Introduction

Venous thromboembolism (VTE) refers to pathological thrombosis of the venous system and embolism to the pulmonary arteries. Deep vein thrombosis (DVT) of the lower extremities is the most common type of VTE [1]. Pulmonary embolism (PE) is the most fatal complication of DVT with fatality rates ranging from 7 % to 11 % according to prospective cohort studies, and other complications include post-thrombotic syndrome and pulmonary hypertension [1, 2]. Therefore, early diagnosis and proper management of DVT are important to prevent complications.

Computed tomography (CT) venography has high accuracy for diagnosing DVT of the lower extremities, which is similar to that of Doppler ultrasound (US) [3–6]. CT venography (CTV) can also evaluate DVT in the pelvic veins, which would be difficult to diagnose on Doppler US, and detect other pathologies, such as, malignancies. However, radiation exposure and the use of iodinated contrast media are disadvantages of CTV [5, 7, 8]. Because of these issues, the value of combining CTV with CT pulmonary angiography (CTPA) remains controversial [5, 9] and CTPA is not routinely included in our institution for detecting DVT.

According to the previous literature, risk for thrombosis is higher in oral contraceptives users and postmenopausal women with hormone replacement therapy [3, 10], and CTV can be used as an accurate diagnostic tool. Furthermore, CTV is currently used to evaluate varicose veins in the lower extremities, which are one of the most common diseases of adults between the second and sixth decades [11, 12]. In these situations, CT exams for diagnosis and follow-up lead to a high dose of radiation exposure, which could be harmful, especially for young women of childbearing age [5, 13]. Therefore, dose reduction techniques should be considered in CTV.

Recently, lower tube voltages have been applied to reduce radiation dose during CT acquisition, and other methods, such as tube current modulation and the usages of noise reduction filters and a higher pitch, have also been shown to be effective [14, 15].

However, lower tube voltages, such as 80 kVp, have not been widely applied to CTV, because of increased image noise and reduced image quality of the standard filtered back-projection (FBP) algorithm at low dose [14, 16, 17]. Furthermore, in patients with high BMI, image quality tends to be poorer due to increased image noise [18, 19]. Therefore, the low tube voltage technique may not be appropriate in these patients.

Recently, new iterative reconstruction (IR) methods, such as, adaptive statistical iterative reconstruction (ASiR, GE Healthcare, Waukesha, WI, USA), and model-based iterative reconstruction (Veo, GE Healthcare, Waukesha, WI, USA) have been introduced, and studies on low dose chest, abdominal, and cardiac CT using IR methods have reported improved image quality and reduced image noise as compared with FBP [16, 17, 20–28].

Therefore, the purpose of this study was to evaluate objective and subjective image qualities and image noise of CTV at 80 kVp with model-based iterative reconstruction (MBIR), and to compare these with those of FBP and adaptive statistical iterative reconstruction (ASIR).

Materials and methods

Patient selection and population

This retrospective study was approved by our institutional review board, which waived the requirement to obtain an informed consent. From November 2013 to February 2014, 48 patients who visited our vascular centre underwent CTV at 80 kVp due to clinical suspicion of DVT. Patients with suspicion of PE underwent combined CTPA and CTV, and they were not enrolled in this study. Patients who underwent arthroplasty of both knees (n = 1), with chronic DVT of both popliteal veins (n = 1), and who underwent repeated CT exams during the period (n = 2) were excluded. Finally, 44 patients were enrolled in the present study (24 men, 20 women; mean age, 56.1 ± 18.1 years). Height (mean, 1.64 ± 0.09 m) and body weight (BW; mean, 67.20 ± 14.77 kg) of the patients were also recorded and body mass index (BMI; mean, 25.03 ± 4.56 kg/m2) was calculated.

CT scanning technique

All 44 patients underwent imaging with a 64-detector CT (Discovery 750 HD, GE Healthcare, Waukesha, WI, USA) with the following parameters: tube voltage, 80 kVp; automatic tube current modulation (ATCM); section thickness, 2.5 mm; slice interval, 2.5 mm; gantry rotation time, 0.6 seconds; and pitch, 0.984. CTV was performed in the supine position from the 12th thoracic vertebra to toes in a craniocaudal direction using a single breath-hold. For contrast enhancement, all patients received 2 mL/kg BW (maximum, 120 mL) of nonionic iodinated contrast medium (Optiray 320 mg/mL; Mallinckrodt Pharmaceuticals, Dublin, Ireland) at a flow rate of 3 mL/s, and followed this with a 25 mL of 0.9 % saline solution at the same flow rate. CTV scans were initiated at 4 minutes after commencing the IV injection of contrast media. ATCM (AutomA, GE Healthcare, Waukesha, WI, USA) was used at a noise index (NI) level of 21, as determined by a pilot study performed at our institution that compared two groups (NI, 24 [20] and NI, 21) of 20 patients. The specific noise index was chosen by two subspecialty radiologists with consensus.

Image reconstruction

Raw data were reconstructed in the axial plane using a 2.5 mm slice thickness with traditional FBP as the reference standard. A blending factor of 60 % was used for ASIR, as this provides acceptable image quality according to previous studies [29, 30] and our experience with the use of ASIR for diverse indications. Images were also reconstructed with MBIR (Veo 2.0), which is a pure IR technique, and mean reconstruction time for MBIR was 120.8 ± 17.0 minutes. All images were anonymized.

Analysis of image quality and image noise

Images were loaded onto a picture archiving and communication system (PACS) workstation (Marosis, Infinitt, Seoul, Korea), and image analysis was performed using axial images and a window level of 40 Hounsfield units (HU) and a window width of 400 HU. Two readers (one radiologist with 2 years and another with 17 years of experience in vascular imaging) independently reviewed CTV images for quality and noise.

For objective analysis, vascular enhancement was quantitatively evaluated using attenuation values of the following; inferior vena cava (IVC) at the level of the left renal vein, right femoral vein at the level of the femoral head, and right popliteal vein at the level of the knee joint. When a patient had DVT of a right side vein, measurements were performed on the contralateral left side vein. Mean attenuation values for each vein were measured using a region of interest (ROI). The two readers independently placed a circular ROI on the vein to be measured to include more than two thirds of the vessel diameter and vascular enhancements were measured in HU with standard deviations (image noise) in the same slices using identical ROIs on FBP, ASIR, and MBIR image series. The readers also measured the contrast-to-noise ratio (CNR) of each vein, by placing a 90-110 mm2 circular ROI in homogeneous subcutaneous fat at the mid level of the medial thigh and in adductor muscle. CNRs were calculated by using, CNR = (VHU – MHU) / FSD (where VHU and MHU are the attenuation values of each vein and adductor muscle, and FSD is the noise of subcutaneous fat) [31]. The mean values obtained by the two readers were calculated (Fig. 1).

Fig. 1
figure 1

CT images obtained at the level of the inferior vena cava (a–c), femoral vein (d–f), and popliteal vein (g–i) for objective analysis. Contrast-to-noise ratio (CNR) was higher for MBIR (c, f, and i) than for FBP (a, d, and g) or ASIR (b, e, and h). Image noise (SD) was lower for MBIR than FBP or ASIR. However, vascular enhancements (Mean) were not different between the three reconstruction techniques. *FBP, filtered back-projection; ASIR, adaptive statistical iterative reconstruction; MBIR, model-based iterative reconstruction

Full size image

For subjective assessments, four images of each reconstructed series (IVC at the level of the left renal vein, right common iliac vein at the level of L5 vertebra, right femoral vein at the level of the femoral head, and right popliteal vein at the level of the knee joint) were saved as DICOM files and stored in image folders in random order (performed using Microsoft Office Excel 2007; Microsoft, Redmond, WA, USA). The readers subjectively evaluated image sets independently using a PACS workstation and recorded findings using a previously described template [32–35] and a 3 to 5 point scale for the following attributes: image quality, image noise, and confidence of detecting DVT (Table 1). Mean values of individual folders were subjected to statistical analysis (Fig. 2). Objective and subjective evaluations were performed at least 4 weeks apart.

Table 1 Scales used for the subjective scoring of image quality characteristics
Full size table
Fig. 2
figure 2

CT images obtained at the level of the inferior vena cava (a–c), common iliac vein (d–f), femoral vein (g–i), and popliteal vein (j–l) reconstructed by FBP (a, d, g, and j), ASIR (b, e, h, and k), and MBIR (c, f, i, and l) for subjective analysis. Image quality score was higher in the MBIR group (c, 4; f, 5; i, 5; l, 5; mean, 4.25) than in the FBP (a, 2; d, 3; g, 3; j, 3; mean, 2.75) or ASIR groups (b, 3; e, 3; h, 4; k, 4; mean, 3.50), and the confidence of detecting deep vein thrombosis (DVT) score was greater for MBIR (5; 5; 5; 5; mean, 5.00) than for FBP (2; 4; 3; 2; mean, 2.75) or ASIR (4; 5; 5; 4; mean, 4.50). Furthermore, image noise was lower for MBIR (1; 1; 1; 1; mean, 1.00) than in the FBP (3; 2; 2; 2; mean, 2.25) or ASIR groups (2; 1; 1; 1; mean, 1.25). *FBP, filtered back-projection; ASIR, adaptive statistical iterative reconstruction; MBIR, model-based iterative reconstruction

Full size image

Evaluation of radiation dose

Dose-length product (DLP) was used as a CT radiation dose descriptor and was provided by the imaging system.

Statistical analysis

The data were analyzed using dedicated statistical software (SPSS version 18; SPSS, Chicago, IL, USA). Objective and subjective image data obtained using the three reconstruction algorithms were compared using repeated measures ANOVA. A P < .050 was considered statistically significant. To determine the inter-observer reliabilities of objective and subjective assessments, intraclass correlation coefficients (ICCs) were used, and an ICC was defined by 0.00-0.20 as poor, 0.21-0.40 as fair, 0.41-0.60 as moderate, 0.61-0.80 as good, and 0.81-1.00 as excellent. Pearson's correlation analysis was performed to compare BMI with CNR among the three reconstruction techniques.

Results

Of the 44 patients, eight had DVT (18.2 %, five men and three women; mean age, 57.0 ± 23.2 years). The distribution of thrombi was as follows: common iliac vein in two patients, femoral vein in two, from common iliac vein to popliteal vein in two, and from femoral vein to popliteal vein in two (Fig. 3). All images reconstructed by MBIR were acceptable for the diagnosis of DVT.

Fig. 3
figure 3

Coronal multiplanar reconstruction images with FBP (a), ASIR (b), and MBIR (c) methods. Intraluminal filling defect was detected in left femoral vein, which was diagnosed as deep vein thrombosis. On axial image at this level, image qualities in both objective and subjective analysis were higher in MBIR than in FBP or ASIR. *FBP, filtered back-projection; ASIR, adaptive statistical iterative reconstruction; MBIR, model-based iterative reconstruction

Full size image

The mean CNR of MBIR was significantly higher than those of FBP and ASIR at the IVC, femoral vein, and popliteal vein (p < .001), and MBIR images had significantly lower objective image noise (p < .001). However, mean vascular enhancement by MBIR was not significantly different from those obtained by FBP or ASIR (p = .928 at the IVC, p = .170 at the femoral vein, and p = .900 at the popliteal vein) (Table 2). In addition, subjective image qualities (p < .001 for reader 1, R1 and p = .031 for reader 2, R2) and confidence of detecting DVT (p = .033 for R1 and p < .001 for R2) for MBIR were significantly higher than for FBP or ASIR. However, MBIR had the lowest level of subjective image noise (p < .001 for R1 and R2) (Table 3).

Table 2 Results of the objective analysis of reconstruction techniques
Full size table
Table 3 Results of the subjective analysis of reconstruction techniques
Full size table

The inter-observer reliability of MBIR was excellent by objective analysis (ICC for vascular enhancement of 0.988 and for image noise of 0.962) and good to excellent by subjective assessment (ICC for image quality 0.751, for image noise 0.828, and for confidence of detecting DVT of 0.629).

Mean DLP of 80 kVp CTV was 364.31 ± 61.20 mGy cm, and this was 10 %-70 % lower than the values reported by previous studies [33, 36–39].

There was no significant correlation between BMI and image quality in FBP (r = -0.195, p = 0.205 at the IVC; r = -0.040, p = 0.796 at the femoral vein; and r = 0.074, p = 0.635 at the popliteal vein), ASIR (r = -0.213, p = 0.165 at the IVC; r = -0.225, p = 0.872 at the femoral vein; and r = 0.610, p = 0.696 at the popliteal vein), and MBIR (r = -0.199, p = 0.195 at the IVC; r = -0.033, p = 0.831 at the femoral vein; and r = 0.062, p = 0.691 at the popliteal vein) (Fig. 4).

Fig. 4
figure 4

Scatter plots of image noise (CNR) against body mass index show no correlation at the level of the inferior vena cava (a–c), femoral vein (d–f), and popliteal vein (g–i) with FBP (a, d, and g), ASIR (b, e, and h), and MBIR (c, f, and i). * IVC, inferior vena cava; FV, femoral vein; PopV, popliteal vein. ** FBP, filtered back-projection; ASIR, adaptive statistical iterative reconstruction; MBIR, model-based iterative reconstruction

Full size image

Discussion

In this retrospective study of 80 kVp CTV with MBIR, image quality was found to be significantly increased and image noise was significantly reduced as compared with 80 kVp CTV with FBP or ASIR. However, no significant difference was found between the vascular enhancements achieved by the three reconstruction techniques. Previous studies reported that CTV at 80 kVp reduced radiation dose and improved image quality [9, 11, 38, 39]. However, application of their results to Westerners was uncertain because of increased image noise and decreased image quality in obese and heavy patients [9, 18, 19, 38, 39]. The present results demonstrate that the low tube voltage CTV with MBIR can reduce these problems.

In the present study, there was no significant correlation between BMI and CNR in all three reconstruction techniques. That could be probably due to small-sized patients that were enrolled in this study (BMI; mean, 25.03 ± 4.56 kg/m2). If many obese patients were included in our study, image quality would be decreased when BMI increased, as consistent with previous studies [18, 19].

Reduced tube voltage has the advantage of improved vascular enhancement with radiation dose reduction [36, 39]. At CTV, previously reported mean venous attenuation range with various concentrations and amounts of contrast media was 91 to 115 HU [40–43] and mean DVT clot attenuation was 51 HU or greater [40, 44]. The present study with 80 kVp CTV showed greater venous attenuation than previous reports (mean, 153.9 HU with FBP; 153.8 HU with ASIR; and 153.4 HU with MBIR), and these results were similar to those reported in the literature [9]. Furthermore, mean DLP of this study was 364.31 ± 61.20 mGy cm, and this was 10 %-70 % lower than the values reported by previous articles [33, 36–39].

The present results that dose-reduced CTV with MBIR shows significant improvements in image quality and image noise over FBP and ASIR are consistent with previous reports. Yasaka et al. [14] evaluated ultralow-dose abdominopelvic CT images reconstructed using MBIR and ASIR, and reported that MBIR improved image noise and streak artefacts significantly, and that it achieved radiation dose reductions without compromising image quality. Ichikawa et al. [16] suggested that MBIR has better image quality and lower image noise than ASIR for the detection of enlarged mediastinal lymph nodes and lower lung attenuation (bulla, emphysema, or cyst) on low-dose chest CT. Singh et al. [45] reported 50 mAs abdominal CT with MBIR provides acceptable image quality and diagnostic confidence, but that CT with FBP or ASIR does not.

FBP has remained the standard CT image reconstruction technique for decades. It is based on several assumptions that simplify CT geometry to accomplish rapid reconstruction. However, it suffers from relatively high levels of image noise and streaky artefacts, especially when low dose CT acquisition is used [16, 17, 23].

ASIR is a currently used hybrid IR algorithm. It models photons and electronic noise statistically and compares the data obtained with FBP until the algorithm converges [14, 24, 25]. Using this method, technicians can blend ASIR with FBP images to obtain final images [27]. It reduces image noise without compromising image quality, and 25-40 % dose reductions have been reported for this algorithm [20, 46–49].

The recently developed MBIR technique is a pure IR and does not involve blending with FBP data. It relies on a more complex, accurate IR algorithm than ASIR [14, 16], and in addition to using the statistical model used in ASIR, MBIR also predicts more complex models that included the modelling of system optics (size of the focal x-ray spot, shape and size of image voxels, and the size of the active detector) [21, 24]. In other words, MBIR is not based on FBP, and it is used to develop a synthesized projection model using knowledge of three main key models in the IR algorithm: the forward model (system optics model and all geometry-related effects), the noise model, and the image model. Therefore, it provides considerably reduced noise as well as much improved spatial resolution [50].

Several studies have reported significant dose reductions for low dose CT with MBIR of up to 80 % for various parts of the body [14, 16, 20, 22, 23, 51]. However, MBIR has also been suggested to have some limitations. Firstly, the MBIR algorithm requires greater computational capacity, and thus, reconstruction times are longer even when the most modern processors are used [21–23]. It takes one second to obtain a reconstructed image with MBIR, whereas FBP and ASIR can reconstruct 15 and ten images, respectively, in the same time [23]. For instance, mean reconstruction time for MBIR was 120.8 ± 17.0 minutes in this study. For this reason, CT with MBIR may be limited in emergency conditions [52, 53]. Secondly, MBIR images tend to have a unique, blotchy, pixilated texture, whereas FBP and ASIR images do not. The exact reasons for this are unknown [14, 23]. Xu et al. [54] suggested that statistical reconstruction might reduce diagnostic ability due to radiologists’ familiarity with FBP. However, in the present study, all images obtained using MBIR were acceptable in terms of detecting DVT.

The present study has several limitations. First, it was conducted using a retrospective design on a relatively small number of patients. A large prospective study is required to further assess the use of low dose CTV conducted using MBIR in DVT. Second, the diagnostic accuracies of the reconstruction techniques were not compared. However, all images reconstructed using MBIR were diagnostically acceptable, and confidence of detecting DVT with MBIR was significantly greater than with FBP or ASIR by subjective assessment. Third, due to the unique appearance of MBIR images, it was not possible to blind radiologists to the technique used during subjective analysis. Nevertheless, all image sets were randomly ordered and times between objective and subjective evaluations were more than 4 weeks to avoid adaptation. Finally, present results may not be applied to CT data performed from other vendors, because MBIR is vendor-specific.

In conclusion, CTV at 80 kVp using the MBIR algorithm provided acceptable image quality for the evaluation of DVT, and was found to be superior to FBP and ASIR in terms of objective and subjective image qualities.