Over the last decades, computed tomography (CT) has proven to be an invaluable tool in clinical diagnostics. State-of-the-art systems allow for the acquisition of whole-body patient data in seconds and the resulting reconstructions provide a spatial resolution in the order of less than a millimeter. Besides conventional single energy imaging based on the integrated information from a single x-ray spectrum, dual-energy CT (DECT) techniques have been introduced to clinical practice extending diagnostic capabilities even further. Popular applications include the computation of virtual monochromatic (VMI) and virtual non-contrast (VNC) images or material characterization and decomposition [1,2,3]. Various techniques have been proposed and are currently used in clinical CT systems to realize DECT. Most commonly, DECT is performed by using two distinct source-detector-pairs, rapid kV-switching, dual-layer sandwich detectors, or split filters [4,5,6,7]. Recently, prototypes of novel photon-counting (PC) detectors became available extending the range of dual- and multi-energy acquisition strategies. In the case of energy-integrating (EI) detectors, an incoming x-ray photon is absorbed in the scintillator, typically Gd2O2S, resulting in the emission of optical photons that are detected in photo diodes eventually forming the desired signal. In the case of photon-counting detectors, the scintillator is replaced by a semiconductor, usually cadmium telluride (CdTe). The absorption of an incoming x-ray photon results in the formation of a charge cloud that is transported to electrodes that are constituting the different detector pixels using a bias voltage in the order of 1 kV [8,9,10]. The formed signals allow not only for the counting of single photons but also for the quantification of their energy. Typically, x-ray photons are sorted into two to four bins according to their energy, facilitating dual- or multi-energy acquisitions. In the case of two energy bins, a single threshold is used and photons with energy below this threshold are sorted to the lower bin and photons with energy above this threshold are sorted to the upper bin. This threshold can usually be placed almost arbitrarily with respect to the detected x-ray spectrum and influences the image quality of derived image sets, e.g., iodine maps or VMIs. Since the detectors always acquire at least two energy bins, spectral data are available even if no dedicated dual-energy acquisition was performed. A retrospective dual-energy evaluation of the acquired data might be desired, e.g., in the case of incidental findings or to retrospectively improve contrast agent enhancement. Thus, thresholds should always be set to maximize image quality even if a DECT evaluation is not within the initial scope of the examination. Hence, the influence of threshold position on the image quality of derived image sets, e.g., the iodine material maps, is of high interest and should be prospectively considered when setting up scan protocols [11]. This optimization problem shall be evaluated in this work. Hence, our focus is on contrast-enhanced imaging. The optimal threshold positions for other applications, e.g., virtual non-calcium images or imaging of potential other contrast agents [12, 13], might differ given the spectral properties of the materials of interest. In particular, an experimental CT system equipped with a prototype photon-counting detector is used in all studies presented here. This detector has already proven to allow for dual- and multi-energy decomposition of conventional and potential novel contrast agents, to provide iodine quantification with high accuracy, and to allow for the computation of VMIs, among others, and several contrast-enhanced patient studies have been published [14,15,16,17,18,19,20]. We here evaluate the quality of derived iodine maps as a function of threshold, patient size, and tube voltage in a phantom study. Since the ionizing nature of x-rays prohibits multiple examinations of patients, the results will be verified in a post-mortem CT angiography study.
making every photon count pdf download
Left: Experimental photon-counting CT and the used semi-anthropomorphic phantom. Right: Semi-anthropomorphic liver phantom used in the experiments. The numbers indicate the iodine concentration in the vials measured in mg/mL. Zero denotes pure water
Reference measurements were conducted using an energy-integrating dual-source dual-energy CT system (Definition Flash) using tube voltage pairs of 80 kV/Sn140 kV and 100 kV/Sn140 kV with Sn denoting an additional 0.4-mm tin prefilter to improve the spectral separation. The tube currents were adjusted to match the radiation dose level used at the photon-counting CT. The FOM of the DSCT is 500 mm and 330 mm for the first and second imaging chains, respectively. To match the pixel size of reconstructions obtained using the PC detector, reconstructions of the DSCT cover a FOV of 275 mm. A summary of all relevant acquisition and reconstruction parameters can be found in Table 1.
Figure 2 shows R for all available tube voltages obtained in the insert containing 15 mg I/mL. Colored, solid lines in this figure represent different phantom sizes, i.e., the S-phantom (green), the M-phantom (blue), and the L-phantom (red). For each tube voltage, the threshold was varied in steps of 5 keV between 50 keV and the maximum possible value for the respective tube voltage. Black dashed-dotted lines show R measured in an energy-integrating, dual-energy CT and represent the average overall phantom sizes. In the case of the photon-counting detector, R steadily increases with an increasing threshold. E.g., in the case of 140 kV, R is about 1.5 for a threshold of 50 keV and rises to about 2.5 at a threshold of 90 keV. This holds for all investigated tube voltages. At any given threshold, R increases with increasing tube voltage. E.g., for a threshold of 60 keV, a tube voltage of 80 kV results in a R of about 1.4, and in 1.6 using 100 kV, 1.65 for 120 kV and finally in 1.76 using 140 kV. The reference DECT system achieves relative contrast media ratios of about 2.8 for 80 kV/Sn140 kV and about 2.2 for 100 kV/Sn140 kV which is consistent with values already published in the literature [23]. Hence, the reference system generally shows a higher relative contrast media ratio compared to the investigated photon-counting detector. Only higher thresholds and tube voltages of 120 kV and 140 kV, respectively, exhibit similar values to the 100 kV/Sn140 kV reference protocol.
Figure 3 shows the CNRD (see Eq. 7, colored solid lines) for all available tube voltages using thresholds varied in 5-keV steps similar to Fig. 2. The line colors encode the phantom sizes. Black solid lines show the CNRD obtained using the reference DECT system using the S-phantom. Furthermore, the figure shows RelNoise (see Eq. 8, dashed lines). In general, CNRD decreases with increasing phantom size and increasing intersection lengths, due to the increase in image noise. All CNRD plots obtained using the investigated photon-counting detector show distinct maxima. In the case of the S-phantom, the optimal CNRD using a tube voltage of 80 kV is found as 0.71 mGy-1/2 at a threshold of 60 keV. For 100 kV, the optimal CNRD of 0.84 mGy-1/2 is achieved at a threshold of 65 keV and tube voltages of 120 kV and 140 kV show maxima at 70 keV with CNRDs of 0.98 mGy-1/2 and 1.04 mGy-1/2, respectively. While CNRD decreases with increasing phantom size, the position of the optimal threshold remains unchanged. This holds for all investigated tube voltages and all phantom sizes. The noise quotient RelNoise illustrates that unlike conventional DECT, maximum CNRD is not found for balanced background noise levels between bins. In particular, maximum CNRD is found for a noise quotient of 0.63 in the case of 80 kV, 0.72 for 100 kV, 0.73 for 120 kV, and 0.74 for 140 kV, respectively. The noise quotient does not show a dependence on phantom size. As indicated in Fig. 2, the used reference system achieves higher CNRDs of 1.29 mGy-1/2 and 1.99 mGy-1/2 for the protocols 80 kV/Sn140 kV and 100 kV/Sn140 kV. Previous studies illustrated that the root-mean-square errors between ground truth iodine concentrations and the ones measured in material maps are similar between the reference DECT system and the experimental system equipped with a photon-counting detector [11].
The results presented in Fig. 3 are further visualized in Fig. 4. The figure shows iodine material maps obtained of the S-phantom for all available tube voltages acquired at the optimal threshold and acquired at the same dose level. It is evident that acquisitions and material decomposition with 80 kV result in the highest noise among all acquisitions. The lowest noise and highest CNRD for the photon-counting detector are found at 140 kV and using a threshold of 70 keV.
Our study shows that the threshold of photon-counting detectors should be set according to the used tube voltage to ensure that image quality in material maps of iodine in terms of noise and CNRD is maximized. Since the investigated scan mode of the photon-counting detector always acquires two energy bins simultaneously, this also applies to standard examinations that are usually not intended as dual-energy examinations, but might be subject to a material decomposition retrospectively in case of accidental findings. Figure 2 illustrates that the relative contrast media ratio, a measure for the spectral separation between acquired bin data, is a function of the used threshold and increases with an increase thereof. Hence, if the noise in the low and high energy bin was constant, the best image quality would be achieved at the highest available threshold. However, since a change in threshold also affects the detected bin spectra and image noise, iodine CNRD shows distinct maxima as a function of tube voltage. This was illustrated using phantom measurements and in a post-mortem CT angiography study. Unlike in conventional dual-energy CT, the best image quality in the iodine maps was not achieved for a balanced image noise between bins, but for noise ratios between 0.6 and 0.8 (see Fig. 3). A comparison to iodine maps obtained using a conventional, energy-integrating dual-source dual-energy system revealed that the relative contrast media ratio of the investigated photon-counting detector is similar to the one observed in this system. However, given that the threshold cannot be changed without changing image noise and relative contrast media ratio, the image quality of the investigated photon-counting detector is slightly inferior to the reference system. This is consistent with the observations in the literature [29]. Other authors proposed the usage of dual-source photon-counting systems equipped with two photon-counting detectors to overcome this fact [30]. However, such systems do not yet exist and the cost-efficient future system might only be equipped with a single detector. The experiments presented herein assume a detector with two distinct energy bins. Other scan modes or detectors might provide even more simultaneously acquired energy bins. The used detector for example also provides a mode that acquires four bins simultaneously. However, R is not a suitable metric to describe the spectral separation between more than two energy bins. The design of suitable metrics for systems with more than two energy bins and the actual number of bins required to maximize e.g. the CNRD in iodine maps is a topic of ongoing research [31]. Photon-counting detectors not only allow for the simultaneous acquisition of multiple energy bins but also provide a variety of other benefits. Popular examples include but are not limited to a reduction in radiation and contrast media dose [32,33,34]. Hence, our experiments provide guidelines for threshold settings maximizing image quality in iodine maps, even if a dual-energy decomposition is only performed retrospectively. 2ff7e9595c
Comments