were subsequently withdrawn from statistical analysis. Standard-of-care chest CT was assessed jointly by the two readers (consensus reading). The three reconstructions of RDCT (FBP, iDose4, and IMR) were independently assessed by the two readers. Sessions were separated by a minimum of 2 weeks to minimize recall bias. In case of disagreement, a consensus reading between these two readers was made afterwards during a new session. Images were displayed with a lung window setting (level = −600 HU, width = 1600 HU) and a soft window setting (level = 40 HU, width = 400 HU), but the radiologists were allowed to modify the window settings. Five parameters were assessed with a 4or 5-point scale: subjective noise, normal anatomic structure detectability (subpleural vessels, main fissures), artifacts (helical artifacts, streak artifacts, beam-hardening artifacts, and blotchy pixilated appearance), lesion conspicuity, and diagnostic confidence. In case of multiple parenchymal abnormalities, grading for lesion conspicuity was attributed considering the less visible lesion. Diagnostic confidence was rated considering the evolution assessment. Grading scales are detailed in Table 2. For lesion detection study, standard-of-care CT was considered as the reference. Observers were asked to record any lesion in each pulmonary lobe, the left upper lobe being divided into culmen (upper part) and lingula (lower part). Lung lesions were defined according to the glossary of terms for thoracic imaging from the Fleischner Society (21). The exact number of micronodules (<3 mm) was recorded if fewer or equal to 5, conversely to cases with more than 5 micronodules, for which the mention >5 was noted. Finally, evolution assessment (lesion stability, progression, or regression) was evaluated compared to the initial evaluation.
Objective Measurements
The mean Hounsfield unit (HU) and its standard deviation representing objective noise were measured for all the CT image series by one radiologist on a clinical workstation (IntelliSpace Portal, Philips Healthcare). Circular regions of interest were drawn inside the trachea immediately over the carina and in the descending thoracic aorta at the carina level. The ROIs were duplicated in the same location for each image reconstruction. For standard-of-care acquisition, aorta noise was measured using soft tissue kernel and trachea noise using lung parenchyma kernel. Measurements were repeated three times in three adjacent slices and the mean value was considered. Subjective Measurements
The image quality criteria chosen in the study have previously been totally or partially used in the literature (18–20). Subjective image quality of all data sets was independently assessed by two radiologists (C. de M.-M., C. de B) with 6 and 15 years of experience in general radiology. They were first trained together on two image data sets to get themselves familiar with the grading of subjective measurements and with
Statistical Analysis
The results were expressed as median [interquartile ranges] for continuous and semiquantitative variables, and described as numbers and percentages for qualitative and ordinal variables. For subjective quality criteria and lesion detection, analysis was performed with the consensus ratings. Objective noise was compared between each reconstruction using pairwise Wilcoxon test, and subjective image quality scores were compared using chi-square trend tests. For lesion detection, each pulmonary lobe was considered as an independent observation; 216 lobes (36 patients × 6 lobes) were then analyzed. Sensitivities and specificities for each RDCT were assessed using standard-of-care CT as the gold standard. Comparisons between sensitivities were performed using adjusted Wald confidence intervals (CIs) for a difference of proportions with matched pairs (22). Interobserver agreement between the two radiologists for the subjective quality rating and the lesion detection was calculated using percentage of agreement. Statistical analysis was performed using R software (R 2.14.0, R Foundation for Statistical Computing, Vienna, Austria). P < 0.05 (two-sided) was the criterion for statistical significance in all analyses. 3
Academic Radiology, Vol ■, No ■, ■■ 2016
DE MARGERIE-MELLON ET AL
TABLE 2. Grading for Subjective Image Quality 1 Subjective noise Normal anatomic structure detectability* Artifacts†
Lesion conspicuity Diagnostic confidence
2
3
4
5
Above average noise Unacceptable visualization of small structures Artifacts affecting diagnostic information Probable artifact mimicking a lesion Poor confidence
Unacceptable image noise Not applicable
Minimal image noise Excellent visualization
Less than average noise
Average image noise
Above average visibility
Acceptable visibility
No artifacts
Minor artifacts not interfering with diagnosis decision making Well-seen lesion with poorly demarcated margins
Major artifacts but possible diagnosis
Probably confident
Confident only for limited clinical situations
Well-seen lesion with sharp margins Completely confident
Subtle lesion
Not applicable
Definite artifact mimicking a lesion Not applicable
* Subpleural vessels and main fissures. † Helical artifacts, streak artifacts, beam-hardening artifacts, and blotchy pixilated appearance.
Figure 1. Boxplots representing the trachea and aorta noise of the standardof-care protocol and of the three reconstructions of reduced-dose chest computed tomography. Boxes indicate interquartile range; whiskers extend the data extremes, and horizontal lines the median value. All pairwise comparisons were statistically significant.
RESULTS Technical Parameters
FBP and iDose4 reconstructions took a few seconds. IMR reconstruction time was 126 seconds for 777 images. The median [interquartile range] for CTDIvol, DLP, sizespecific dose estimation, and estimated effective radiation dose were 3.4 mGy [3.1–3.9], 129.0 mGy·cm [112.8–144.0], 4.5 mGy [4.2–5.2], and 1.8 mSv [1.6–2.0], respectively, for the standard-of-care protocol, and 0.43 mGy for all patients, 16.0 mGy·cm [15.5–16.9], 0.59 mGy [0.55–0.62], and 0.22 mSv [0.22–0.24], respectively, for RDCT (P < 0.0001). 4
Objective Measurements
Noise measurements are presented in Figure 1. The aorta and trachea noise were, respectively, 13.2 [12.0–15.4] HU and 48.8 [45.2–53.0] HU for the standard-of-care protocol, 61.5 [51.6–83.6] HU and 42.8[38.3–50.8] HU with FBP, 26.7 [23.7–33.0] HU and 28.0 [21.3–32.3] HU with iDose4, and 14.0 [11.8–16.7] HU and 13.3 [9.3–16.3] HU with IMR for RDCT. Measurements were significantly different between all pairs with RDCT (P < 0.0001 for all pairwise comparison), showing a 77% median decrease in aorta noise and a 70% median decrease in trachea noise between FBP and IMR. Concerning standard-of-care CT and IMR reconstruction of
Academic Radiology, Vol ■, No ■, ■■ 2016
REDUCING RADIATION DOSE AT CHEST CT
RDCT, aorta noise was significantly lower (P = 0.03) and trachea noise was significantly higher (P < 0.0001) with standard-of-care CT. Subjective Measurements
Subjective scores are detailed in Table 3. For RDCT, normal anatomic structure detectability score was identical with the three reconstructions. Other subjective criteria (subjective noise, artifacts, lesion conspicuity, and diagnostic confidence) were significantly different between all pairs, lower with iDose4 (corresponding to higher performances) compared to FBP, and lower with IMR compared to iDose4 (P < 0.0001 for all pairwise comparison). Concerning standard-of-care CT and IMR reconstruction of RDCT, all measurements were also significantly different: standard-of-care CT scores were lower than IMR scores, except for subjective noise, which was lower with IMR (P < 0.001). Examples are provided in Figures 2–4. For all RDCT reconstructions, the percentages of agreement between the two readers range from 69% to 94% for artifacts and subjective noise, and from 50% to 94% for diagnostic confidence, and from 31% to 56% for lesion conspicuity. The percentage of agreement for normal structure detectability was 100% for all RDCT reconstructions. Sixty-one countable nodules were identified on standardof-care CT. FBP and iDose4 reconstructions of RDCT identified, respectively, 53 and 55 countable nodules without
any false-positive. IMR reconstruction identified 56 countable nodules, including one false-positive. Thus, detectability rates (the number of true-positive nodules divided by the number of nodules detected on standard-of-care CT) were 87% (95% CI: 76–94) for FBP and 90% (95% CI: 80–96) for iDose4 and IMR. These differences were not statistically significant. The percentage of agreement for the nodule number per lobe was higher than 90% in all reconstructions. Standard-of-care examination identified 83 countable micronodules among all patients. Forty micronodules were identified with FBP reconstruction of RDCT, including one false-positive, 57 with iDose4 without any false-positive, and 65 with IMR including one false-positive. Countable micronodule detectability rates were 47% (95% CI: 36–58), 69% (95% CI: 57–78), and 77% (95% CI: 66–86), respectively. Countable micronodules detection was significantly higher with IMR and iDose4 than with FBP. The percentage of agreement for the micronodule number per lobe was higher than 80% in all reconstructions. Table 4 summarizes the performances in lesion detection for the three reconstructions of RDCT compared to standardof-care CT (reference) considering 216 lobes. For groundglass opacities, detection sensitivities of IMR and iDose4 were significantly higher than FBP. For emphysema detection, sensitivity of IMR was significantly higher than FBP and iDose4. Other differences in sensitivities among the three reconstructions were found to be nonsignificant. The percentage of
TABLE 3. Subjective Quality Scores (Consensus Reading) Ultra Reduced-dose CT
Subjective noise (1–5) 2 3 4 5 Normal anatomic structure detectability (1–4) 2 3 4 Artifacts (1–4) 1 2 3 4 Lesion conspicuity (1–5) 1 2 3 Diagnostic confidence (1–4) 1 2 3
Standard-of-care CT
FBP
iDose4
IMR
7 (19) 29 (81) 0 (0) 0 (0)
0 (0) 0 (0) 18 (50) 18 (50)
0 (0) 10 (28) 26 (72) 0 (0)
32 (89) 4 (11) 0 (0) 0 (0)
10 (28) 26 (72) 0 (0)
0 (0) 0 (0) 36 (100)
0 (0) 0 (0) 36 (100)
0 (0) 0 (0) 36 (100)
21 (58) 15 (42) 0 (0) 0 (0)
0 (0) 2 (6) 32 (88) 2 (6)
0 (0) 25 (69) 11 (31) 0 (0)
0 (0) 36 (100) 0 (0) 0 (0)
35 (97) 1 (3) 0 (0)
0 (0) 14 (39) 22 (61)
0 (0) 31 (86) 5 (14)
25 (69) 11 (31) 0 (0)
36 (100) 0 (0) 0 (0)
0 (0) 20 (56) 16 (44)
0 (0) 35 (97) 1 (3)
15 (42) 21 (58) 0 (0)
CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction. Data are expressed as count (percentages).
5
DE MARGERIE-MELLON ET AL
Figure 2. Example of normal anatomic lung structure detectability scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show pulmonary fissures, which are well seen as a complete thin line in (a), but which are ill-defined in (b), (c), and (d). Detectability of normal anatomic lung structures was ranked 3 in (a), and ranked 4 in (b), (c), and (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
agreement between the two readers was higher than 94% for all lesions in the three reconstructions, except for groundglass opacities (87%–93%). There were 12 pleural effusions on standard-of-care CT, which were all detected with iDose4 and IMR RDCT reconstructions. One small pleural effusion was missed with FBP. Over the four pericardial effusions diagnosed on standardof-care CT, two were detected on FBP reconstruction and three were detected on iDose4 and IMR. Evolution assessment (stability, progression, and regression) was made on the standard-of-care CT by comparison with the previous examination. FBP reconstruction of RDCT provided the correct evolution assessment in 34 cases over 36. With IMR and iDose4 reconstructions, the evolution assessment was correct in all cases. DISCUSSION Lowering radiation dose induced by CT scan examinations without compromising image quality is a critical issue today, in order to 6
Academic Radiology, Vol ■, No ■, ■■ 2016
respect the “as low as reasonably achievable” (ALARA) principle. The current prospective study suggests that chest RDCT (CTDI = 0.43 mGy, effective dose of 0.22 mSv) reconstructed with IR allows for an accurate assessment of the evolution of lung affections in a selected population. As expected, objective noise was lower in iDose4 compared to FBP and higher in iDose4 compared to IMR, as well as the artifacts and the subjective noise ratings. Because of the low dose level, artifacts were mainly due to beam hardening. IMR is the latest algorithm that aims to improve noise and artifact suppression by modeling not only noise statistics but also system geometry. Noise and artifacts are important factors that alter image quality, and in parallel with their decrease we observed an improvement of diagnostic confidence when using iDose4 and IMR. Diagnostic confidence was greater with IMR than with iDose4 (“completely confident” in 42% and 0% of cases, respectively); however, readers were mostly “completely confident” or “probably confident” with both reconstructions (100% or 97%, respectively, for IMR and iDose4), which can be considered as clinically acceptable. On the opposite, diagnostic confidence of FBP reconstruction (“confident only for limited clinical situations” in 44% of the cases) was insufficient. However, the loss of small anatomic elements like lung fissures was observed in all reconstruction methods. IR algorithms did not allow subjective improvement of spatial resolution at this very low dose level. Concerning lesion detection, no significant difference was observed among the three reconstruction methods for nodule detectability, which was close to 90% in all cases. Nodules are high contrast lesions, and their sizes (>3 mm) made them visible even with high background noise. Conversely, for countable micronodule detectability, IMR and iDose4 showed better performances than FBP but remained below 80%. Likewise, ground-glass opacity detectability was close to 100% with iDose4 and IMR and much higher than with FBP. It is likely that high background noise in FBP significantly hampered micronodules and ground-glass opacity identification. Finally, emphysema detectability was performing well with IMR but insufficient with iDose4 and FBP. This appears mainly related to the reduction of streak artifacts at the level of the shoulders, which may preclude visualization of emphysema. Although emphysema is usually not relevant data in case of immunocompromised lung disease, other low density abnormalities (mosaic perfusion, thin-walled cysts) can be encountered in such pathologies. At the clinical level, accurate evolution assessment was obtained in all cases with both iDose4 and IMR. On the contrary, RDCT images obtained with FBP led to misdiagnosis in two cases. The estimated effective dose delivered by RDCT protocol (0.22 mSv) in our study was much lower than the standard low-dose CT scans (0.8–1.5 mSv) (23), coming close to the level of a posteroanterior and a lateral chest x-ray (0.05– 0.24 mSv) (1). Moreover, IMR reconstruction time was low (126 seconds for 777 images), close to those reported in the literature (from 80 seconds for CT colonography [24] to 217 seconds for a chest CT with 777 images [4]).
Academic Radiology, Vol ■, No ■, ■■ 2016
REDUCING RADIATION DOSE AT CHEST CT
Figure 3. Example 1 of lesion conspicuity scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show a ground-glass opacity in the left lung. Definition of this opacity grows from (b) to (d). Lesion conspicuity was ranked 1 in (a), 3 in (b), 2 in (c), and 1 in (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Figure 4. Example 2 of lesion conspicuity scoring. Corresponding images for (a) standard-of-care CT, (b) RDCT, FBP reconstruction, (c) RDCT, iDose4 reconstruction, and (d) RDCT, IMR reconstruction. Arrows show emphysema of the right apex. Definition of emphysema grows from (b) to (d). Lesion conspicuity was ranked 1 in (a), 3 in (b), 2 in (c), and 1 in (d). CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Similar dose reductions were achieved in several studies. Some of them (25–27) used MBIR algorithm (model-based iterative reconstruction), a fully iterative algorithm from GE Healthcare (Waukesha, WI, USA). Like IMR, it incorporates a system geometry modeling in addition to statistics model. However, MBIR requires high computational power, resulting in long processing time of up to 1 hour (25,28), which may limit routine applicability. Concerning lesion detection, several publications (26,27) only focused on nodule detection in the context of lung cancer screening. They showed the great potential of MBIR with RDCT images (effective dose: 0.17–0.20 mSv). Neroladaki et al. (25) confirmed MBIR high diagnostic performances for nodules, but demonstrated its limitations at this dose level (mean effective dose: 0.16 mSv) for the detection of ground-glass opacities and emphysema, contrary to the results obtained in the current study with
IMR. Some studies (29,30) with a similar dose level used SAFIRE algorithm (sinogram-affirmed iterative reconstruction, Siemens Healthcare, Forchheim, Germany). SAFIRE is an IR method that processes both in raw data and in image domains. Lee et al. (30) demonstrated that reduced-dose (mean effective dose: 0.29 mSv) chest CT with SAFIRE generates sufficient image diagnostic quality in 90% of the cases. Like with MBIR, performances were also lower in case of lesions of decreased attenuation, like emphysema and ground-glass opacities. Finally, Nagatani et al. (31) used AIDR 3D hybrid algorithm (adaptative iterative dose reduction using threedimensional processing (Toshiba Medical Systems Corporation, Otawara, Japan) with reduced dose (0.29 mSv) to assess lung nodule detection. They showed comparable lung nodule detection to reference CT, except for smaller ground-glass nodules. 7
Academic Radiology, Vol ■, No ■, ■■ 2016
8
0.95 (0.87–0.99) 0.96 (0.91–0.98) 0.9 (0.77–0.97) 0.99 (0.96–1) 1 (0.89–1) 1 (0.98–1) 0.86 (0.64–0.97) 1 (0.98–1) 0.6 (0.32–0.84) 1 (0.98–1) 0.57 (0.18–0.9) 1 (0.98–1) 77 45 32 18 9 4 0.96 (0.89–0.99) 0.98 (0.94–1) 0.67 (0.52–0.8) 0.99 (0.97–1) 0.94 (0.79–0.99) 1 (0.98–1) 1 (0.84–1) 1 (0.98–1) 0.73 (0.45–0.92) 1 (0.98–1) 0.57 (0.18–0.9) 1 (0.98–1) 75 33 30 21 11 4 0.67 (0.55–0.77) 0.98 (0.94–0.97) 0.65 (0.50–0.78) 0.99 (0.96–1) 0.94 (0.79–0.99) 1 (0.98–1) 0.95 (0.76–1) 0.99 (0.96–1) 0.53 (0.27–0.79) 1 (0.98–1) 0.43 (0.01–0.82) 1 (0.98–1) 53 33 30 22 8 3 75 48 32 21 15 7
CI, confidence interval; CT, computed tomography; FBP, filtered back projection; IMR, iterative model reconstruction; RDCT, reduced-dose chest computed tomography.
Lesion
Ground-glass opacity Emphysema Consolidation Linear atelectasis Bronchiectasis Septal thickening
Sensitivity (95% CI) Sensitivity (95% CI) Sensitivity (95% CI) Number of Lesion with Number of Standard-of- Lesion with care CT RDCT
TABLE 4. Lesion Detection
FBP
Specificity (95% CI)
Number of Lesion with RDCT
iDose4
Specificity (95% CI)
Number of Lesion with RDCT
IMR
Specificity (95% CI)
DE MARGERIE-MELLON ET AL
To our knowledge, few publications (6,7) studied IMR clinical application for the chest. Both studies demonstrated great potential for IMR in image quality. Khawaja et al. (7) also assessed lesion detection in a group of 10 patients. All lesions could be detected with IMR. Dose level was higher for the low-dose protocol (0.9 mSv), but the clinical context was different. Indeed, to give clinical relevance to our study, we focused on a specific group of patients who underwent a fulldose chest CT examination in the previous 2 months. The main question of the control CT scan was only to assess evolution of a known lung disease. We showed that RDCT reconstructed with iDose4 and IMR was able to answer this question. Further studies are needed to evaluate IMR performances at this reduced dose level for initial diagnosis, owing to the fact that subtle details could be suppressed. This study has some limitations. First, the iDose4 levels for standard-of-care CT and RDCT were subjectively selected, as our preferred compromise for noise reduction and usual image texture preservation for the selected dose level and according to the settings used in our department, taking into account that no consensus has been reported in the literature regarding the optimal level of this hybrid IR (7,15,32,33). Second, blinding to the reconstruction mode was not applicable. Indeed, the visual aspects of FBP, iDose4, and IMR images were easily recognizable. It may have induced some bias by favoring the most recent technology. Third, quality image scoring and lesion detection are intrinsically subjective and difficult to transpose to other studies, which is why we associated objective measurements. Fourth, the sample size was small because of the strict inclusion criteria and the limited duration of the study. Therefore, most of the comparisons can suffer from a lack of power. Those inclusion criteria were chosen to obtain a homogeneous group of patients in order to give immediate clinical relevance to our results. Finally, as expected in a population of patients treated for hematologic malignancy, the median body mass index (BMI) of the cohort was low (22 kg/m2), with only four patients with a BMI > 25 kg/m2. In the study by Lee et al. (30) using RDCT with similar dose (mean effective dose: 0.29 mSv) reconstructed with SAFIRE, images of diagnostic quality were obtained in more than 95% of cases for patients with a BMI of less than 25 kg/m2, whereas this rate decreased to 70% in patients with BMI > 25 kg/m2. It is probable that our RDCT protocol parameters should be adapted for patients with higher BMI. Nevertheless, these parameters could also benefit a pediatric population. To conclude, our study shows that chest CT may be achievable with a dose close to 1 posteroanterior and lateral chest x-ray using the last generation of IR for lung affection monitoring in a specific population usually requiring repeated followup studies. Similar evolution assessment can be obtained for all patients either with iDose4 and IMR. Nevertheless, IMR appears to improve diagnostic confidence and lesion conspicuity, as well as detection of reduced-contrast lesions like emphysema. Contrary to other recent IR algorithms, reconstruction time is compatible with clinical exercise. However, it has to
Academic Radiology, Vol ■, No ■, ■■ 2016
be demonstrated that relevant information is not missing with such dose and reconstructions for initial diagnosis of lung disorders. ACKNOWLEDGMENTS This work was supported by grants from Société Française de Radiologie and from Fonds d’Etude et de Recherche du Corps Médical. REFERENCES 1. Mettler FA, Jr, Bhargavan M, Faulkner K, et al. Radiologic and nuclear medicine studies in the United States and worldwide: frequency, radiation dose, and comparison with other radiation sources–1950-2007. Radiology 2009; 253:520–531. doi:10.1148/radiol.2532082010. 2. Brenner DJ, Hall EJ. Computed tomography—an increasing source of radiation exposure. N Engl J Med 2007; 357:2277–2284. doi:10.1056/ NEJMra072149. 3. Brooks RA, Di Chiro G. Theory of image reconstruction in computed tomography. Radiology 1975; 117:561–572. doi:10.1148/117.3.561. 4. Mehta D, Thompson R, Morton T, et al. Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality. Med Phys Int J 2013; 1:147–155. 5. Kim H, Park CM, Song YS, et al. Influence of radiation dose and iterative reconstruction algorithms for measurement accuracy and reproducibility of pulmonary nodule volumetry: a phantom study. Eur J Radiol 2014; 83:848–857. doi:10.1016/j.ejrad.2014.01.025. 6. Yuki H, Oda S, Utsunomiya D, et al. Clinical impact of model-based type iterative reconstruction with fast reconstruction time on image quality of low-dose screening chest CT. Acta Radiol 2015; 57:295–302. doi:10.1177/ 0284185115575537. 7. Khawaja RDA, Singh S, Gilman M, et al. Computed tomography (CT) of the chest at less than 1 mSv: an ongoing prospective clinical trial of chest CT at submillisievert radiation doses with iterative model image reconstruction and iDose4 technique. J Comput Assist Tomogr 2014; 38:613– 619. doi:10.1097/RCT.0000000000000087. 8. Koh TT, Colby TV, Müller NL. Myeloid leukemias and lung involvement. Semin Respir Crit Care Med 2005; 26:514–519. doi:10.1055/s-2005922034. 9. Georgiadou SP, Sipsas NV, Marom EM, et al. The diagnostic value of halo and reversed halo signs for invasive mold infections in compromised hosts. Clin Infect Dis 2011; 52:1144–1155. doi:10.1093/cid/cir122. 10. Bissinger AL, Einsele H, Hamprecht K, et al. Infectious pulmonary complications after stem cell transplantation or chemotherapy: diagnostic yield of bronchoalveolar lavage. Diagn Microbiol Infect Dis 2005; 52:275– 280. doi:10.1016/j.diagmicrobio.2005.03.005. 11. Vahid B, Marik PE. Infiltrative lung diseases: complications of novel antineoplastic agents in patients with hematological malignancies. Can Respir J 2008; 15:211. 12. Nakaseko C, Ozawa S, Sakaida E, et al. Incidence, risk factors and outcomes of bronchiolitis obliterans after allogeneic stem cell transplantation. Int J Hematol 2011; 93:375–382. doi:10.1007/s12185-011-0809-8. 13. Tack D, Kalra MK, Genevois PA. ALARA concept for MDCT optimization: what is reasonable, what is achievable? In: Radiation dose multidetector CT. Berlin Heidelberg: Springer, 2012; 231–258. Editors: Tack D, Kalra MK, Gevenois PA. 14. Wang Y, de Bock GH, van Klaveren RJ, et al. Volumetric measurement of pulmonary nodules at low-dose chest CT: effect of reconstruction setting on measurement variability. Eur Radiol 2010; 20:1180–1187. doi:10.1007/s00330-009-1634-9. 15. Nakajo C, Heinzer S, Montandon S, et al. Chest CT at a dose below 0.3 mSv: impact of iterative reconstruction on image quality and lung analysis. Acta Radiol 2015; 57:311–317. doi:10.1177/0284185115578469.
REDUCING RADIATION DOSE AT CHEST CT
16. Christner JA, Kofler JM, McCollough CH. Estimating effective dose for CT using dose-length product compared with using organ doses: consequences of adopting International Commission on Radiological Protection publication 103 or dual-energy scanning. AJR Am J Roentgenol 2010; 194:881–889. doi:10.2214/AJR.09.3462. 17. American Association of Physicist in Medicine. Size-specific dose estimates (SSDE) in pediatric and adult body CT examinations. Task Group 204, 2011. 18. Singh S, Kalra MK, Gilman MD, et al. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study. Radiology 2011; 259:565–573. doi:10.1148/radiol.11101450. 19. Singh S, Kalra MK, Moore MA, et al. Dose reduction and compliance with pediatric CT protocols adapted to patient size, clinical indication, and number of prior studies. Radiology 2009; 252:200–208. doi:10.1148/radiol.2521081554. 20. Yamada Y, Jinzaki M, Hosokawa T, et al. Dose reduction in chest CT: comparison of the adaptive iterative dose reduction 3D, adaptive iterative dose reduction, and filtered back projection reconstruction techniques. Eur J Radiol 2012; 81:4185–4195. doi:10.1016/j.ejrad .2012.07.013. 21. Hansell DM, Bankier AA, MacMahon H, et al. Fleischner Society: glossary of terms for thoracic imaging. Radiology 2008; 246:697–722. doi:10.1148/radiol.2462070712. 22. Agresti A, Min Y. Simple improved confidence intervals for comparing matched proportions. Stat Med 2005; 24:729–740. doi:10.1002/sim.1781. 23. Larke FJ, Kruger RL, Cagnon CH, et al. Estimated radiation dose associated with low-dose chest CT of average-size participants in the national lung screening trial. Am J Roentgenol 2011; 197:1165–1169. doi:10.2214/AJR.11.6533. 24. Lambert L, Ourednicek P, Briza J, et al. Sub-milliSievert ultralow-dose CT colonography with iterative model reconstruction technique. PeerJ 2016; 4:e1883. http://dx.doi.org/10.7717/peerj.1883. 25. Neroladaki A, Botsikas D, Boudabbous S, et al. Computed tomography of the chest with model-based iterative reconstruction using a radiation exposure similar to chest x-ray examination: preliminary observations. Eur Radiol 2012; 23:360–366. doi:10.1007/s00330-012-2627-7. 26. Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for ultralow-dose chest CT. Invest Radiol 2013; 48:206– 212. doi:10.1097/RLI.0b013e31827efc3a. 27. Yamada Y, Jinzaki M, Tanami Y, et al. Model-based iterative reconstruction technique for ultralow-dose computed tomography of the lung: a pilot study. Invest Radiol 2012; 47:482–489. doi:10.1097/ RLI.0b013e3182562a89. 28. Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. Eur Radiol 2012; 22:1613–1623. doi:10.1007/s00330-012-2452-z. 29. Kim Y, Kim YK, Lee BE, et al. Ultra-low-dose CT of the thorax using iterative reconstruction: evaluation of image quality and radiation dose reduction. AJR Am J Roentgenol 2015; 204:1197–1202. doi:10.2214/ AJR.14.13629. 30. Lee SW, Kim Y, Shim SS, et al. Image quality assessment of ultra lowdose chest CT using sinogram-affirmed iterative reconstruction. Eur Radiol 2014; 24:817–826. doi:10.1007/s00330-013-3090-9. 31. Nagatani Y, Takahashi M, Murata K, et al. Lung nodule detection performance in five observers on computed tomography (CT) with adaptive iterative dose reduction using three-dimensional processing (AIDR 3D) in a Japanese multicenter study: comparison between ultra-low-dose CT and low-dose CT by receiver-operating characteristic analysis. Eur J Radiol 2015; 84:1401–1412. doi:10.1016/j.ejrad.2015.03.012. 32. Laqmani A, Buhk JH, Henes FO, et al. Impact of a 4th generation iterative reconstruction technique on image quality in low-dose computed tomography of the chest in immunocompromised patients. Rofo 2013; 185:749–757. doi:10.1055/s-0033-1335577. 33. Willemink MJ, Borstlap J, Takx RAP, et al. The effects of computed tomography with iterative reconstruction on solid pulmonary nodule volume quantification. PLoS ONE 2013; 8:e58053. doi:10.1371/journal. pone.0058053.
9