In 2021, photon-counting detector (PCD) computed tomography (CT) was introduced into clinical practice. Photon-counting detector CT quantifies the energy of each incident photon, which enables energy-selective CT acquisition with a single tube voltage and a single detector.1 In PCD-CT, virtual monoenergetic images (VMIs) have been established as the reference image set for clinical reading.2 In addition, a PCD-CT-specific “quantum iterative reconstruction” (QIR) algorithm with 4 strength levels has been implemented, which is specifically tailored to the data complexity and multienergy information of PCD-CT.3,4

Photon-counting detector CT has demonstrated its benefits in several areas of diagnostic imaging.5–14 Recently, QIR has been shown to improve image quality in the abdomen and lung,3,4 as well as the conspicuity of hypoattenuating focal liver lesions.3 However, the impact of PCD-CT combined with iterative reconstruction on diagnostic accuracy of specific diseases has not yet been assessed.

Former studies have shown the benefits of VMIs and iterative reconstruction for energy-integrating detector CT (EID-CT) with regard to image quality, lesion conspicuity, and diagnostic confidence for liver lesions,15–19 but some studies reported that these improvements in image quality with iterative reconstruction did not necessarily translate into increased overall diagnostic accuracy.20–22 In addition, the routine clinical use of VMIs combined with iterative reconstruction from EID-CT has been limited because of their altered image impression, particularly their blotchier noise texture at low-keV levels.23–25 Virtual monoenergetic images from PCD-CT combined with QIR might offer an improvement in image perception as QIR aims to replicate the noise texture of traditional filtered back projection. Moreover, based on the improved noise properties and the perfect anatomic acquisition registration, PCD-CT has the potential to improve focal liver lesion detection.

Our phantom study aimed to assess the image noise texture, diagnostic performance, and potential for radiation dose reduction of PCD-CT with QIR at different strength levels in detecting hypoattenuating and hyperattenuating focal liver lesions compared with EID-CT.

MATERIALS AND METHODS Image Acquisition and Reconstruction

An anthropomorphic abdominal phantom with uniform background and a liver insert including liver parenchyma and different lesions was used (QSA-453 and QSA-637; QRM, Moehrendorf, Germany). The liver insert had 90 Hounsfield units (HU) background attenuation at 120 kVp (Fig. 1). Five iodinated hypoattenuating and 4 hyperattenuating spherical and capsular focal liver lesions were arranged at different locations within the liver parenchyma. These lesions had the following characteristics: size of 5 to 10 mm and lesion-to-liver contrast difference of −30, −45, +30, and +90 HU at 120 kVp. A ring of fat-equivalent material was placed around the abdominal phantom to simulate a patient with a body mass index of approximately 25 kg × m−2 and axial dimensions of 350 × 250 mm.

FIGURE 1:

Phantom setup. Representative axial CT images of the anthropomorphic abdominal phantom with iodinated hypoattenuating and hyperattenuating liver lesions and a ring of fat-equivalent material at 5 mGy (left), 2.5 mGy (middle), and 1.25 mGy (right).

The phantom was imaged on 2 different CT systems: (a) in the dual-energy mode on a third-generation dual-source EID-CT (SOMATOM Force, Siemens Healthcare GmbH, Forchheim, Germany) at the tube voltage potential combination of 100 and 150 kVp with tin filtration and (b) in the single-source mode on a dual-source PCD-CT at 120 kVp (NAEOTOM Alpha, Siemens) equipped with 2 cadmium telluride PCDs. On each scanner, 3 different radiation dose levels with a volume CT dose index of 5, 2.5, and 1.25 mGy were applied by adapting the reference tube current of the automated tube current modulation system (CAREDose4D, Siemens). For each dose level, acquisition was repeated 10 times without changing the position of the phantom. Each dataset was reconstructed as VMIs at 60 keV for both scanners and as linear-blended images (LBIs) with a weighing ratio of 0.5 for EID-CT26 in the axial plane. For PCD-CT, VMIs were reconstructed with a novel iterative reconstruction algorithm (QIR, VA50, Siemens) at each of 4 strength levels (QIR 1–4) and without QIR (QIR-off). For EID-CT, VMI and LBI datasets were reconstructed using advanced modeled iterative reconstruction at a strength level of 3, which is our institutional clinical standard and has shown benefits in abdominal imaging.27,28 Images of the 10 repeated scans of each setup were combined to create a single dataset to compute accurate image quality metrics.29 Acquisition and image reconstruction parameters are summarized in Table 1. Volume CT dose index was used as a figure of merit for the radiation dose exposure.

TABLE 1 - Data Acquisition and Image Reconstruction Parameters PCD-CT EID-CT Radiation dose, CTDIvol in mGy 5/2.5/1.25 5/2.5/1.25 Data acquisition  Tube potential, kVp 120 100/Sn150  Quality reference, mAs 104/52/26 Tube A 117-tube B 59/tube A 57-tube B 29/tube A 25-tube B 13  Gantry revolution time, s 0.5 0.5  Beam collimation, mm 144 × 0.4 128 × 0.6  Pitch 0.8 0.8 Image reconstruction  Display field of view, cm 37 × 37 37 × 37  Section thickness, mm 3.0 3.0  Section overlap, mm 1.5 1.5  Kernel Br36 Br36  Algorithm QIR-off, 1–4 ADMIRE 3  Reconstruction VMIs at 60 keV VMIs at 60 keV and linear-blended images (weighing factor of 0.5)

In this study, QIR-off was used as a surrogate for weighted filtered back projection.

PCD-CT indicates photon-counting detector computed tomography; EID-CT, energy-integrating detector computed tomography; CTDIvol, volume CT dose index; QIR, quantum iterative reconstruction; ADMIRE, advanced modeled iterative reconstruction; VMIs, virtual monoenergetic images.

Image Noise Characteristics

Image noise was assessed by computing the noise power spectrum (NPS) in the background of the phantom according to recommendations from the International Commission on Radiation Units and Measurements reports 54 and 87. An in-house algorithm was programmed using the IGOR Pro 6 (Wavemetrics, Inc, Portland, OR) language. A total of 280 quadratic regions of interest of 110 × 110 pixels (area of approximately 60 cm2) were semiautomatically extracted to obtain 2-dimensional (2D) NPS as described in detail elsewhere.30 The 2D NPS were then radially averaged to yield 1D NPS.31 No detrending correction was applied to subtract large inhomogeneities.

Image noise magnitude was quantified by computing the NPS filtered by a human eye visual response function (NPSe), to account for a human observer's variable perception of noise at different spatial frequencies.32,33 Image noise magnitude values of the different algorithm strengths were used to calculate a relative noise magnitude ratio value for QIR-off and QIR 1–4 compared with QIR-off and EID-CT VMIs and LBIs. This metric described the impact of each algorithm's strength on the noise magnitude. Noise magnitude ratios close to 0 indicated substantial noise reduction, ratios close to 1 indicated no reduction, and ratios greater than 1 indicated an increase in noise magnitude. Noise magnitude ratio was calculated according to the following formula:

Noisemagnitude ratioi=∫NPSealgorithmidosek∫NPSealgorithmjdosek,

where i corresponds to QIR-off and QIR 1–4 and j corresponds to QIR-off, EID-CT VMIs, or EID-CT LBIs. k corresponds to the radiation doses of 5, 2.5, or 1.25 mGy.

Image noise texture was compared using the root-mean-square deviation (RMSD) of the normalized NPSe (nNPSe) curves31 (normalization by its integral across all frequencies) for QIR-off and each QIR strength level in comparison with QIR-off, EID-CT VMIs, or EID-CT LBIs according to the following formula:

RMSD=∑0∞xr−yr,t2N

where xr represents the nNPSe value at the frequency r for QIR-off, EID-CT VMIs, or EID-CT LBIs and yr,t represents the nNPSe value at the frequency r for images of the PCD-CT acquired at the dose levels (t = 5, 2.5, or 1.25 mGy) and reconstructed using QIR-off and QIR 1–4. N represents the total number of sample values. For an equivalent noise texture, curves have a similar shape, in which case the RMSD value is close to 0. The peak frequency difference (PFD) was also calculated with respect to QIR-off, EID-CT VMIs, or EID-CT LBIs. A positive PFD implies a sharper noise texture, whereas a negative PFD indicates a blotchier noise texture.

Lesion Detectability

Lesion detectability was assessed using a channelized Hotelling model observer with 10 dense differences of Gaussian channel.34 For this study, 150 signal-absent images and 50 or 40 signal-present images for hypoattenuating or hyperattenuating lesions, respectively, were used to compute an average area under the receiver operating curve (AUC). The AUC was used as figure of merit for lesion detectability. The internal noise was defined as the diagonal matrix multiplied by a proportional factor added to the covariance matrix to emulate the results found in humans.35 The covariance matrix was calculated from 150 signal-absent images. The 95% confidence interval estimators of channelized Hotelling model observer performance were calculated using the method developed by Wunderlich et al.36 A monotonic function was used to link signal-to-noise ratio (SNR) and AUC37:

SNR=2Φ−1AUC,

where Φ=∫−∞xφydy is the cumulative Gaussian function and φ is a Gaussian function with φ=12πe−x22.

Potential for Radiation Dose Reduction

The potential for radiation dose reduction without impairment in lesion detection was computed for the QIR strength level with the highest AUC compared with EID-CT VMIs and LBIs for each of the 3 radiation doses. By fitting the curve between AUC and radiation dose, we evaluated the radiation dose for which PCD-CT would achieve the same AUC as EID-CT applying the steps described in a previous study.38

RESULTS Image Noise Magnitude and Texture

The NPSe curves for all image reconstructions are displayed in Figure 2a–c. Noise magnitude ratio, RMSD of the nNPSe, and PFD shifts of the NPSe for the different strength levels of QIR in comparison with QIR-off, EID-CT VMIs, and EID-CT LBIs are summarized in Table 2.

FIGURE 2:

Image noise magnitude and texture. Noise power spectrum curves filtered by a human visual response function (NPSe) for the different scanner and reconstruction combinations at 5 mGy (a), 2.5 mGy (b), and 1.25 mGy (c). Vertical lines indicate the noise peak frequency. Note the pronounced image noise reduction when using QIR-4 at low radiation doses.

TABLE 2 - Noise Power Spectrum Comparison Versus QIR-Off Versus EID-CT VMIs Versus EID-CT LBIs Noise Magnitude ratio RMSD, HU2 mm2 PFD, mm−1 Noise magnitude ratio RMSD, HU2 mm2 PFD, mm−1 Noise magnitude ratio RMSD, HU2 mm2 PFD, mm−1 5 mGy  QIR-off / / / 1.349 0.082 0.020 2.012 0.019 0.010  QIR-1 0.775 0.011 −0.002 1.045 0.071 0.018 1.559 0.009 0.008  QIR-2 0.663 0.022 −0.004 0.894 0.060 0.016 1.333 0.007 0.006  QIR-3 0.551 0.037 −0.010 0.743 0.045 0.010 1.108 0.020 0.000  QIR-4 0.440 0.072 −0.020 0.593 0.012 0.000 0.885 0.055 −0.010 2.5 mGy  QIR-off / / / 1.197 0.048 0.024 1.798 0.008 0.008  QIR-1 0.782 0.012 0.000 0.937 0.036 0.024 1.407 0.017 0.008  QIR-2 0.673 0.021 −0.012 0.806 0.028 0.012 1.210 0.025 −0.004  QIR-3 0.564 0.032 −0.014 0.676 0.017 0.010 1.014 0.036 −0.006  QIR-4 0.457 0.052 −0.024 0.547 0.007 0.000 0.821 0.056 −0.016 1.25 mGy  QIR-off / / / 0.854 0.019 −0.002 1.396 0.034 −0.008  QIR-1 0.791 0.012 −0.012 0.675 0.022 −0.014 1.104 0.045 −0.020  QIR-2 0.686 0.022 −0.012 0.586 0.028 −0.014 0.957 0.054 −0.020  QIR-3 0.582 0.038 −0.014 0.497 0.040 −0.016 0.812 0.068 −0.022  QIR-4 0.479 0.058 −0.016 0.409 0.057 −0.018 0.669 0.087 −0.024

Comparison of the noise power spectrum filtered by human visual response function (NPSe) parameters among the different PCD-CT QIR strength levels versus QIR-off, EID-CT VMIs, and EID-CT LBIs. Noise magnitude ratios close to 0 indicate substantial reductions and noise magnitude ratios close to 1 indicate no noise reduction. Image noise texture is preserved if RMSD and PFD are close to 0. A negative PFD implies that the image noise texture is blotchier.

QIR indicates quantum iterative reconstruction; EID-CT, energy-integrating detector computed tomography; VMIs, virtual monoenergetic images; PCD-CT, photon-counting detector computed tomography; LBIs, linear-blended images; RMSD, root-mean-square deviation; PFD, peak frequency deviation.

Compared with QIR-off, increasing QIR level substantially decreased noise magnitude at all 3 dose levels (noise magnitude ratio from 0.791 for QIR-1 at 1.25 mGy to 0.440 for QIR-4 at 5 mGy). Peak frequency difference decreased and RMSD increased slightly with increasing QIR level compared with QIR-off (eg, PFD −0.024 mm−1 and RMSD 0.052 HU2 mm2 for QIR-4 at 2.5 mGy).

Compared with EID-CT VMIs, noise magnitude ratio was lower for all QIR levels at all 3 radiation doses except for QIR-1 at 5 mGy (ratio of 1.045). Peak spatial frequency was higher or similar for all QIR levels compared with EID-CT VMIs at 5 and 2.5 mGy (eg, PFD range from 0.0 mm−1 for QIR-4 at 2.5 mGy to 0.024 mm−1 for QIR-1 at 2.5 mGy), whereas peak spatial frequency was slightly lower at 1.25 mGy (PFD of −0.018 mm−1 for QIR-4). The RMSD decreased at higher QIR levels for 5 and 2.5 mGy, whereas it increased at higher QIR levels for 1.25 mGy.

Compared with EID-CT LBIs, noise magnitude ratio was lower for QIR-4 at 5 and 2.5 mGy and for QIR 2–4 at 1.25 mGy (lowest ratio of 0.669 for QIR-4 at 1.25 mGy). Highest levels of QIR decreased PFD slightly compared with EID-CT LBIs. This decrease was largest at the lowest radiation dose (PFD −0.024 mm−1 for QIR-4 at 1.25 mGy). The RMSD increased at higher QIR levels at all 3 radiation doses (RMSD 0.087 HU2 mm2 for QIR-4 at 1.25 mGy).

Lesion Detectability Hypoattenuating Lesions

Figure 3 and Table 3 show the AUC for detecting hypoattenuating focal liver lesions as a function of radiation dose for all reconstructions. Corresponding images of the lesions are shown in Figure 4.

FIGURE 3:

Diagnostic accuracy for hypoattenuating focal liver lesions. AUC for hypoattenuating focal liver lesions as a function of radiation dose, scanner, and reconstruction algorithm. Note the increase in detection at higher strength levels of QIR and the higher detection of high QIR levels as compared with EID-CT at low radiation doses.

TABLE 3 - Lesion Detection for Hypoattenuating Lesions AUC at 5 mGy AUC at 2.5 mGy AUC at 1.25 mGy EID-CT VMIs 0.974 ± 0.007 0.883 ± 0.016 0.726 ± 0.016 EID-CT LBIs 0.995 ± 0.002 0.928 ± 0.014 0.854 ± 0.017 PCD-CT QIR-off 0.958 ± 0.010* 0.910 ± 0.012 0.817 ± 0.015† PCD-CT QIR-1 0.974 ± 0.008* 0.936 ± 0.010† 0.847 ± 0.014† PCD-CT QIR-2 0.981 ± 0.006 0.949 ± 0.009† 0.863 ± 0.013† PCD-CT QIR-3 0.988 ± 0.004 0.962 ± 0.007*,† 0.881 ± 0.014† PCD-CT QIR-4 0.992 ± 0.003† 0.975 ± 0.006*,† 0.900 ± 0.012*,†

AUC values for the detection of hypoattenuating focal liver lesions as a function of the scanner and algorithm. Data are mean ± 95% confidence intervals.

AUC indicates area under the receiver operating curve; EID-CT, energy-integrating detector computed tomography; VMIs, virtual monoenergetic images; PCD-CT, photon-counting detector computed tomography; LBIs, linear-blended images; QIR, quantum iterative reconstruction.

*Statistical significance between PCD-CT and EID-CT LBIs (P < 0.05).

†Statistical significance between PCD-CT and EID-CT VMIs (P < 0.05).

FIGURE 4:

Regions of interest of each hypoattenuating focal liver lesion as a function of radiation dose, scanner, and algorithm. AUC values are listed. Images were obtained by averaging 10 repeated scans. Note the subjective decrease in lesion conspicuity with EID-CT VMIs, particularly at low radiation dose (bottom row).

For both scanners, AUC increased with increasing radiation dose. Overall, PCD-CT with QIR showed higher AUC compared with EID-CT VMIs in all conditions except for QIR-off at 5 mGy. The AUC was highest for EID-CT LBIs (0.995) and QIR-4 (0.992) at 5 mGy. Compared with EID-CT LBIs, AUC was higher for QIR 1–4 at 2.5 mGy and for QIR 2–4 at 1.25 mGy (eg, AUC of 0.900 for QIR-4 compared with 0.854 for EID-CT LBIs at 1.25 mGy).

Hyperattenuating Lesions

Figure 5 and Table 4 show the accuracy for detecting hyperattenuating focal liver lesions as a function of radiation dose for all reconstructions. Corresponding images of the lesions are shown in Figure 6.

FIGURE 5:

Diagnostic accuracy for hyperattenuating focal liver lesions. AUC for hyperattenuating focal liver lesions as a function of radiation dose, scanner, and reconstruction algorithm. Similar to hypoattenuating lesions, the detection increased at higher strength levels of QIR and increased compared with EID-CT at low radiation doses.

TABLE 4 - Lesion Detection for Hyperattenuating Lesions AUC at 5 mGy AUC at 2.5 mGy AUC at 1.25 mGy EID-CT VMIs 0.980 ± 0.006 0.904 ± 0.020 0.677 ± 0.023 EID-CT LBIs 0.997 ± 0.002 0.947 ± 0.011 0.859 ± 0.020 PCD-CT QIR-off 0.921 ± 0.022* 0.859 ± 0.025† 0.753 ± 0.027*,† PCD-CT QIR-1 0.983 ± 0.007 0.950 ± 0.011* 0.856 ± 0.016* PCD-CT QIR-2 0.988 ± 0.005 0.963 ± 0.009* 0.875 ± 0.016* PCD-CT QIR-3 0.993 ± 0.003 0.975 ± 0.007*,† 0.895 ± 0.014* PCD-CT QIR-4 0.997 ± 0.001* 0.985 ± 0.005*,† 0.917 ± 0.012*,†

AUC values for the detection of hyperattenuating focal liver lesions as a function of the scanner and algorithm. Data are mean ± 95% confidence intervals.

AUC indicates area under the receiver operating curve; EID-CT, energy-integrating detector computed tomography; VMIs, virtual monoenergetic images; PCD-CT, photon-counting detector computed tomography; LBIs, linear-blended images; QIR, quantum iterative reconstruction.

*Statistical significance between PCD-CT and EID-CT VMIs (P < 0.05).

†Statistical significance between PCD-CT and EID-CT LBIs (P < 0.05).

FIGURE 6:

Regions of interest of each hyperattenuating focal liver lesion as a function of radiation dose, scanner, and algorithm. AUC values are listed. Note the subjective decrease in lesion conspicuity with EID-CT VMIs, particularly at low radiation dose (bottom row).

For both scanners, AUC increased with increasing radiation dose. Overall, PCD-CT showed higher AUC compared with EID-CT VMIs in all conditions except for QIR-off at 5 and 2.5 mGy. The AUC was highest for EID-CT LBIs and QIR-4 at 5 mGy (both, AUC of 0.997). Compared with EID-CT linear-blended, AUC was higher for QIR 1–4 at 2.5 mGy and for QIR 2–4 at 1.25 mGy (eg, AUC of 0.917 for QIR-4 compared with 0.859 for EID-CT LBIs at 1.25 mGy).

Potential for Radiation Dose Reduction

Table 5 summarizes the potential for radiation dose reduction obtained with QIR-4 compared with EID-CT VMIs and LBIs. Mean estimated radiation dose reduction ranged between 44% and 54% compared with VMIs. Compared with EID-CT LBIs, there was no potential for radiation dose reduction at 5 mGy, whereas the radiation dose reduction potential of QIR-4 was 19% at 1.25 mGy and 39% at 2.5 mGy.

TABLE 5 - Potential for Radiation Dose Reduction Radiation Dose Dose Reduction Potential VMIs LBIs 5 mGy −54%* (−64% to −45%) / 2.5 mGy −53%* (−61% to −45%) −39%* (−49% to −29%) 1.25 mGy −44%* (−61% to −28%) −19%* (−35% to −4%)

Radiation dose reduction potential of PCD-CT with QIR-4 as compared with EID-CT VMIs and LBIs. Confidence intervals are provided in parentheses.

VMIs indicates virtual monoenergetic images; LBIs, linear-blended images; PCD-CT, photon-counting detector computed tomography; QIR, quantum iterative reconstruction; EID-CT, energy-integrating detector computed tomography.

*Statistically significant (P < 0.05).

DISCUSSION

In this phantom study, we compared the accuracy for detecting hypoattenuating and hyperattenuating focal liver lesions between PCD-CT at different strength levels of QIR and third-generation dual-source EID-CT with advanced modeled iterative reconstruction at a strength level of 3. Virtual monoenergetic images at 60 keV were used for both scanners and LBIs for EID-CT. In addition, we assessed the image noise characteristics of PCD-CT combined with QIR and its potential for radiation dose reduction compared with EID-CT.

Our results indicate an improvement in the detection of hypoattenuating and hyperattenuating focal liver lesions using PCD-CT with QIR at all available strength levels compared with EID-CT VMIs. This improvement was strongest at very low radiation dose and high strength levels of QIR (up to 24% for QIR-4 at 1.25 mGy). Compared with EID-CT LBIs, detection was higher for QIR 1–4 at 2.5 mGy and for QIR 2–4 at 1.25 mGy. Energy-integrating detector CT LBIs demonstrated high accuracy at the highest dose of 5 mGy, which could only be achieved using QIR-4. Radiation dose reduction potential of PCD-CT with QIR-4 was 54% at 5 mGy compared with VMIs and 39% at 2.5 mGy compared with LBIs.

This is the first study to compare focal liver lesion detection between PCD-CT and EID-CT. We focused our investigation on small focal liver lesions (5–10 mm) and relatively low radiation doses (1.25–5 mGy) to improve the discrimination of detection differences and to challenge these 2 state-of-the-art CT systems with difficult image conditions. Interestingly, the difference in detection between PCD-CT and EID-CT was highest at the lowest radiation dose of 1.25 mGy, with substantially lower detection for EID-CT VMIs (eg, detection accuracy of 92% for QIR-4 vs 68% for EID-CT VMIs for hyperattenuating lesions and 90% for QIR-4 vs 73% for EID-CT VMIs for hypoattenuating lesions). We attribute this difference to the reduced electronic noise in PCD-CT, which is particularly beneficial in very low radiation dose conditions.

Former studies on EID-CT reported that improvements in image quality with iterative reconstruction do not necessarily translate into an improvement in diagnostic accuracy.20–22 Some studies even found that iterative reconstruction did not preserve detectability when reducing radiation dose.39–41 This finding was confirmed in our study with a substantial decrease in detectability for EID-CT and PCD-CT at the lowest radiation dose. However, the overall detectability of PCD-CT with QIR could still be considered good to high even at the lowest dose level (minimum AUC of 0.847 for QIR-1 and of 0.900 for QIR-4). Regarding VMIs, similar detection between PCD-CT and EID-CT was found only if QIR was deactivated (for QIR-off at the highest radiation dose of 5 mGy for hypoattenuating lesions and QIR-off at radiation doses of 5 and 2.5 mGy for hyperattenuating lesions). In addition, PCD-CT with a QIR level of 2 or higher led to higher detection accuracy compared with EID-CT LBIs at low radiation doses.

Several former studies used simple objective image quality metrics, for example, contrast-to-noise ratio as a surrogate metric for lesion detectability.39,42 However, these metrics may provide erroneous results as they do not take into account lesion size, lesion contrast (ie, the metric is not task based), or noise texture.37,43 Using a human model observer such as the channelized Hotelling model observer as a surrogate for image quality is a potential solution to overcome the limitation of first-order statistical metrics (ie, signal-to-noise ratio, contrast-to-noise ratio, standard deviation). These model observers already demonstrated a strong correlation with human observer performance.44,45

A previous phantom study by Mileto et al22 compared the detectability of hypoattenuating lesions less than 1 cm at different radiation doses between filtered back projection and iterative reconstruction with EID-CT. The authors found a limited radiation optimization potential in the detectability of small low-contrast hypoattenuating focal lesions using iterative reconstruction. In our study, in which small lesions were also used, the detection increased from 0.817 to 0.900 using QIR-4 compared with QIR-off at 1.25 mGy and from 0.921 to 0.997 at 5 mGy.

Based on our estimations, QIR-4 could allow a mean d