Comparison of enhancement quantification from virtual unenhanced images to true unenhanced images in multiphase renal Dual‐Energy computed tomography: A phantom study

1. INTRODUCTION

Multiphase computed tomography (CT) exams are a noninvasive imaging technique commonly used for the diagnosis of renal masses. 1 These exams include a true unenhanced (TUE) phase of imaging, the administration of an iodinated contrast agent, followed by one or more postcontrast phases. Enhancement can be quantified by calculating the difference in CT number between TUE and postcontrast images for a given region of interest (ROI). It is crucial that the quantification of enhancement be accurate, as it has been characterized as the most important criteria in determining surgical from nonsurgical renal masses. 2 Previously, a change in CT number of 10 Hounsfield units (HU) or more between pre‐ and postcontrast images was considered positive for enhancement; however, with the advent of helical CT it has been proposed that this threshold should be increased to account for helical interpolation. 3 Enhancement is now commonly characterized by a change of 20 or more in measured CT number between TUE and postcontrast images, although this number is not universally agreed upon. 2 As a result, a mass with enhancement measuring between 10‐20 HU can be considered “intermediate” and may require further evaluation. 2

Renal cell carcinoma (RCC) is the most common kidney cancer in adults, accounting for approximately 90% of renal neoplasms and 3% of all adult malignancies. 4 RCC is an aggressive disease that has a 5‐year survival rate of 95% for Stage 1 disease, but less than 20% for Stage 4 disease. 5 The diagnosis of RCC based on the appearance of a lesion on CT imaging can vary widely in difficulty. While the diagnosis of a simple nonenhancing cyst is straightforward, classifying complex lesions can be much more challenging. 1 Studies have shown that if the patient has an enhancing renal mass, such as RCC, the mass will have a substantial noncalcified region with a CT number measuring within a range of 20‐70 HU on unenhanced CT. 6 In a postcontrast scan acquired during the corticomedullary phase, studies have shown that RCC will enhance significantly more than a benign cyst (81.4 HU vs 27.4 HU, respectively) and that a difference of >42 HU in measured enhancement during the corticomedullary phase was highly predictive of RCC with 97.1% sensitivity and 85.7% specificity. 7

Dual‐Energy CT (DECT) is an extension of conventional CT in which two datasets are acquired using different photon spectra nearly simultaneously. 8 This can be achieved either by using a single X‐ray tube that rapidly switches between a high and low kVp at each projection angle, scanning the patient twice using different kVp, scanning the patient with dual X‐ray sources and detector arrays, or using a dual‐layer detector with a single X‐ray source. This work uses the rapid kV‐switching technique in which the X‐ray tube alternates between 80 and 140 kVp at each projection with a constant tube current of approximately 600 mA to acquire co‐registered dual‐energy projections. 9 Benefits of rapid kV switching include excellent temporal registration, which reduces the potential for motion artifacts, and the availability of the entire scan field of view (SFOV) for DECT image acquisition. 10 A technical challenge of this technique is the rise and fall times of the high voltage waveforms, which complicates the determination of the effective energy for the high‐ and low‐kVp projections. 11

DECT provides the ability to exploit the attenuation properties of materials to apply material decomposition techniques. This is achievable because each material has a unique attenuation coefficient, based on a unique combination of Compton and photoelectric interaction probabilities. A basis pair of materials with a large separation in linear attenuation coefficients can be chosen, commonly water and iodine, and used for material decomposition. By assuming each voxel is a weighted combination of the basis pair, the amount of iodine in each voxel can be estimated when the object is imaged at different energies. Theoretically, material decomposition can be generalized to decompose an arbitrary number of materials 12 ; however, this work focuses on basis pair decomposition. Material decomposition is the basis for the reconstruction of virtual unenhanced (VUE) images, in which the estimated volume of iodine in each voxel is replaced by an equivalent volume of blood. 13

The use of VUE imaging provides the potential to use VUE images in place of TUE images in multiphase renal CT exams. Eliminating the precontrast phase of imaging could reduce patient dose and increase patient throughput, consequently improving clinical efficiency. Previous studies have investigated the feasibility of using VUE images in place of TUE images for patients with gastric tumors, resulting in a dose reduction of 30.5%, and in the diagnosis of patients with subarachnoid haemorrhage. 14 , 15 For imaging of renal lesions, it has been shown that a threshold of 2 mg/cm ³ is the most accurate in distinguishing enhancing from nonenhancing lesions using iodine density images generated from DECT. 16 Other studies have investigated the feasibility of replacing precontrast images with virtual noncontrast images in renal DECT exams. 8 , 17 , 18 , 19 To our knowledge, there has not been a study conducted specifically assessing the feasibility of replacing precontrast images with VUE images for evaluation of renal masses across a wide range of clinical scenarios for the rapid kVp‐switching DECT technique. The aim of this phantom study was to investigate the accuracy and sensitivity when measuring enhancement using VUE images across a variety of clinical conditions to assess the potential of replacing TUE images in diagnostic renal CT exams with VUE images derived from rapid‐kV‐switching DECT technology.

2. METHODS

The technique employed in the VUE image reconstruction is believed to utilize a two‐material decomposition technique, namely water and iodine. 8 It can be assumed that iodine has displaced blood in postcontrast imaging; therefore, the amount of iodine estimated in each voxel can be replaced by an equivalent volume of blood to generate a VUE image (Fig. 1 ). 13 This method of VUE image reconstruction is based on the assumption that materials within each voxel mix to form an ideal solution.

Phantoms were constructed and used to compare the accuracy of measured enhancement when VUE images were used as the baseline versus TUE images across a range of simulated clinical scenarios. Several variables known to affect measured CT number were evaluated. These variables included patient size, lesion size, Gemstone Spectral Imaging (GSI) protocol used, and level of simulated enhancement.

Three elliptical cylinder phantoms were designed and constructed for this study, referred to here as the small, medium, and large phantoms. Each phantom was composed of four plates made of high density polyethylene (Fig. 2 ). The major/minor axes of the phantoms were selected to correspond to the 5 ^th (28.3/17.4 cm), 50 ^th (36.1/22.2 cm), and 95 ^th (47.9/29.4 cm) percentiles of the adult population of the United States. Dimensions were calculated from the PeopleSize (Open Ergonomics, Ltd., Leicestershire, UK) anthropometric database. Lesion size and location were variable within each phantom, which was facilitated by fabricating three interchangeable sets of the two interior plates for each phantom. Using these plates, each phantom could contain a 1.0, 2.0, or 3.0 cm diameter spherical insert in the periphery. Additionally, the phantom included a 1.0‐cm spherical insert and a 1.0‐cm‐diameter cylindrical insert to the left and right of a Delrin rod, which was included to represent the spine (Fig. 3 ).

The phantom study was designed to simulate nonenhancing, intermediately enhancing, and highly enhancing renal lesions. A conceptual summary of all enhancement scenarios simulated is given in Table 1 . Data for each scenario were acquired in the same general fashion using a single Discovery HD750 CT scanner (GE Healthcare, Waukesha, WI) in helical mode. Dual‐energy data processing and VUE image reconstruction was performed using version 2.0 of the GSI Volume Viewer (GE Healthcare, Waukesha, WI). For all simulations, the phantom was positioned identically for the pre‐ and postcontrast scans. In the postcontrast configuration, a single‐energy CT (SECT) scan was acquired at 120 kVp followed by the DECT dataset. The acquisition and reconstruction parameters used for both pre‐ and postcontrast CT data acquisition are provided in Table 2 , where the mAs for each GSI protocol used in the study is specifically detailed. Note that the mAs used for dual energy acquisition is linked to the GSI protocol selected. VUE images were reconstructed from DECT datasets using the Material Suppressed Iodine (MSI) algorithm.

Zero enhancement was simulated by imaging each phantom configuration with distilled water in the phantom’s inserts for pre‐ and postcontrast imaging. Intermediate enhancement was simulated by acquiring TUE images with a water‐blood surrogate mixture in each insert to achieve precontrast CT densities of 20 and 40 HU. These values were chosen because they corresponded to the lower bound and typical value for the known RCC “danger zone” of 20‐70 HU on precontrast imaging, which allows for simulation of borderline lesions. 6 Note that apple juice was used as a blood surrogate in this study, as it was found to have a similar effective atomic number, density, and CT number to blood. Iodinated contrast (Optiray 320, GE Healthcare, Waukesha, WI) was added to each insert for postcontrast imaging to simulate 10 HU (low) and 20 HU (borderline) enhancement levels for each of the baseline precontrast values (Table 1 ).

Previous studies have found that the typical precontrast CT number of RCC is approximately 35‐40 HU, 20 and that the known unenhanced CT number range for RCC is 20‐70 HU. 6 Therefore, enhancing lesions were simulated by first acquiring TUE images with a water‐blood surrogate mixture to achieve a CT number of 40 HU. Enhancement was simulated by adding iodinated contrast to achieve a postcontrast CT number of approximately 80 HU in the spherical simulated lesion inserts. This value was chosen because enhancement of 42 HU or more has been shown to be highly predictive of RCC. 7

An empirical relationship between CT number (HU) at 120 kVp and iodinated contrast concentration (mg/mL) was calculated and used to determine the contrast needed to achieve each desired enhancement. DECT GSI protocols were selected to evaluate the widest range of acquisition variables possible. The selection of GSI protocols was informed by a previous experiment which compared the CT numbers measured in VUE image for all GSI protocols to the CT number measured in the TUE image, which was considered the ground truth. A subset of four GSI protocols were selected for this experiment. The GSI 10 protocol was used for all phantom sizes to allow for a direct comparison of the effect of phantom size on measured enhancement for a given technique. VUE images were reconstructed from the DECT dataset using the MSI algorithm available on the GSI volume viewer. Measured enhancement was quantified for the SECT dataset by calculating the difference in measured CT number between the 120 kVp postcontrast scan and the TUE images (henceforth referred to as ΔTUE). Similarly, enhancement was calculated for the DECT dataset as the difference in CT number between 70‐keV monochromatic images processed from the DECT postcontrast dataset and the VUE images (henceforth referred to as ΔVUE). The sizes of ROIs used to measure enhancement were identical in the pre‐ and postcontrast images and across experiments. The ROI size used was 14.4 mm ² (1‐cm sphere and hollow rod), 70‐80 mm ² (2‐cm insert), and 140‐150 mm ² (3‐cm insert). Each ROI was placed in the center of the insert on the central slice of the phantom. Variation between ROI measurement locations was considered as a source of random error in the statistical model. All images were reconstructed at an image thickness of 5.0 mm using the STANDARD kernel with no iterative reconstruction applied.

Data were analyzed using Bayesian hierarchical models that incorporated all sources of experimental uncertainty. In each model, the Bayesian method was used to estimate the parameters of the posterior distribution in multiple levels, then combined to form the hierarchical model. This method allowed for accounting of all sources of uncertainty in the study. The result of this analysis was the posterior probability. Separate models were built for nonenhancing, intermediate enhancing, and positively enhancing simulated lesions. Sources of random error incorporated into the model included CT scanner variability (σ _S ), variation between ROI measurement location within the simulated lesion (σ _M ), and variation across the ROI, or noise (σ _N ). For the experimental scenarios, all images needed for each phantom configuration were acquired using a single preparation of the solution used to fill the inserts. A 2‐mL serological pipette with a specified precision of ± 0.01 mL (Fisherbrand, Fisher Scientific, Pittsburgh, PA) was used to pipette the Optiray 320. The total experimental variation was calculated by adding each source of variation in quadrature for each measurement. The fixed effects incorporated in the model were lesion size, lesion location, phantom size, enhancement, and GSI protocol used.

A posterior probability > 0.95 for the difference between ΔVUE and ΔTUE returned by the Bayesian hierarchical model indicated that ΔVUE was significantly higher than ΔTUE enhancement measurements, while a posterior probability < 0.05 indicated that ΔVUE was significantly lower than ΔTUE. A posterior probability of 0.05‐0.95 indicated no significant difference between ΔVUE and ΔTUE. Note that the credible interval (CI) is a range within which lies some predetermined percentage (e.g., 95%) of the posterior distribution of the parameters given the data and can be interpreted as the Bayesian analogue of the confidence interval.

3. RESULTS