CT pulmonary angiography: dose reduction via a next generation iterative reconstruction algorithm

Abstract

Background

Computed tomography pulmonary angiography (CTPA) is the standard imaging modality for detection or rule out of pulmonary embolism (PE); however, radiation exposure is a serious concern. With iterative reconstruction algorithms a distinct dose reduction could be achievable.

Purpose

To evaluate a next generation iterative reconstruction algorithm for detection or rule-out of PE in simulated low-dose CTPA.

Material and Methods

Low-dose CT datasets with 50%, 25%, and 12.5% of the original tube current were simulated based on CTPA examinations of 92 patients with suspected PE. All datasets were reconstructed with two reconstruction algorithms: standard filtered back-projection (FBP) and iterative model reconstruction (IMR). In total, 736 CTPA datasets were evaluated by three blinded radiologists regarding image quality, diagnostic confidence, and detectability of PE. Furthermore, contrast-to-noise ratio (CNR) was calculated.

Results

Images reconstructed with IMR showed better detectability of PE than images reconstructed with FBP, especially at lower dose levels. With IMR, sensitivity was over 95% for central and segmental PE down to a dose level of 25%. Significantly higher subjective image quality was shown at lower dose levels (25% and 12.5%) for IMR images whereas it was higher for FBP images at higher dose levels. FBP was rated as showing less artificial image appearance. CNR was significantly higher with IMR at all dose levels.

Conclusion

By using IMR, a dose reduction of up to 50% while maintaining satisfactory image quality seems feasible in standard clinical situations, resulting in a mean effective dose of 1.38 mSv for CTPA.

Keywords

Thorax computed tomography CT CT angiography lung adults computer applications – detection diagnosis

Introduction

Pulmonary embolism (PE) is a common disease, especially in immobilized and hospitalized patients, causing approximately 37,000 deaths in the European Union and 60,000–100,000 deaths in the USA annually (1). The overall risk for PE in hospitalized patients is around 1% with a rate of fatal PE of 2% for these patients (2). About 15% of all sudden deaths are attributed to PE (1).

If there is a clinical likelihood for PE (i.e. suggestive Wells score), computed tomography pulmonary angiography (CTPA) is the gold standard to confirm or rule out suspected pulmonary embolism (PE) (3). Due to improvements in CT technology, high diagnostic accuracy with sensitivities and specificities of >90% can be achieved (4 –6). As a drawback, CTPA involves a high radiation dose with a reported average effective dose (ED) of up to 10 mSv (7,8).

Radiation exposure is of concern because of the potentially increased lifetime risk for malignancy and gametal damage, especially when applied to young or pregnant patients (9). For example, radiation is an accepted cause of breast cancer and there is an especially high radiation exposure to the breast in CT scans of the thorax (10). An estimated 1.5–2.0% of all future cancer diseases can be attributed to radiation applied by CT studies with a higher risk for younger patients (9,11). Therefore, dose reduction with suitable image quality is needed, according to the principle of “as low as reasonably achievable” (ALARA).

There are multiple techniques for dose reduction that have been implemented in clinical routine. Reduction of tube voltage (kVp), tube current (mA), or scan time, as well as the use of new methods such as z-axis dose modulation, have led to a reduced radiation exposure (12,13). Suitable image quality of CTPA can be achieved with reduced tube voltages of 100 kVp or 80 kVp, resulting in a significant dose reduction (14,15). But reduction of tube voltage leads to higher image noise, especially in obese patients. To reduce image noise, higher tube currents are needed, leading to only limited or even missing reduction of total radiation exposure (16,17).

Progress in computing power and image reconstruction algorithms have enabled widespread use of iterative reconstruction methods. These offer advanced image quality and/or reduction of radiation exposure, even in obese patients (16,18,19). First generation iterative algorithms have already led to a significant reduction of image noise (20). In a previous study, low-dose images with a reduced radiation dose by 75% provided full diagnostic confidence and sensitivity regarding detection of PE when the next generation iterative reconstruction algorithm “iterative model reconstruction” (IMR) was used (21). This study, however, was performed only with a small number of patients (16). Thus, before reduction in tube current can be enabled for all patients, results must be confirmed in a follow-up study with a greater number of patients to exclude the possibility of missed PE at lower dose levels.

In the present study, IMR was compared to the formerly standard algorithm FBP at different simulated tube current levels in images obtained from clinical scans of 92 patients. The aim was to examine the performance of images reconstructed with IMR with respect to image quality, diagnostic confidence regarding PE, and diagnostic performance at lower dose levels. Establishment of this reconstruction algorithm in clinical routine could lead to a significant reduction in radiation dose.

Material and Methods

Patient population

Written informed consent was waived by the institutional review board, as all patients were included retrospectively. All scans were performed exclusively for clinical use with clinical standard protocols.

Ninety-two consecutive patients (48 men, 44 women) with suspected PE were included into this study. Mean age was 63.1 years (age range = 30–89 years) for male patients and 62.4 years (age range = 16–93 years) for female patients. Thirty-two patients (34.8%) were diagnosed positive for PE and 60 patients were diagnosed negative for PE. PE was classified as central, segmental, or subsegmental; PE in multiple locations was also possible. No preselection regarding patient weight, age, sex, or other characteristics was performed. Examinations primarily not suitable for clinical evaluation (i.e. due to contrast phase or motion artefacts) were not included.

CTPA image acquisition

All patients were examined using a state-of-the-art CT system (Brilliance iCT; Philips Healthcare, Cleveland, OH, USA), with a standard CTPA protocol. Scans were performed with 60 mL of intravenous contrast agent (Imeron 400, Bracco Imaging Deutschland GmbH, Konstanz, Germany) followed by a 50-mL saline chaser with an injection rate of 3.5 mL/s. The scan range included the whole thorax and was defined by an anteroposterior scout. The bolus tracker was placed within a region of interest (ROI) in the pulmonary trunk (threshold for scan start = 100 HU). The scan was performed craniocaudally with a pitch of 0.9 and a 128 × 0.625 mm detector configuration. Tube voltage depended on body mass index (BMI) with 120 kVp for a BMI > 25 kg/m² and 100 kVp for a BMI < 25 kg/m².

Tomographic slices were obtained with a field of view of 350–500 mm, based on the diameter of the patient. A 512 × 512 image matrix with a slice thickness of 0.625 mm was used.

For every patient, the automatically generated dose protocol was extracted after the examination. Tube voltage (kVp), tube current (mA), volume-weighted CT dose index (CTDI_vol), and dose length product (DLP) were collected. By multiplication of DLP by the chest conversion factor (0.0145), the effective dose (ED) could be calculated (22).

Simulation of low-tube-current images and reconstruction

Simulation of lower dose images was performed via the addition of noise including effects from photon starvation and detector noise using a previously described formula (23). Using this method, it is unnecessary to perform repetitive scans of the same patient with different dose levels as images are identical apart from dose settings (24). CTPA raw data were used to simulate CTPA scans with tube currents of 50%, 25%, and 12.5% of the original current. All other parameters were identical. All obtained CTPA images (tube current levels of 100% down to 12.5%) were reconstructed with two different reconstruction algorithms, FBP (standard soft tissue kernel) and IMR (IMR3 with strong iterative reconstruction, standard soft tissue kernel), resulting in a total of 736 datasets (four tube current levels, two algorithms, 92 patients). All images were reconstructed in axial view with slice thickness of 3 mm.

Objective image quality

For objective evaluation of image quality, the contrast-to-noise ratio (CNR) for the pulmonary trunk as a central vessel (CcNR) and cCNR for a peripheral vessel such as a segmental artery (CpNR) was obtained. A circular ROI was placed in the pulmonary trunk (HU_VC), the paraspinal muscle (HU_M), and a segmental artery (HU_VP). Standard deviation (SD) of the HU for all ROIs was obtained. SD in the central and peripheral vessel was defined as noise, respectively. CcNR was calculated as [HU_VC-HU_M]/noise and CpNR was calculated as [HU_VP-HU_M]/noise.

Subjective image quality

Each dataset was independently evaluated by three blinded radiologists with five, four, and three years of clinical experience regarding following criteria:

image quality in four levels (1 = unacceptable for diagnostic purposes, 2 = suboptimal, 3 = good, 4 = excellent); in this context, subjective image noise, artefacts, and overall image impression were considered;

artificial image appearance (1 = strong, 2 = moderate, 3 = weak, 4 = none).

Diagnostic confidence regarding detection of PE

Diagnostic confidence regarding central and peripheral PE as well as overall diagnostic confidence was rated in four levels (1 = poor confidence/diagnosis not possible due to bad image quality, 2 = poor confidence/confident only for limited clinical situation, 3 = probably confident, 4 = completely confident). Hereby, diagnostic confidence regarding central PE corresponds to the same region as detection of central PE (see below). Diagnostic confidence regrading peripheral PE corresponds to segmental and subsegmental PE combined.

Diagnostic performance

Readers could rate cases as positive for PE (PE_p) or as negative for PE (PE_n) in the central, segmental, and subsegmental vessels of the lung. In addition, they could state that a diagnosis was impossible due to bad image quality. PE should only be rated as positive in the specific location when a clear embolus was visible. Two authors, each with more than ten years of experience in CT imaging, discussed each case regarding presence of PE in all locations. They defined cases as positive (PE_p) or negative (PE_n) for central, segmental, and subsegmental PE based on the concerted evaluation of the full dose images; this consensus was defined as ground truth.

Statistical analysis

Statistical analysis was performed by dedicated software packages (SPSS, IBM, USA; Excel 2016, Microsoft, USA; Prism 7, Version 7.0c). Continuous data are expressed as arithmetic mean ± SD. Data were tested for Gaussian distribution via D’Agostino–Pearson omnibus test. As Gaussian distribution was present, two-sided paired t-test was used for comparison of CcNR and CpNR values, respectively. Graphs for subjective image quality assessment are shown in box-whisker plots, including mean, median, 25/75% quartile, whiskers (1–99 percentile) as well as outliers. Statistical evaluation of subjective image criteria was performed using Wilcoxon signed-rank test. A P value ≤0.05 was considered to indicate statistical significance. Inter-reader agreement was evaluated by using Fleiss’ kappa (κ). Hereby, values <0 were regarded as indicating no agreement and 0–0.20 as slight, 0.21–0.40 as fair, 0.41–0.60 as moderate, 0.61–0.80 as substantial, and 0.81–1 as almost perfect agreement (25).

If a reader evaluated the case as negative for PE but there was a PE present, this was regarded as false negative (fn). If a reader stated that the diagnosis was not possible due to image quality but there was a PE present in this case, this was regarded as false negative due to images quality (fn_IQ). Sensitivity was calculated using the following equation:

Sensitivity = 100 % - \frac{\sum fn + \sum {fn}_{IQ}}{(\sum {PE}_{p}) * 3}

Note that all cases positive for PE (PE_p) were included ×3 because every reader rated all cases.

If a reader evaluated the case as positive for PE but there was no PE present, this was regarded as false positive (fp). If a reader stated that the diagnosis was not possible due to image quality and there was no PE present in this case, this was regarded as false positive due to images quality (fp_IQ).

Specificity = 100 % - \frac{\sum fp + \sum {fp}_{IQ}}{(\sum {PE}_{n}) * 3}

Note that all cases negative for PE (PE_n) were also included ×3 because every reader rated all cases.

Results

Overview of patient results

Of 92 patients, 26 (28%) were examined using a tube voltage of 120 kVp, 66 (72%) patients were examined using a tube voltage of 100 kVp. Tube current was automatically adjusted via z-axis modulation; the resulting mean tube current was 105 mAs (range = 49–368 mAs). DLP was 190 mGy*cm (range =72–695 mGy*cm) with a mean effective dose of 2.76 ± 1.72 mSv (range = 1.04–10.08 mSv). In patients examined with 120 kVp, the mean ED was 4.6 ± 2.27 mSv. In patients examined with 100 kVp, mean ED was 2.07 ± 0.68 mSv. Thirty-two patients were diagnosed as positive for PE: central PE was present in 13 cases, segmental PE was present in 22 cases, and subsegmental PE was present in 26 cases.

CNR

At all dose levels, CcNR and CpNR were significantly higher in images reconstructed with IMR compared to images reconstructed with FBP (P < 0.001). CcNR and CpNR were higher in images reconstructed with IMR at the 12.5% dose level compared to images reconstructed with FBP at the 100% dose level. All values for CcCR and CpNR are shown in Table 1.

Table 1.

Subjective and objective image quality.

Location	Dose level (%)	Reconstruction algorithm
Location	Dose level (%)	IMR	FBP
Diagnostic confidence
Overall	12.5*	3.00	2.33
	25	3.33	3.33
	50	3.67	3.33
	100*	3.67	4.00
Central	12.5*	3.33	2.67
	25*	3.67	3.33
	50	3.67	3.67
	100	3.67	4.00
Peripheral	12.5*	2.67	2.00
	25	3.00	3.00
	50^†	3.33	3.67
	100	3.67	3.67
Image quality
	12.5*	2.33	1.33
	25*	2.67	2.33
	50	3.00	3.00
	100*	3.17	3.33
Artificial image appearance
	12.5*	2.67	1.00
	25*	2.67	1.00
	50*	2.33	1.00
	100*	2.00	1.00
CNR
CcNR	12.5*	13.70	1.45
	25*	19.82	2.31
	50*	21.17	3.75
	100*	24.52	5.40
CpNR	12.5*	9.90	1.01
	25*	11.20	2.11
	50*	12.80	3.07
	100*	16.75	4.80

For diagnostic confidence, image quality, and artificial image appearance, median values for all dose levels are shown. For CNR, absolute values are shown for central (CnNR) and peripheral (CpNR) vessels.

P < 0.001.

^†P < 0.05.

IMR, iterative model reconstruction; FBP, filtered back projection; CNR, contrast-to-noise ratio; CcNR, CNR for a central vessel; CpNR, CNR for a peripheral vessel.

Subjective image quality

A comparison of images reconstructed with FBP and IMR at the different dose levels is shown in Fig. 1.

Fig. 1.

Comparison of CT images reconstructed with IMR and FBP at different dose levels. Selected emboli are marked with red arrows in the top figure panel (100% dose level IMR). Window level is set to 200 HU and window width is set to 400 HU for all images. (a) CT images of a patient with a central and several segmental/subsegmental PEs. Note the increasing noise and thus worse detectability even of the central PE in the FBP-reconstructed CT images at low dose levels. The small subsegmental PE in the right lower lobe is visible in the FBP and IMR images at 100% dose and in the IMR image at 50% dose. In all other images, this subsegmental PE is not clearly visible. (b) CT images of a patient with multiple segmental and subsegmental PE in both lungs. Small emboli are detectable in images reconstructed with IMR even at low dose levels, whereas detectability at low dose levels is impaired when images are reconstructed with FBP.

At the highest dose level, images reconstructed with FBP showed a higher median subjective image quality than images reconstructed with IMR (median of 3.33 vs. 3.17). At the 50% dose level, the median was 3.0 for both IMR and FBP. At lower dose levels, the median was higher for IMR (2.67 at the 25% dose level and 2.33 at the 12.5% dose level) than for FBP (2.33 at the 25% dose level and 1.33 at the 12.5% dose level) (Fig. 2). Differences in image quality were significant at the 100%, 25%, and 12.5% dose level (P < 0.001).

Fig. 2.

Image quality for CT images reconstructed with IMR and FBP at the different dose levels. At the highest dose level, image quality was rated higher for FBP images whereas at lower dose levels, IMR performed better. Differences are significant at the 100%, 25%, and 12.5% dose level (P < 0.001).

At all dose levels, readers rated artificial image appearance of images reconstructed with IMR higher than of images reconstructed with FBP (P < 0.001). For IMR, artificial image appearance was lower at higher dose levels (2.0 at the 100% dose level and 2.33 at the 50% dose level) and showed the highest median (2.67) at the lowest dose level. All medians are shown in Table 1.

Diagnostic confidence regarding detection of PE

Overall diagnostic confidence was higher for FBP at the 100% dose level (median of 4.0 for FBP vs. 3.67 for IMR; P < 0.001). At both the 50% and 25% dose levels, no difference between FBP and IMR was shown (3.67 at the 50% dose level and 3.33 at the 25% dose level). At the 12.5% dose level, a significant difference (P < 0.001) was shown with higher diagnostic confidence for images reconstructed with IMR (median 3.0 vs. 2.33).

Diagnostic confidence regarding central PE was higher for images reconstructed with FBP than for images reconstructed with IMR at the 100% dose level (4.0 vs. 3.67; P = 0.086) and did not differ at the 50% dose level (median of 3.67; P = 0.051). At the 25% and 12.5% dose levels, diagnostic confidence regarding central PE was significantly (P < 0.001) higher for images reconstructed with IMR (median of 3.67 and 3.33) than for images reconstructed with FBP (median of 3.33 and 2.67) as shown in Fig. 3.

Fig. 3.

Diagnostic confidence regarding detection or rule out of central PE for different dose levels for IMR and FBP reconstructed CT images. At higher dose levels (50% and 100%), there is no significant difference between images reconstructed with FBP and IMR. At lower dose levels (25% and 12.5%), images reconstructed with IMR show significantly (P < 0.001) higher diagnostic confidence than images reconstructed with FBP.

Diagnostic confidence regarding peripheral PE generally showed lower medians than overall diagnostic confidence or diagnostic confidence regarding central PE. Identical medians were shown for both algorithms at the 100% and 25% dose levels (3.67 and 3.0, respectively). At the 50% dose level, median diagnostic confidence was higher for FBP than for IMR (3.67 vs. 3.33, P = 0.46). At the 12.5% dose level, the median for IMR was significantly (P < 0.001) higher than for FBP (2.67 vs. 2.0).

Regarding all dose levels and subdivisions, median diagnostic confidence was 3.0 (probably confident) or higher for all subgroups when images were reconstructed with IMR (excluding peripheral PE at the 12.5% dose level). All medians are shown in Table 1.

Diagnostic performance

No central PE was missed by any reader in images reconstructed with IMR at dose levels of 100% and 50% (Table 2). Sensitivities regarding segmental and subsegmental PE were higher for images reconstructed with IMR than for images reconstructed with FBP at any dose level apart from segmental PE at the 100% dose level (Table 3). For images reconstructed with IMR, sensitivities regarding segmental PE were >95% at all dose levels down to the 25% dose level and >90% regarding subsegmental PE at the 100% and 50% dose levels.

Table 2.

Diagnostic performance regarding detection of pulmonary embolism.

Dose level (%)	Reconstruction algorithm	Central PE				Segmental PE				Subsegmental PE
Dose level (%)	Reconstruction algorithm	FN	TN	FP	TP	FN	TN	FP	TP	FN	TN	FP	TP
12.5	FBP	2	37	35	202	14	52	81	129	40	38	116	82
12.5	IMR	2	37	8	229	9	57	24	186	19	59	52	146
25	FBP	2	37	6	231	6	60	29	181	21	57	61	137
25	IMR	1	38	1	236	2	64	7	203	8	70	30	168
50	FBP	2	37	2	235	3	63	11	199	8	70	32	166
50	IMR	0	39	0	237	2	64	5	205	7	71	18	180
100	FBP	0	39	1	23 6	0	66	2	208	4	74	13	185
100	IMR	0	39	1	236	1	65	1	209	3	75	10	188

Absolute numbers regarding detection of central, segmental, and segmental pulmonary embolism (PE). Shown are false negative (FN), true negative (TN), false positive (FP), and true positive (TP) numbers of all readers combined. There were 13 cases positive for central PE, 22 cases positive for segmental PE, and 26 cases positive for subsegmental PE. Note that all cases were included ×3 because there were three readers.

IMR, iterative model reconstruction; FBP, filtered back projection.

Table 3.

Detection rates of pulmonary embolism.

Dose level (%)	Reconstruction algorithm	Central PE		Segmental PE		Subsegmental PE
Dose level (%)	Reconstruction algorithm	Sensitivity (%)	Specificity (%)	Sensitivity (%)	Specificity (%)	Sensitivity (%)	Specificity (%)
12.5	FBP	94.9	85.2	78.8	61.4	48.7	41.4
12.5	IMR	94.9	96.6	86.4	88.6	75.6	73.7
25	FBP	94.9	97.5	90.9	86.2	73.1	69.2
25	IMR	97.4	99.6	97.0	96.7	89.7	84.8
50	FBP	94.9	99.2	95.5	94.8	89.7	83.8
50	IMR	100	100	97.0	97.6	91.0	90.9
100	FBP	100	99.6	100	99.0	94.9	93.4
100	IMR	100	99.6	98.5	99.5	96.2	94.9

Sensitivity and specificity values regarding the detection of central/segmental/subsegmental pulmonary embolism (PE.) Both sensitivity and specificity are >90% in all locations for images reconstructed with IMR at the 100% and 50% dose levels. For central and segmental PE, the sensitivity and specificity are >96% for images reconstructed with IMR at the 25%, 50%, and 100% dose levels.

IMR, iterative model reconstruction; FBP, filtered back projection.

False positive diagnoses were made in <1% of all cases negative for central PE in images reconstructed with IMR at the 100%, 50%, and 25% dose levels. For images reconstructed with FBP, specificities for central PE were lower. Inter-reader agreement was ≥0.5 for all subgroups. Inter-reader agreement was higher for central PE and for higher dose levels with moderately higher values for IMR (Table 4).

Table 4.

Concordance of ratings.

Dose level (%)	Reconstruction algorithm	Central PE	Segmental PE	Subsegmental PE
12.5	FBP	0.65	0.57	0.50
12.5	IMR	0.79	0.57	0.52
25	FBP	0.80	0.67	0.61
25	IMR	0.88	0.76	0.59
50	FBP	0.80	0.75	0.58
50	IMR	0.97	0.81	0.66
100	FBP	0.86	0.84	0.71
100	IMR	0.86	0.86	0.72

Inter-reader agreement κ regarding detection of central/segmental/subsegmental pulmonary embolism (PE). κ is higher for central PE as well as for higher dose levels. Minimal κ is 0.5, still showing moderate agreement. For images reconstructed with IMR, κ is >0.80 (almost perfect agreement) for central and segmental PE at the 100% and 50% dose levels.

IMR, iterative model reconstruction; FBP, filtered back projection.

By summation of all incorrectly diagnosed cases (false negative, false positive, and cases in which a diagnosis was not possible due to image quality), the total percentage of incorrect diagnosis for central, segmental, and subsegmental PE was calculated (Fig. 4).

Fig. 4.

Total percentage of incorrect diagnosis. The total percentage of incorrect diagnoses and the fraction of false negative diagnoses, false positive diagnoses, and cases in which a diagnosis was not possible due to image quality for each subset, for (a) central, (b) segmental, and (c) subsegmental PE.

Discussion

In the present study, the possibility of a further dose reduction in CTPA scans by using the next generation iterative reconstruction algorithm, IMR, was evaluated.

In our study, CNR was significantly higher in images reconstructed with IMR compared to images reconstructed with FBP at all dose levels. In addition, CNR was significantly higher in images reconstructed with IMR at the 12.5% dose level compared to images reconstructed with FBP at the 100% dose level. This result was expected as IMR is an iterative reconstruction algorithm designed to reduce noise in contrast to the conventional reconstruction algorithm FBP.

Images reconstructed with IMR show a blurred image appearance and were assessed as having an artificial image appearance whereas images reconstructed with FBP were assessed as having no artificial image appearance. This could be due to the lower noise in IMR images. In other studies, images reconstructed with iterative algorithms were also rated as having an artificial image appearance (26). As sensitivity and specificity were higher for images reconstructed with IMR compared to images reconstructed with FBP at all dose levels, artificial image appearance does not seem to have a negative influence on the diagnostic value of an examination. Therefore, artificial image appearance is not to be considered a disadvantage of IMR. It seems likely that radiologists will become accustomed to this appearance and that other quality criteria such as detectability of PE should be rated higher than a “natural” image appearance

At the full dose level (100%), image quality and diagnostic confidence regarding detection or rule out of PE were rated higher for images reconstructed with FBP compared to images reconstructed with IMR. However, sensitivity did not differ between IMR and FBP images (central and segmental PE) or was higher in images reconstructed with IMR (subsegmental PE). At the 50% dose level, image quality and diagnostic confidence did not differ between IMR and FBP images; here, sensitivity and specificity were higher for images reconstructed with IMR at all locations (central, segmental, and subsegmental PE). At lower dose levels (25% and 12.5%) diagnostic confidence was higher for IMR images than for FBP images and IMR images did show considerably higher sensitivities and specificities regarding detection of PE at all locations. Taking all this into account, diagnostic confidence does not necessarily correlate with diagnostic performance. This especially accounts for higher dose levels as at these dose levels, IMR images were rated with lower diagnostic confidence but showed better diagnostic performance. This effect is most likely due to the artificial image appearance of IMR images, which readers are not adjusted to.

As the present study shows, sensitivities and specificities >90% for central, segmental, and subsegmental PE can be maintained despite a dose reduction of 50% when images are reconstructed with IMR. Regarding central and segmental PE, sensitivities and specificities were >96% for images reconstructed with IMR down to a dose level of 25%. Multiple studies examined sensitivities and specificities in CTPA for diagnosis or rule out of PE. Hereby, three studies with >100 examined patients found sensitivities of 82–83% and specificities of 90–96% (5,27,28). Compared to these results, images reconstructed with IMR show high values for sensitivity and specificity down to a dose level of 50% even for subsegmental PE. In addition, it must be considered that embolisms in peripheral subsegmental arteries now are detected at optimum conditions with modern CT systems which could not be detected due to technical conditions in earlier studies.

Nevertheless, sensitivity and specificity decrease even at the 50% dose level compared to full dose images, especially regarding the diagnosis of subsegmental PE (sensitivity of 96.2% vs. 91% and specificity of 94.9% vs. 90.9%). According to the principle of ALARA, one must weigh up dose reduction against optimum image quality and in consequence higher sensitivity/specificity. Here, no universal suggestion can be made. However, it seems reasonable to choose a lower dose for young patients as especially in this case, dose reduction must be pursued. For older patients, a dose reduction seems not as essential as the actually used dose is already low and risk of future malignancy is lower in these patients. Additionally, in cases with shortage of time, dose should not be generally reduced as the need of a repetition of the examinations should be avoided.

We observed a sensitivity for segmental and subsegmental PE <100% even at full dose images. A sensitivity of 100% seems not achievable even by an experienced radiologist, especially in a study setting when a high number of cases has to be evaluated. Additionally, an absolute certain diagnosis regarding segmental and subsegmental PE is not always possible due to suboptimal contrast phase or due to small vessel diameter. However, each reader had to determine if a PE was present or absent in each location. In consequence, even an incorrect diagnosis by a reader is not necessarily wrong, e.g. a PE could be determined as segmental and subsegmental but was only diagnosed as a subsegmental PE by the reader. Here, a false negative diagnosis regarding segmental PE would have been noted even if the reader did not miss the PE itself.

Inter-reader agreement was higher at higher dose levels and for central/segmental PE. This is most likely as for lower dose levels and for subsegmental PE, some readers did rate images as “not diagnostic” and others did rate them as present/not present for PE. At higher dose levels and for central/segmental PE, κ was >0.8, showing excellent inter-reader agreement.

The present study has some limitations. First, iDose as a first-generation iterative reconstruction algorithm was not evaluated. In a previous study, IMR was slightly superior to iDose with both algorithms showing very good image quality even at low radiation doses. As datasets of 92 patients (736 CTPA datasets) had to be rated in this study, we decided to only evaluate FBP and IMR. Furthermore, in contrast to the previous study (21), we only used simulated tube currents of 50%, 25%, and 12.5% as image quality and detectability of PE suffered clearly at very low dose levels of 6.3% and 3.6%. In consequence, these dose levels are not of clinical relevance in the current setting. Second, readers could only use images in 3-mm axial view for evaluation. In clinical routine, multiplanar reformations as well as images with a lower slice thickness (i.e. 1 mm) are available. Despite the benefit of these additional reformations on diagnostic sensitivity and specificity, these images were not available for readers due to reasons of practicability. However, providing axial 3-mm images was sufficient for the purposes of this study, especially as identical conditions were present for all cases, reconstruction algorithms, and dose levels. Third, ground truth had to be defined by the consensus of two experienced radiologists as there is no gold standard regarding detection of PE. However, the evaluation of the full dose images by two experienced radiologists seem to be the most accurate way of defining a ground truth. Fourth, examinations with 100 kVp and 120 kVp tube voltage were included into this study and the evaluated parameters could be influenced by tube voltage. As the diagnostic performance of IMR should be evaluated for clinical routine (where both 100 kVp and 120 kVp are used), no preselection regarding tube voltage was performed deliberately to represent an average patient population. To evaluate the influence of tube voltage on the performance of IMR, further studies are needed.

In conclusion, simulated low-dose CTPA examinations with dose levels of down to 12.5% were assessed. Sensitivity and specificity >90% could be maintained even for subsegmental PE at a dose level of 50% (mean ED = 1.38 mSv); for central and segmental PE, sensitivity and specificity >97% could be maintained at the 50% dose level. Consequently, a dose reduction of 50% based on the presently used protocol in our department seems possible for a selected patient group by using IMR.

Footnotes

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Thomas Köhler and Bernhard Brendel are employees of Philips GmbH, Innovative Technologies.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Goldhaber

Pulmonary thromboembolism. In: Longo

Fauci

Kasper

et al. (eds). Harrison’s Principles of Internal Medicine , 18th edn. New York, NY: McGraw-Hill, 2012, pp. 1561–1565.

Wood

. An approach to venous thromboembolism/pulmonary embolism in the critically ill. critical care medicine: perioperative management , San Francisco, CA: Society of Critical Care Anesthesiologists (SOCCA), 2002.

Konstantinides

Torbicki

Agnelli

et al.

2014 ESC Guidelines on the diagnosis and management of acute pulmonary embolism. Eur Heart J 2014; 35: 3033–3069.

Remy-Jardin

Remy

Baghaie

et al.

Clinical value of thin collimation in the diagnostic workup of pulmonary embolism. Am J Roentgenol 2000; 175: 407–411.

Stein

Fowler

Goodman

et al.

Multidetector computed tomography for acute pulmonary embolism. N Engl J Med 2006; 354: 2317–2327.

Phillips

Straiton

Staff

. Planar and SPECT ventilation/perfusion imaging and computed tomography for the diagnosis of pulmonary embolism: A systematic review and meta-analysis of the literature, and cost and dose comparison. Eur J Radiol 2015; 84: 1392–1400.

Astani

Davis

Harkness

et al.

Detection of pulmonary embolism during pregnancy: comparing radiation doses of CTPA and pulmonary scintigraphy. Nucl Med Commun 2014; 35: 704–711.

Takahashi

Yoon

H-C

. Four-year cumulative radiation exposure in patients undergoing computed tomography angiography for suspected pulmonary embolism. Radiol Res Pract 2013; 2013: 482403–482403.

Brenner

Hall

. Computed tomography–an increasing source of radiation exposure. N Engl J Med 2007; 357: 2277–2284.

10.

Hill

Preston-Martin

Ross

et al.

Medical radiation, family history of cancer, and benign breast disease in relation to breast cancer risk in young women, USA. Cancer Causes Control 2002; 13: 711–718.

11.

Berrington de Gonzalez

Mahesh

Kim

et al.

Projected cancer risks from computed tomographic scans performed in the United States in 2007. Arch Intern Med 2009; 169: 2071–2077.

12.

Huda

Atherton

Ware

et al.

An approach for the estimation of effective radiation dose at CT in pediatric patients. Radiology 1997; 203: 417–422.

13.

Kalra

Rizzo

Maher

et al.

Chest CT performed with z-axis modulation: scanning protocol and radiation dose. Radiology 2005; 237: 303–308.

14.

Faggioni

Neri

Sbragia

et al.

80-kV pulmonary CT angiography with 40 mL of iodinated contrast material in lean patients: comparison of vascular enhancement with iodixanol (320 mg I/mL) and iomeprol (400 mg I/mL). Am J Roentgenol 2012; 199: 1220–1225.

15.

Szucs-Farkas

Megyeri

Christe

et al.

Prospective randomised comparison of diagnostic confidence and image quality with normal-dose and low-dose CT pulmonary angiography at various body weights. Eur Radiol 2014; 24: 1868–1877.

16.

Marin

Nelson

Schindera

et al.

Low-tube-voltage, high-tube-current multidetector abdominal CT: improved image quality and decreased radiation dose with adaptive statistical iterative reconstruction algorithm–initial clinical experience. Radiology 2010; 254: 145–153.

17.

Henes

Groth

Begemann

et al.

Impact of tube current-time and tube voltage reduction in 64-detector-row computed tomography pulmonary angiography for pulmonary embolism in a porcine model. J Thorac Imaging 2012; 27: 113–120.

18.

Itatani

Oda

Utsunomiya

et al.

Reduction in radiation and contrast medium dose via optimization of low-kilovoltage CT protocols using a hybrid iterative reconstruction algorithm at 256-slice body CT: phantom study and clinical correlation. Clin Radiol 2013; 68: e128–135.

19.

Kligerman

Mehta

Farnadesh

et al.

Use of a hybrid iterative reconstruction technique to reduce image noise and improve image quality in obese patients undergoing computed tomographic pulmonary angiography. J Thorac Imaging 2013; 28: 49–59.

20.

Noël

Renger

Fiebich

et al.

Does iterative reconstruction lower CT radiation dose: evaluation of 15,000 examinations. PLoS One 2013; 8: e81141–e81141.

21.

Sauter

Koehler

Fingerle

et al.

Ultra low dose CT pulmonary angiography with iterative reconstruction. PLoS One 2016; 11: e0162716–e0162716.

22.

Huda

Ogden

Khorasani

. Converting dose-length product to effective dose at CT. Radiology 2008; 248: 995–1003.

23.

Zabic

Wang

Morton

et al.

A low dose simulation tool for CT systems with energy integrating detectors. Med Phys 2013; 40: 031102–031102.

24.

Muenzel

Koehler

Brown

et al.

Validation of a low dose simulation technique for computed tomography images. PLoS One 2014; 9: e107843–e107843.

25.

Landis

Koch

. The measurement of observer agreement for categorical data. Biometrics 1977; 33: 159–174.

26.

Yamada

Jinzaki

Tanami

et al.

Model-based iterative reconstruction technique for ultralow-dose computed tomography of the lung: a pilot study. Invest Radiol 2012; 47: 482–489.

27.

Wang

Dai

et al.

Chinese multi-center study of lung scintigraphy and CT pulmonary angiography for the diagnosis of pulmonary embolism. Int J Cardiovasc Imaging 2012; 28: 1799–1805.

28.

Macdonald

Patrikeos

Thompson

et al.

Diagnosis of pulmonary embolism: ventilation perfusion scintigraphy versus helical computed tomography pulmonary angiography. Australas Radol 2005; 49: 32–38.