This study used a paediatric chest phantom (PH- 1 C) measuring 32 cm × 17 cm × 38 cm with a lung density insert (Kyoto Kagaku, Kyoto, Japan). The PH- 1 C phantom is designed to replicate key anatomical structures of the paediatric chest, including the lung, heart, trachea, and surrounding soft tissues. It features a lung density insert with tissue-equivalent materials that mimic the radiographic properties of a 5-year-old child’s lungs providing realistic attenuation and scatter characteristics for CT imaging [16].

The phantom was examined on a first-generation dual-source PCD-CT system (NAEOTOM Alpha, software version VB10; Siemens Healthineers, Forchheim, Germany) equipped with two cadmium telluride detectors. Scans were performed in the high-pitch mode with a pitch factor of 3.2, a gantry rotation time 0.25 s, and at a tube voltage of 100 kV, using tin pre-filtration. The scans were conducted twice, first in the standard mode with a detector collimation of 144 × 0.4 mm, and then in the ultra-high resolution mode with a detector collimation of 120 × 0.2 mm (Table 1). Exposure times during scan acquisition were 0.42 s for the standard mode and 0.84 s for the ultra-high resolution mode. Radiation doses were varied by adjusting the image quality (IQ) level to achieve a volume CT dose index (CTDIvol) of 0.45 mGy, 0.30 mGy, 0.15 mGy, and 0.07 mGy, respectively. Additionally, to explore the scanner’s technical limits, the tube current was fixed at the lowest possible value to achieve the minimum radiation dose, resulting in a CTDIvol of 0.01 mGy.

All scans were reconstructed as polychromatic images with a single energy threshold set at 20 keV, given that dual-source ultra-high-resolution scans do not yield energy-resolved data. Slice thickness was set to 1 mm and increment to 0.7 mm, applying the sharp lung kernel Bl60 and using a matrix size of 512 pixels × 512 pixels. All scans were reconstructed without quantum iterative reconstruction (quantum iterative reconstruction off), and with quantum iterative reconstruction at strengths 2 and 4. Figures 1 and 2 depict representative images of the paediatric phantom scans performed at different dose levels and reconstructed with different quantum iterative reconstruction strengths.

Representative axial images of scans acquired in both standard and ultra-high resolution modes are presented. A rectangle highlights the magnified region within the phantom (a). The magnified images compare scans from the standard mode (b-f) and the ultra-high resolution mode (g-k) across varying radiation doses. All reconstructions were performed using quantum iterative reconstruction at strength 4. Note the worsening visibility of small peripheral structures, such as the three subpleural nodules marked with a circle, with a lower radiation dose. CT computed tomography, CTDIvol volume CT dose index

Representative axial images of scans acquired in the standard mode (a-d) and the ultra-high resolution mode (e–h) at the two lowest radiation doses (volume CT dose index, CTDIvol) of 0.07 mGy (a, b, e, f) and 0.01 mGy (c, d, g, h). Scans were reconstructed both without using quantum iterative reconstruction (QIR off) (a, c, e, g) and with QIR at strength 4 (b, d, f, h). Note the considerably reduced image noise and augmented visualization of small peripheral structures in reconstructions with QIR 4 (b, d, f, h) compared to those without QIR (a, c, e, g). CT computed tomography, CTDIvol volume CT dose index, QIR quantum iterative reconstruction

Two board-certified paediatric radiologists, one with 32 years (reader 1, C.K.) and the other with 11 years (reader 2, M.Z.) of experience in paediatric lung CT, independently evaluated the images. Both readers were blinded to the scanning parameters and radiation doses. Reader 2 (M.Z.) performed the subjective image evaluation twice, with a time interval of 10 weeks. Lung images were assessed on axial image sections with fixed window width of 1500 HU and centre of − 500 HU. Six criteria were investigated: overall image quality, image noise and reticular pattern, presence of streak artefacts, pleural sharpness, sharpness of lung structures, and visibility and detection of small lung structures.

Overall image quality was assessed using a 4-point discrete visual scale, where 1 indicated unacceptable quality (non-diagnostic images), 2 indicated limited quality (sufficient only for restricted clinical interpretation due to high noise), 3 indicated adequate quality (small peripheral structures remained visible), and 4 indicated quality exceeding diagnostic requirements (minimal or no noise). Supplementary material 1 depicts examples with an overall image quality exceeding diagnostic requirement, and an unacceptable, non-diagnostic overall image quality.

Image noise and reticular patterns, as well as streak artefacts, were rated using a 4-point discrete visual scale, where 1 was severe, 2 was moderate, 3 was mild, and 4 was absent. Pleural sharpness represents the clarity of the peripheral lung-to-soft-tissue interface, as well as visibility and detection of small lung structures, where three predefined subpleural nodules in the left lower lobe were evaluated using a 4-point discrete visual scale. Here, 1 corresponded to unacceptable, 2 corresponded to significantly reduced, 3 corresponded to mildly reduced, and 4 corresponded to excellent sharpness or visibility.

Global noise index and global signal-to-noise ratio index were computed as previously described [10, 17]. In brief, after lung segmentation and extraction of all lung voxels, noise maps were generated slice-wise. The standard deviation of pixel values in the immediate vicinity of a given pixel was calculated to create these noise maps. These noise maps allow for the creation of a histogram representing the noise distribution. Subsequently, the mode value, i.e. the most common noise value, is extracted from the histogram and utilized as the global noise index [18].

To compute the global signal-to-noise ratio index, signal-to-noise ratio maps for the entire lung were first generated. These signal-to-noise maps were obtained by dividing the attenuation by the noise within the target region (i.e. a single pixel and its immediate surrounding). A histogram of the signal-to-noise distribution across the lungs then served to extract the mode value, which was utilized as the global signal-to-noise ratio index [10].

In addition, the mean attenuation of all lung voxels was measured. All steps were performed applying a fully automated computational pipeline in the R programming language.

Size-specific dose estimates in milligray (mGy) and effective doses in millisievert (mSv) were calculated using the effective chest diameter combined with size-dependent correction factors obtained from the American Association of Physicists in Medicine [19] and Romanyukha et al. [20]. The effective chest diameter was calculated applying the following formula,

$$\text= \sqrt\bullet \text}$$

Effective doses were determined by multiplying the dose-length product (DLP) by a diameter-specific conversion factor of 0.052 [19, 20].

Variables are presented as mean ± standard deviation when normally distributed and as median and interquartile range when non-normally distributed. Categorical variables are reported as counts and percentages. Inter-observer reliability between both readers and intra-observer reliability for reader 2 were evaluated using two-way mixed, average measures intraclass correlation coefficients (ICC [3, k]). ICC values below 0.5 indicate a poor, 0.5 to 0.75 a moderate, 0.75 to 0.9 a good, and higher than 0.9 an excellent reliability [21]. Differences of quantitative metrics were investigated by Wilcoxon signed-rank tests. P-values were adjusted with the Benjamini–Hochberg procedure for multiple comparisons. A two-tailed P-value less than 0.05 was considered to indicate statistical significance. Analyses were performed using SPSS Statistics version 26 (IBM Corp., Armonk, NY).