Comparison of true non-contrast and virtual non-contrast images in the characterization of renal lesions using detector-based spectral CT

Abstract

Objectives:

Renal lesions are sometimes incidentally detected during computed tomography (CT) examinations in which an unenhanced series is not included, preventing the lesions from being fully characterized. The aim of this study was to investigate the feasibility to use virtual non-contrast (VNC) images, acquired using a detector-based dual-energy CT, for the characterization of renal lesions.

Methods:

Twenty-seven patients (12 women) underwent a renal CT scan, including a non-contrast, an arterial, and a venous phase contrast-enhanced series, using a detector-based dual-energy CT scanner. VNC images were reconstructed from the venous contrast-enhanced series. The mean attenuation values of 65 renal lesions in both the VNC and true non-contrast (TNC) images were measured and compared quantitatively. Three radiologists blindly assessed all lesions using either VNC or TNC images in combination with contrast-enhanced images.

Results:

Sixteen patients had cystic lesions, five had angiomyolipoma (AML), and six had suspected renal cell carcinomas (RCC). Attenuation values in VNC and TNC images were strongly correlated (ρ = 0.7, mean difference −6.0 ± 13 HU). The largest differences were found for unenhanced high-attenuation lesions. Radiologists classified 86% of the lesions correctly using VNC images.

Conclusions:

In 70% of the patients, incidentally detected renal lesions could be accurately characterized using VNC images, resulting in less patient burden and a reduction in radiation exposure.

Advances in knowledge:

This study shows that renal lesions can be accurately characterized using VNC images acquired by detector-based dual-energy CT, which is in agreement with previous studies using dual-source and rapid X-ray tube potential switching technique.

Introduction

As a result of the increased use of computed tomography (CT) and improvements in image quality, the number of incidentally detected renal lesions has increased. ¹ Of all patients undergoing CT, 41% have renal cysts, while 2% have renal cell carcinoma (RCC). ² Incidentally detected RCCs are mostly early stage, treatable tumors. Therefore, it is important to fully characterize incidental renal lesions to exclude malignancies or to set up a treatment plan in the case of malignancies.

To fully characterize renal lesions, both an unenhanced and a contrast-enhanced CT scan are needed. Unfortunately, incidental renal lesions are commonly found during examinations in which an unenhanced series is not available. Consequently, additional imaging is recommended in the form of a multiphasic renal lesion CT protocol, including an unenhanced CT scan. ³ This is detrimental in terms of radiation dose, costs, resources, and patient burden.

Dual-energy CT or spectral CT can alleviate the above problem as it allows for the reconstruction of so-called virtual non-contrast (VNC) images. These VNC images are usually reconstructed using a three-material decomposition algorithm. ⁴ Using such algorithm, the contribution of iodine to the total attenuation value is effectively subtracted in each voxel. VNC images are useful for the management of incidentally detected renal lesions as they obviate the need for an additional scan, which can lead to a substantial reduction of exposure to radiation and eliminates possible inaccuracies due to spatial misalignment. ^3,5

The clinical use of VNC images is currently still limited due to a relatively high noise level, susceptibility to artefacts and reduced sensitivity for small calcifications. ⁶ Studies have shown that VNC images derived from detector based dual-energy CT demonstrate attenuation values similar to unenhanced images in various tissues. ^7,8 Specifically, dual-energy CT can provide high-quality VNC images for the characterization of renal lesions, as well as for other imaging applications including cardiac protocols. ^9–11 In a study by Meyer et al, 379 renal lesions were characterized by four different radiologists, using VNC and unenhanced (true non-contrast, TNC) images. ³ Despite a small reduction in diagnostic performance compared to TNC images, this study demonstrated that VNC images enabled accurate renal lesion characterization. However, in that and other studies, images were acquired using dual-source CT or using rapid X-ray tube potential switching. To the best of our knowledge, the capability to assess renal lesions using VNC images acquired using detector-based dual-energy CT has not yet been investigated. Therefore, the aim of the current study was to investigate the accuracy in characterizing renal lesions using VNC images, compared to TNC images, acquired with a detector-based dual-energy CT scanner.

Methods and materials

Patients

Twenty-seven patients (12 women) with a mean age of 70 (range 46–81) years were retrospectively included in this study. Between December 2020 and March 2021, these patients had undergone a renal dual-energy CT with tumor protocol, including a non-contrast, an arterial phase contrast-enhanced, and a venous phase contrast-enhanced scan. Patients in which one or more renal lesions were detected according to the reports of the radiologists were included in the study. There were no further specific exclusion criteria. For patients with multiple renal lesions, up to three lesions were randomly selected.

Data acquisition

A detector-based dual-energy CT scanner (IQon Spectral CT, Philips Healthcare, The Netherlands) was used in this study. Scans were obtained using a tube potential of 120 kVp, a detector configuration of 64 × 0.625 mm, a detector width of 40 mm, a pitch factor of 0.925, a revolution time of 0.4 s and were reconstructed using a slice thickness of 2 mm. All images were acquired during breath hold. The unenhanced scan was acquired first. Then, 100 ml of iodine containing contrast fluid (Omnipaque 300 mg ml⁻¹, GE Healthcare, The Netherlands) was injected intravenously to the patient via an antecubital vein. The arterial phase contrast-enhanced scan was acquired 35 to 40 s after injection. After another 65 s, the venous phase contrast-enhanced scan was performed. The VNC images were reconstructed from the venous phase contrast-enhanced images using a proprietary algorithm on the CT scanner. Axial images were reconstructed with a slice thickness of 2 mm using iterative reconstruction (IMR 1) and a routine filter kernel, and were saved in DICOM format. Post-processing was performed in MATLAB (R2020a, MathWorks, Natick, MA, USA).

Image analysis

The study was divided into two parts: a quantitative, and a qualitative analysis. The quantitative analysis was carried out to measure and compare renal lesion attenuation values in VNC images with the golden standard, TNC images. For the qualitative analysis, three radiologists characterized the renal lesions using visual grading characteristics (VGC) analysis of VNC and TNC images.

Quantitative analysis

Regions of interest (ROI) were manually selected in the venous phase contrast-enhanced and the TNC image. All ROI’s were freehand regions, which were drawn by the same person, i.e. the first author. Lesions were usually clearly visible and delineated on the venous phase contrast image. Care was taken to exactly encompass the lesion in the corresponding TNC images. The ROI from the contrast-enhanced image was copied and pasted directly onto the VNC image, since the VNC and venous phase contrast-enhanced scan are perfectly aligned. The attenuation values were expressed as the mean ± 1 SD (HU) within the ROI. This procedure was repeated for all of 65 included lesions.

The lesions were divided into four groups based on their characteristics in the TNC images combined with the venous phase contrast-enhanced images. The first step was to determine if the lesion was enhanced or not. A lesion was defined as enhanced when the change in attenuation value after the injection of the contrast fluid exceeded 20 HU (ΔCT number>20 HU). ¹² The enhanced lesions were divided in fat containing (CT number < −10 in TNC image) and non-fat containing (CT number > −10 HU in TNC image). The non-enhanced lesions were divided in high-attenuating (CT number>20), and low-attenuating (CT number<20). ³ The process to divide the lesions in four groups is summarized in the flowchart in Figure 1.

Figure 1.

Flowchart used to divide the 65 renal lesions into four groups based on CT numbers. There were 13 enhanced lesions and 52 non-enhanced lesions. The enhanced lesions were divided into four fat-containing, and nine non-fat containing lesions. The non-enhanced lesions were divided into 20 high-attenuating, and 32 low-attenuating lesions.

The correlation and agreement between the attenuation values in VNC and TNC images, and the sensitivity and specificity of the classification of enhanced lesions using VNC images were calculated using Spearman’s ρ. Enhanced lesions were considered true positive when the change in attenuation value between the venous phase contrast-enhanced image and the TNC image exceeded 20 HU. ¹² The sensitivity, and specificity was investigated for other threshold values as well.

Qualitative analysis

Three radiologists, two with three years, and one with 12 years of clinical experience were asked to characterize the 65 renal lesions. The lesions could be characterized as a solid angiomyolipoma (AML), a renal cell carcinoma (RCC), or a cyst. For cystic renal lesions, the Bosniak classification was used to distinguish between different types of cysts. ¹³ An unenhanced lesion (∆CT number<20 HU) was classified as a cyst of type Bosniak I, II, or IIF. ¹ If parts of the lesion were enhanced (∆CT number>20 HU), it was classified as a cyst of type Bosniak III or IV. In this case, the radiologist looked for the presence of fat. A fat-containing lesion was defined as a lesion with an attenuation value in the unenhanced image < −10 HU and was classified as AML. ¹ When the lesion did not contain fat, it was classified as RCC. The characterization of lesions is clinically used to determine the need for follow-up and type of treatment. However, the clinical follow-up was outside the scope of this study.

Assessment of the images was performed using the ViewDEX software on a review screen (NX242, EIZO Europe GmbH, Eindhoven, The Netherlands). ^14–16 For each patient, the arterial phase contrast-enhanced, the venous phase contrast-enhanced, and the unenhanced images were presented to the radiologists. The radiologists characterized all 65 lesions twice: the first time using VNC images, and the second time using TNC images. The characterization using VNC images was performed first, so that the radiologists could not be biased by the TNC images, which were defined as the ground truth. The radiologists were allowed to scroll through the slices, zoom, adapt window/level, and measure lesion attenuation values by manually selecting ROI. They did not have access to the patient study reports during the image assessment.

Data analysis

Data in text and tables are given as mean ± SD unless indicated otherwise. The correlation and agreement between the attenuation values in VNC and TNC images, was assessed using Spearman’s correlation coefficient, the Wilcoxon signed rank test, and a Bland Altman plot. Fisher’s exact test was used to determine the agreement in the classification of renal lesions using VNC and TNC images. The Fleiss κ statistic was used to evaluate the interrater agreement among the three radiologists. A p-value of <0.05 was considered as statistically significant. Statistical analyses were done using MATLAB (R2020a, MathWorks, Natick, MA, USA).

Results

A total of 65 renal lesions were investigated. Sixteen patients had cystic lesions, five patients had angiomyolipoma (AML), and six patients were suspected of having a renal cell carcinoma (RCC). The scanning range was 289 ± 55 mm for the unenhanced series, 271 ± 30 mm for the arterial phase and 474 ± 51 mm for the venous phase contrast-enhanced series. The mean CTDI_vol was 6.8 ± 3.2 mGy for the unenhanced series, 7.6 ± 3.6 mGy for the arterial phase and 6.7 ± 2.7 mGy for the venous phase contrast-enhanced series. The mean DLP was 211 ± 101 mGycm for the unenhanced series, 225 ± 110 mGycm for the arterial phase and 336 ± 136 mGycm for the venous phase contrast-enhanced series. The mean total DLP for the studies was 833 ± 382 mGycm.

Quantitative analysis

Attenuation values measured in VNC and TNC images are summarized per group in Table 1. The median CT numbers of lesions in VNC images were lower than in TNC images in all groups, except for the enhanced fat-containing lesions. A Wilcoxon signed-rank test revealed a significant difference in means and standard deviations of renal lesion attenuation between VNC and TNC images in all groups (p < 0.01).

Table 1.

CT numbers (median±interquartile range) measured in renal lesions in true non-contrast (TNC) and virtual non-contrast (VNC) images

	TNC a	VNC b	p-value
Unenhanced
Low attenuation	8.4 (5.8–13.2)	5.0 (-0.04–9.4)	< 0.01
High attenuation	27.7 (24.7–33.0)	16.3 (5.6–23.7)	< 0.01
Enhanced
Non-fat containing	34.0 (24.4–36.9)	28.0 (12.5–34.3)	< 0.01
Fat-containing	−50.5 (-62.0–-28.8)	−36.6 (-48.1–-10.7)	< 0.01

^aTrue non-contrast

^bVirtual non-contrast

There was a strong positive relationship between the renal lesion attenuation values in VNC and TNC images (Figure 2a). The Spearman’s correlation coefficient was ρ = 0.7 (p < 0.001). The mean difference between attenuation values in VNC and TNC images was −6 ± 13 HU, with a 95% confidence interval between −31.6 HU and 19.6 HU (Figure 2b). The largest differences were present in the group of unenhanced high-attenuation lesions. The difference was positive for low, and negative for high attenuation values in TNC images.

Figure 2.

Correlation and agreement between lesion attenuation values in VNC and TNC images. (a) Renal lesion CT number in the VNC image as a function of CT number in the TNC image. There was a strong positive relation between the VNC and TNC attenuation values. (b) Bland-Altman plot to visualize agreement between lesion CT numbers in VNC and in TNC images. The mean difference was −6.0 HU and the upper and lower limits of the 95% confidence interval were 19.6 and −31.6 HU, respectively.

The accuracy of the characterization of renal lesions using VNC images is visualized in Figure 3. The conventional enhancement threshold of 20 HU yielded three false negatives and ten false positives, which corresponds to a sensitivity of 0.77, and a specificity of 0.79. Decreasing the enhancement threshold of VNC images to 10 HU led to a sensitivity of 1.0. However, the specificity decreased to 0.63. On the other hand, increasing the threshold to 51 HU resulted in maximal specificity of 1.0, and a strongly decreased sensitivity of 0.23.

Figure 3.

Renal lesion enhancement in the TNC images versus the enhancement in the VNC images. Using the conventional enhancement threshold of 20 HU, the green areas indicate true negatives (TN) and true positives (TP). The red areas indicate false positives (FP), and false negatives (FN). The lines of maximal sensitivity and specificity are indicated at 10 and 51 HU, respectively.

Qualitative analysis

Table 2 shows the occurrence of all possible VNC-TNC combinations of categories that could be assigned to the lesions. It was found that, when assessing VNC images, 86% of the lesions were classified identically by the radiologists as when assessing the TNC images, which were considered the ground truth. Using VNC versus TNC images, radiologists classified 64.1 versus 66.2% of the lesions as Bosniak I cyst, 8.2 versus 9.2% as Bosniak II, 5.6 versus 4.1% as Bosniak IIF, 3.1 versus 2.1% as Bosniak III, 1.5 versus 4.1% as Bosniak IV cysts, 9.2 versus 7.7% as AML, and 8.2 versus 6.7% as RCC. Fisher’s exact test revealed a high association between classifications using VNC and TNC images (p < 0.01). The interrater agreement using VNC images was 0.37 (fair agreement), while it was 0.42 (moderate agreement) using TNC images. ¹⁷

Table 2.

Combinations of lesion categories that could be chosen by the radiologists using virtual non-contrast (VNC) (rows) versus true non-contrast (TNC) images (columns)

	TNC imagea
VNC image	Bosniak I	Bosniak II	Bosniak IIF	Bosniak III	Bosniak IV	AML	RCC suspect
Bosniak I	117	3	1	1	1	1	1
Bosniak II	2	12	0	0	0	2	0
Bosniak IIF	3	1	7	0	0	0	0
Bosniak III	1	0	0	3	2	0	0
Bosniak IV	0	0	0	0	3	0	0
AML	5	0	0	0	0	11	2
RCC suspect	0	2	0	0	2	1	10

^a Numbers indicate the number of cases for which the combination occurred. The numbers marked in boldface indicate the lesions that were classified correctly. TNC: true non-contrast, VNC: virtual non-contrast, AML: angiomyolipoma, RCC: renal cell carcinoma.

Figure 4 shows the difference between VNC and TNC attenuation values for the 25 VNC-TNC classification pairs as characterized by the radiologists. For the correctly classified lesions, there was a small difference between VNC and TNC attenuation values (−4.9 ± 12.2 HU), although a relatively large difference (+20 HU) in attenuation between VNC and TNC images was found for the correctly classified AML lesions. For the incorrectly classified lesions, the difference in attenuation values was higher (−11.5 ± 15.4 HU).

Figure 4.

Boxplots showing the difference between VNC and TNC attenuation values for the 25 VNC-TNC classification pairs as characterized by three radiologists. For each pair, the classification assigned using VNC images is indicated on the lower x-axis, and the that using TNC images on the upper x-axis. The Bosniak I up to IV cysts are abbreviated with I, II, IIF, III, and IV. The grey boxes indicate the lesions that were classified correctly.

Discussion

The characterization of renal lesions is dependent on the difference in lesion attenuation value between the unenhanced and contrast-enhanced CT image series. Therefore, to adequately classify renal lesions, it is important that the attenuation values in VNC images are sufficiently close to those of true non-contrast (TNC) images.

We found a statistically significant difference in the mean and standard deviation of the attenuation values between VNC and TNC images (p < 0.01), which is in agreement with previous studies. ⁷ However, despite the measured differences, 86% of the lesions were classified correctly by the radiologists during the qualitative image analysis. Fisher’s exact test also revealed that there is a high association between the chosen classifications using VNC and TNC images (p < 0.01). Furthermore, the interrater agreement using VNC was 0.37 (fair agreement), which was close to that of TNC images: 0.42 (moderate agreement). Therefore, this study suggests that VNC images can be valuable in the characterization of renal lesions when a TNC image is unavailable.

The attenuation values in VNC images were lower than those in TNC images for all lesion groups except for the enhanced fat-containing lesions. This is in agreement with previous studies using other dual-energy techniques. ^3,10 The group of lesions that showed the largest difference between VNC and TNC attenuation values were the unenhanced high-attenuation lesions, as was also found in the study of Meyer et al. ³ This is possibly caused by the fact that the content of unenhanced high-attenuation lesions, for example, iron and colloid formation, have the same ratio of high and low-energy attenuation values as iodine. Therefore, the amount of iodine is overestimated, resulting in too much iodine subtraction and overestimation of the enhancement. As a result, some true unenhanced renal lesions were classified as enhanced lesions (false positives). On the other hand, in the group of enhanced fat-containing lesions, the attenuation value was overestimated in VNC images, which caused some true enhanced renal lesions to be classified as unenhanced lesions (false negatives). A clear explanation for this effect was not found. However, this group with only four renal lesions was very small and is possible that these lesions were misclassified due to chance. Including a larger number of this type of lesions in a future study could elucidate this question.

Five true Bosniak I cysts were incorrectly classified as AML’s based on the VNC images. For those lesions, the difference in attenuation between VNC and TNC images was very large (−26.5 ± 12.0 HU). The TNC, contrast-enhanced and VNC images of one of those lesions are shown in Figure 5a. The lesion appeared much darker in the VNC image, compared to the TNC image (-20.1 vs 9.4 HU). Obviously, this resulted in a larger enhancement based on the VNC image, compared with the TNC image (45.6 vs 16.1 HU). Therefore, the lesion was classified as an enhancing fat-containing lesion, i.e., an AML, while it is a true unenhanced lesion with the properties of a Bosniak I cyst.

Figure 5.

Examples of three lesions in the TNC (left), contrast-enhanced (center), and VNC image (right). The lesions are indicated with the red ROIs. The corresponding attenuation values (mean standard deviation) are indicated in the lower left corner, and the mean enhancement values in between the enhanced and unenhanced images. Upper row: A Bosniak I cyst, incorrectly classified as an AML; middle row: a Bosniak II cyst, incorrectly classified as an RCC; and lower row: correctly classified Bosniak III/IV cyst. Calcifications are indicated with blue arrows and are blurred in the VNC image. The enhancing part of the lesion wall is indicated with the blue ROI.

The fact that the attenuation values of lesions in VNC images were lower than in TNC images, sometimes caused lesions to be wrongly classified as malignant carcinomas. An example of this is shown in Figure 5b. The attenuation value in the VNC image was 19.4 HU, while in the TNC image it was 36.4 HU. As a result, the lesion was enhanced (∆CT number = 31.4 HU) based on the VNC image, and unenhanced (∆CT number = 14.4 HU), based on the TNC image. The lesion was classified as a suspect for RCC, whereas it was, in fact, a hyperdense cyst of type Bosniak II.

VNC images are less sensitive for small calcifications, compared to TNC images and this may affect the classification. In Figure 5c, the calcifications indicated with the red arrows appear a smaller in the VNC images, compared to the contrast-enhanced and TNC image, but they are still visible. Furthermore, a part of the lesion wall, indicated with the blue ROI is enhanced using both the TNC and the VNC images (∆CT number = 22.1 versus ∆CT number = 26.3 HU). Therefore, the radiologists were still able to classify this lesion correctly.

In terms of diagnostic accuracy, it would be unwarranted to just replace TNC images with VNC images, because of the limited sensitivity, i.e., the presence of false negatives when using VNC images. False negatives are detrimental, because missing enhanced lesions that are possibly a malignant tumor and could metastasize in the future, is undesirable. According to the quantitative analysis results, false negatives can be prevented by decreasing the enhancement threshold from 20 HU to 10 HU. This reduced threshold would result in a sensitivity of 1. However, the specificity would decrease to 0.63, resulting in more false positives. Performing unnecessary surgeries to remove benign renal lesions is not desirable either. For the data used in this study, an enhancement threshold of 51 would prevent all unnecessary surgeries (specificity of 1), but the sensitivity would decrease to 0.23.

Based on these findings, the following management protocol for incidentally detected renal lesions could be proposed: lesions with an enhancement smaller than 10 HU based on the VNC image should be classified as benign cysts; lesions with an enhancement larger than 30 HU should be considered solid renal lesions that are possibly malignant. For the remaining lesions, an additional TNC series should be obtained to completely characterize the lesion. When applying such protocol to the data in this study, no malignant lesions would have been missed, while 7.7% of the lesions would be wrongly treated as malignant lesions. Moreover, using this protocol, for 51% of the patients, malignancy could have been excluded and for 20% of the patients, the management of solid renal lesions could have been started without the need for an extra scan series. For the remaining 29% of the patients, an additional TNC series would have been required to completely characterize the lesion(s). Thus, for the patient population in this study, by using VNC images for the characterization of incidental renal lesion, additional CT examinations could have been prevented for 71% of the patients.

This study has limitations. Classifications obtained with TNC images were considered as ground truth. However, these classifications are not necessarily true. An example of more reliable ground truth is the outcome of histopathologic analysis after surgery or percutaneous biopsy. Another limitation was that the radiologists assessing the images were under time constraints due to the high workload at the hospital. Moreover, they had to assess the lesions using another software program than they are used to. These factors may have caused some inaccuracies in the renal lesion classification, which would also explain the fair to moderate interrater agreement. Finally, the group of 65 lesions resulting from 27 patients was relatively small, compared to the study of Meyer et al, resulting in particularly small groups of enhanced fat-containing and non-fat-containing lesions. ³

The results of this study should be interpreted with care. Before the assessment of renal lesions can be based on VNC images a larger group of lesions should be investigated first. Future research is needed to determine the minimal size of calcification that can be detected in VNC images. This could be done using a phantom study with different calcification sizes. If the reliability of VNC images, reconstructed using a detector based dual-energy CT scanner can be proven in future research, the unenhanced scan could be removed from all CT imaging protocols. This would result in a significant reduction of the patient radiation dose. VNC images could possibly also be useful for characterization of lesions in other organs, like the adrenal glands or the liver.

It can be concluded that VNC images, acquired with a detector based dual-energy CT scanner, can be used to characterize incidentally detected renal lesions with a high accuracy. The results of this study suggest that in approximately 70% of the patients, the VNC image can be used when an unenhanced series is not available. This could result in less patient burden and a substantial reduction in exposure to ionizing radiation.