Dose-Optimized Image Acquisition Parameters for Neonatal Chest Radiography: A Phantom Study Comparing Computed Radiography and Wireless Digital Radiography Needle Detectors

Abstract

Purpose

To determine dose-optimized image acquisition parameters for good image quality (IQ) in neonatal chest radiography with a computed radiography (CR) CsBr needle detector vs. a wireless digital radiography (DR) CsI detector using different doses and filters.

Materials and Methods

Physical resolution of the two detectors in unprocessed imaging of a contrast-detail phantom was automatically evaluated. Post-processed chest radiographic imaging of a neonatal phantom was used for Visual Grading Analysis (VGA) by three radiology raters. Different kVp, mAs, and filter settings were used. The VGA score (VGAS) and dose area product (DAP) were used to determine image acquisition parameters and dose levels for good image quality. Pixel data from segments corresponding to visual grading characteristics (VGC) was used to calculate signal-to-noise ratio, contrast-to-noise ratio (CNR), and signal profile curves. These results were compared to the raters’ “ground truth” by Spearman’s correlation.

Results

The CR detector had the highest resolution in unprocessed imaging, although this was dependent on a tube voltage of 66 kVp (P < 0.001), and more so than the DR detector. The VGAS showed no significant difference between the CR needle and the DR CsI detectors at the same DAP, or when using standard pediatric filtering of 3.5 mm Al + 0.1 mm Cu (P > 0.05). A lung dose level of 0.017 mSv was needed for good IQ (effective dose (E): 0.010 mSv). This was achievable with different acquisition parameters. Out of 24 segments, only the CNR of bone-to-soft-tissue had a good Spearman’s correlation (ρ > 0.50) to raters’ VGAS (P < 0.0001), mostly due to problems with image registration.

Conclusion

The CR needle and DR CsI detectors have comparable IQ in neonatal chest radiography. In this study, an E of approximately 0.010 mSv was needed for good IQ.

Key Points

• CR needle and DR detectors have comparable image quality in neonatal chest radiography.

• CR needle technology has higher absolute raw image resolution, although this is voltage-dependent and more so than the DR detector.

• We propose ideal image acquisition parameters for neonatal chest radiography.

Citation Format

• Rama K, Esser M, Spogis J et al. Dose-Optimized Image Acquisition Parameters for Neonatal Chest Radiography: A Phantom Study Comparing Computed Radiography and Wireless Digital Radiography Needle Detectors. Rofo 2025; DOI 10.1055/a-2525-9430

Zusammenfassung

Ziel

Bestimmung dosisoptimierter Bildaufnahmeparametern für eine gute Bildqualität (IQ) in der neonatalen Thoraxradiografie mit einem Computerradiografie-CsBr-Nadeldetektor im Vergleich zu einem drahtlosen Digitalradiografie-CsI-Nadeldetektor unter Verwendung unterschiedlicher Dosiswerten und Filter.

Material und Methoden

Die physikalische Auflösung der beiden Detektoren bei nicht-prozessierten Aufnahmen eines Kontrast-Detail-Phantoms wurde automatisch bewertet. Postprozessierte Thorax-Röntgenaufnahmen eines Neugeborenen-Phantoms wurden von drei radiologischen Ratern für eine Visual Grading Analysis (VGA) verwendet. Es wurden unterschiedliche kVp-, mAs- und Filtereinstellungen verwendet. Der VGA-Score (VGAS) und das Dosisflächenprodukt (DAP) dienten der Bestimmung von Bildaufnahmeparametern und Dosiswerten für eine gute Bildqualität. Die Pixeldaten der Segmente, die den Visual Grading Characteristics (VGC) entsprechen, wurden zur Berechnung des Signal-Rausch-Verhältnisses, des Kontrast-Rausch-Verhältnisses (CNR) und der Signalprofilkurven verwendet. Diese Ergebnisse wurden mittels Spearman-Korrelation mit der „Ground Truth“ der Rater verglichen.

Ergebnisse

Der CR-Detektor hatte die höchste Auflösung bei nicht-prozessierten Aufnahmen, obwohl dies von einer Röhrenspannung von 66 kVp abhängig war (P < 0,001), und zwar stärker als der DR-Detektor. Der VGAS zeigte keinen signifikanten Unterschied zwischen den CR-Nadel und DR-CsI-Detektoren bei gleichem DAP oder bei Anwendung der pädiatrischen Standardfilterung von 3,5 mm Al + 0,1 mm Cu (P > 0,05). Für eine gute IQ war eine Lungendosis von 0,017 mSv erforderlich (Effektive Dosis (E): 0,010 mSv), die mit verschiedenen Aufnahmeparametern erreicht werden konnte. Von den 24 Segmenten wies nur das CNR des Knochen-zu-Weichteil-Segmentes eine gute Spearman-Korrelation (ρ > 0,50) zu den VGAS der Rater auf (P < 0,0001), hauptsächlich aufgrund unzureichender Bildregistrierung.

Schlussfolgerung

CR-Nadel- und DR-CsI-Detektoren haben eine vergleichbare IQ in der neonatalen Thoraxradiographie. In dieser Studie war eine E von circa 0,010 mSv für eine gute Bildqualität erforderlich.

Kernaussagen

• CR-Nadel- und DR-Detektoren haben eine vergleichbare Bildqualität in der neonatalen Thorax-Radiografie.

• Die CR-Nadeltechnologie hat eine höhere absolute Rohbildauflösung, dies ist allerdings spannungsabhängig, und zwar stärker als beim DR-Detektor.

• Wir schlagen ideale Bildaufnahmeparameter für die neonatale Thorax-Radiografie vor.

Keywords

Thoracic Radiography - Newborn Infant - Imaging Phantoms - Radiation Dosage - Radiation Protection

Introduction

Chest radiography is the most frequently performed diagnostic radiographic imaging method in the neonatal intensive care unit (NICU) [1] [2]. Thus, it represents the main source of ionizing radiation in hospitalized neonates. Currently, chest X-ray imaging is performed both by using computed radiography (CR) and digital radiography (DR), with DR increasingly replacing CR as it is more time-efficient [3]. A visual grading analysis (VGA) study by Smet et al. has shown no significant difference in image quality (IQ) between CR needle detectors and DR CsI detectors in post-processed neonatal phantom images [4]. However, the optimized image acquisition parameters still must be clarified to avoid inappropriate dose levels [5].

Båth and Månsson introduced the analysis of visual grading characteristics (VGC) [6]. This method provides a statistical framework to analyze the radiologists’ rating of the fulfillment of image quality criteria in an image. In this context, it is essential to know the detectors’ IQ without image post-processing, considering the existing difference in pixel width between the two technologies. Another point of interest is the possible automation of the evaluation using the signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and signal profile curves determined at the corresponding locations used for VGC. These values would allow comparison to the radiologists’ “ground truth”.

The purpose of this study was to compare the image quality of the DR CsI detector technology to the CR NIP detector technology in unprocessed images using a contrast-detail phantom, and in processed images using a neonatal phantom. We also aimed to identify image acquisition parameters and respective dose levels for optimal image acquisition in neonates when considering both IQ and the As Low As Reasonably Achievable (ALARA) principle.

Materials and methods

This study encompasses two parts, both involving the same detectors (an Agfa CR HD5.0: a CR needle CsBr detector, with a pixel width of 100 µm and a pixel matrix of 3408 × 4200; and an Agfa DR 14s: a DR needle CsI detector, with a pixel width of 148 µm and a pixel matrix of 2400 × 2880; Agfa HealthCare N.V., Mortsel, Belgium). The X-ray tube that was used was the Siemens Ysio MAX (Siemens Healthineers, Munich, Germany). A standard minimum filtering of 2.5mm Al was used, with the possibility of adding additional Cu filtering (0.1 mm, 0.2 mm, 0.3 mm) and additional aluminum (or aluminum equivalents). The filtering settings, as well as the voltage and current settings used, are detailed in [Table 1]. These settings are based on and are in accordance with the European Commission’s guidelines on quality criteria for radiographic imaging in pediatrics [7], as well as with the latest edition of Germany’s Medical Chamber guideline on quality assurance in diagnostic radiology [8] and the latest edition of Germany’s guideline on pediatric diagnostic imaging by the Commission for Radiation Protection [9], which are the guidelines that regulate this topic in Germany. Within this frame, we chose to use a relatively wide range of image acquisition parameters, even though some combinations (especially at the “edges” of the ranges of values used) would lead to clinically non-usable settings, e.g., high dose area product (DAP) values. Through this process, we avoided a pre-selection bias in our results.

Table 1 Image acquisition parameters used for the CDRAD 2.0 Phantom and the Gammex 610 neonatal chest phantom.

	CDRAD 2.0 Phantom	Gammex 610 Neonatal Chest Phantom
SID = source to image distance
SID [cm]	115	115
Focus	0.6	0.6
Total filter [mm]	2.5 Al | 3.5 Al + 0.1 Cu | 3.5 Al + 0.2 Cu	2.5 Al | 3.5 Al + 0.1 Cu
Peak voltage [kVp]	66 | 70	57 | 60 | 66 | 77
Tube current · exposure time [mAs]	0.50 | 1.00 | 1.50 | 2.00	0.50 | 0.63 | 0.80 | 1.00 | 1.25 | 1.60 | 2.00

Firstly, we measured the physical resolution capabilities of the detectors using the CDRAD 2.0 contrast-detail phantom (Artinis Medical Systems B.V., The Netherlands), which is a 265 mm × 265 mm × 10 mm acrylic plastic plate containing a 15 × 15 square matrix with holes of different diameters and depths ranging from 0.3 mm to 8.0 mm ± 0.03 mm ([Fig. 1]) [10] [11]. The settings used for image acquisition are listed in [Table 1]. For each setting, six images were acquired and then automatically evaluated without image post-processing by the Artinis CDRAD analyzer software (Artinis Medical Systems B.V., The Netherlands), with an alpha level of significance of 0.0001 and an a priori difference of the mean of 0. There were three missing mAs measurements for the computed radiography (CR) detector at 70 kVp. This automated evaluation provides an image quality figure inverse (IQFinv) value, which increases with higher resolution. It is calculated with the following equation [11],

Fig. 1 Contrast detail phantom CDRAD 2.0 (image courtesy of Artinis Medical Systems B.V., The Netherlands).

where C _i is the depth (i.e., contrast) of the visible hole in the column i, and D _i,th is the lowest diameter (threshold) correctly detected in the column i [10] [11].

For both voltages, i.e., 66 kVp and 70 kVp, the significant differences in IQFinv between the two detectors were determined by linear regression analysis.

Because the use of only a contrast-detail phantom would have limitations with respect to representing the clinical use of the detectors’ technology, this study’s second part involves imaging a commercially available neonatal chest phantom (Gammex 610 Neonatal Phantom, Sun Nuclear Corporation, Melbourne, Florida, USA). This chest phantom represents a neonate in the weight class of 1–2 kg. It contains a right lung simulating infant respiratory distress syndrome (IRDS) and a left lung simulating a pneumothorax. To these characteristics we added an endotracheal tube, a central venous catheter (CVC), a Silastic CVC, and a nasogastric tube ([Fig. 2]a). The images were acquired with different kVp, mAs, and filter settings, as listed in [Table 1]. A total of 112 anterior-posterior images for the neonatal phantom were obtained. All images were processed with Agfa Musica 3 post-processing, presented randomly, and rated via ViewDEX software [12] [13] by three radiology raters with different levels of experience (radiology resident, board-certified radiologist training as a pediatric radiologist, senior pediatric radiologist). The raters were blinded to all information regarding the image acquisition parameters and had a high IQ reference image with marked VGC regions. A three-megapixel Eizo RadiForce RX350 monitor (Eizo Corporation, Hakusan, Japan) was used to rate the images. The raters evaluated 13 criteria, including all European Commission’s guidelines criteria [7] and those applicable from the most recent guidelines of Germany’s Medical Chamber [8], using a five-point Likert scale ([Table 2]). To assess the intra-rater agreement, 10% of the images (11 out of 112) were presented twice to the raters. Considering that the ratings belong to an ordinal scale, the median of all 13 VGC ratings was used to calculate a visual grading analysis score (VGAS) for each image and rater [4] [6]. The VGAS of all criteria was used to calculate intra-class correlation coefficients (ICC) for intra- and inter-rater agreement.

Fig. 2 a Anterior-posterior radiography: Gammex 610 neonatal chest phantom with the addition of an endotracheal tube (1), a central venous catheter (CVC) (2), a Silastic CVC (3), and a nasogastric tube (4). b Anterior-posterior radiography: Colored areas and lines are the segments corresponding to the visual grading characteristics (listed in [Table 2]) used for automated extraction of objective pixel data after manual registration.

Table 2 Visual grading characteristics and rating scale.

Visual grading characteristics	Rating scale
IRDS = infant respiratory distress syndrome, CVC = central venous catheter
1) Trachea	A – insufficient
2) Big airways	B – limited
3) Small airways	C – sufficient
4) Pneumothorax (left lung)	D – good
5) Thoracic vertebra	E – optimal
6) Pericardium-lung border
7) Diaphragm
8) Costodiaphragmatic recess
9) IRDS (right lung)
10) Tip of the endotracheal tube
11) Tip of the Silastic-CVC
12) Tip of the CVC
13) General image quality

To evaluate the specific conditions of neonatal lung imaging, a lung score (VGAS_LUNGS) was calculated as a median of three VGCs: IRDS of the right lung, pneumothorax of the left lung, and small airways. A 3^rd-grade polynomial function of VGAS_LUNGS to DAP was calculated and was used to determine the corresponding DAP values at a rating of “good”, i.e., 4, and thus to determine the corresponding image acquisition parameters needed to obtain a clinically usable image quality without using unnecessary ionizing radiation.

Ordinal logistic regression analysis was calculated to determine the effects of the DAP, detector, and filter on VGAS of all criteria (in this context, it serves as a proxy for IQ), as described by Smet et al. [4]. Four images for each detector were missing DAP values and could not be included in the further analysis requiring this data. DAP values which allow for good image quality (i.e., rating of 4), extracted through the function of the curve of the lungs’ rating criteria (VGAS_LUNGS), the corresponding image acquisition parameters, and the phantom surface area of 100 cm², were used to calculate organ doses, the whole-body dose (WBD), and effective dose (E) values. This was done by using conversion tables available in the literature (the DAP value was divided by the surface area of 100 cm², resulting in the entrance dose. Multiplication of the entrance dose by the conversion factor resulted in a dose value, e.g. organ dose, WBD, etc. [14]. The reference book we used for the conversion factor [14] bases its definition of effective dose on the recommendations of the 2007 publication from the International Commission on Radiological Protection [15]. The rater-based evaluation was compared to objectified image analysis. To do this, manual rigid registration of the images was performed using ImageJ software (NIH, Bethesda, MD, USA) and 3D Slicer software [16]. This was done because the neonatal phantom images would not overlap with each other, which is necessary for segmentation and automated pixel information extraction. Non-rigid registration through algorithms would permit overlapping of the images by distorting them, thus altering the original information. To avoid this, rigid registration was chosen. Segmentation was then performed on a reference image at locations corresponding to the VGC ([Fig. 2]b), and the mean pixel value, standard deviation, and along the middle line signal profile curves were extracted from the segments via an in-house script using MATLAB software (The MathWorks, Inc., Natick, MA, USA). With these values, the SNR, CNR, and slope of the linear regression equation of signal profiles at the border between two structures ([Fig. 3]) were calculated. Spearman’s correlation coefficient compared these objective criteria to the raters’ VGAS (used as a reference for “ground truth”). The level for statistical significance was set at 0.05, and the software used for statistical analyses was IMB SPSS Statistics Version 27 (IBM Corporation, Armonk, NY, USA) and JMP Version 16 (SAS Institute, Inc., Cary, NC, USA).

Fig. 3 Example of visualization of the mean average of all signal profiles, i.e., of all registered neonatal phantom images, along the middle line of the central venous catheter (CVC) horizontal segment (see [Fig. 2]a (2) and [Fig. 2]b (green)). From pixel no. 16 to pixel no. 22, the signal profile intensity changes, as the underlying structure changes from vertebra/mediastinum to the CVC’s tip. For all signal profiles, at the border between different structures, the slope of the linear regression equation was calculated and used as a parameter for image sharpness, representing the rate of change of signal intensity.

Results

The highest IQFinv of all unprocessed recordings was found at 66 kVp for the CR needle CsBr detector (P < 0.001). The CR detector, however, has a more pronounced voltage dependency of IQ than the DR CsI detector ([Fig. 4]). Thus, at 70 kVp, the DR CsI detector has significantly higher IQFinv values (P < 0.001). However, it should be noted that the absolute IQFinv values of the DR CsI detector at 70 kVp are not as high as those of the CR detector at 66 kVp.

Fig. 4 Mean IQFinv results of the CDRAD 2.0 phantom. Higher values represent higher image quality (IQ). Band represents 95% confidence interval (CI). Three missing mAs measurements for the computed radiography (CR) detector at 70 kVp.

The ICC of VGAS (median of all 13 criteria) was high at 0.80 [95% confidence interval (CI) 0.68–0.87]. For the intra-rater agreement, it was also good or high depending on the rater: 0.82 [95% CI 0.35–0.95], 0.86 [95% CI 0.52–0.96], 0.63 [95% CI -0.43–0.90]. The results of the ordinal logistic regression analysis of VGAS as the dependent variable, DAP as the independent covariate, and filter and detector as independent factors are shown in [Table 3]. At the same DAP level, the VGAS between the two detectors had no significant difference. There was also no significant difference in VGAS between standard pediatric total filtering of 3.5 mm Al + 0.1 mm Cu instead of minimal filtering of 2.5 mm Al at the same DAP level.

Table 3 Results of the ordinal logistic regression analysis regarding the effects of dose area product (DAP), filter, and detector on the visual grading analysis score (VGAS, median of the 13 visual grading characteristics). Four images for each detector with missing DAP values not included.

OR = odds ratio, n/a = not available, DAP = dose area product, CR = computed radiography, DR = digital radiography
Model fitting information
Model	-2 Log Likelihood	Chi-Square	df	Sig.
Intercept only	557.876
Final	325.894	231.982	3	0
Parameter estimates	Estimate	Std. error	Wald	df	P-value	95% confidence interval		OR
						Lower bound	Upper bound
[rating = 1]	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a
[rating = 2]	–1.84	0.48	14.78	1	<0.001	–2.78	–0.90	0.16
[rating = 3]	1.16	0.28	17.75	1	<0.001	0.62	1.70	3.20
[rating = 4]	4.26	0.38	125.92	1	<0.001	3.51	5.00	70.60
[rating = 5]	0	.	.	0	.	.	.	.
DAP	10.93	1.08	102.21	1	<0.001	8.81	13.04	55603.42
[Filter = 2.5 mm Al]	–0.21	0.27	0.60	1	0.44	–0.75	0.32	0.81
[Filter = 3.5 mm Al + 0.1 mm Cu]	0	.	.	0	.	.	.	.
[Detector = CR needle CsBr]	0.11	0.24	0.23	1	0.63	–0.35	0.58	1.12
[Detector = DR needle CsI]	0	.	.	0	.	.	.	.

As expected, the DAP has a significant effect on IQ with an odds ratio (OR) of 55,603.42 (95% CI 6687.53–462,314.19, P < 0.001). [Fig. 5] shows the relationship between VGAS_LUNGS and DAP, differentiated by detector and filter. It is apparent that after a certain DAP level, the curve reaches a plateau, after which there is no relevant benefit for IQ as the dose of ionizing radiation increases. Based on the DAP determined at a rating of “good” for VGAS_LUNGS, [Table 4] displays the corresponding organ doses, effective dose (E), and whole-body dose (WBD). Using a standard pediatric total filter of 3.5 mm Al + 0.1 mm Cu, an E of 0.010 mSv is needed for good image quality with the DR CsI detector, which corresponds to a lung dose level of 0.016 mSv [14] [15]. Further organ doses are listed in [Table 4]. With CR needle CsBr, a slightly higher DAP level was required (2.6 mGy · cm²). In our case, it was possible to obtain this range of DAP (2.2–2.6 mGy · cm²) with four different settings (57 kVp & 1.60 mAs; 60 kVp & 1.60 mAs; 66 kVp & 1.25 mAs; 77 kVp & 0.80 mAs), among which 57 kVp & 1.60 mAs, and 77 kVp & 0.80 mAs had the lowest DAP at 2.2 mGy · cm² ([Table 4]).

Fig. 5 Relationship between visual grading analysis score of the lungs (VGASLUNGS) and dose area product (DAP). Image quality increases with DAP but reaches a plateau after which the increments of image quality require a much greater dose. See [Table 4] for DAP and dose values. The band represents the 95% confidence interval (CI). Four images for each detector with missing DAP values not included.

Table 4 Based on DAP values needed to obtain good image quality (rating of 4) according to the visual grading analysis score of the lungs VGAS_LUNGS (median score of pneumothorax, small airways, and infant respiratory distress syndrome (IRDS)), corresponding image acquisition parameters that deliver this DAP were extracted by the function of its curve. Thus, by use of DAP and image acquisition parameters, a calculation of organ doses, whole body dose, and effective dose was performed as described in the methodology section. Four images for each detector with missing DAP values not included.

	2.5 mm Al		3.5mm Al + 0.1 mm Cu
	CR needle CsBr	DR needle CsI	CR needle CsBr	DR needle CsI
VGASLUNGS = visual grading analysis score of the lungs, IRDS = infant respiratory distress syndrome, CR = computed radiography, DR = digital radiography, DAP = dose area product, WBD = whole body dose, E = effective dose, ICRP = International Commission on Radiological Protection
DAP needed for rating of “good” image quality [mGy1 · cm2]	2.6	2.9	2.6	2.2
Corresponding peak voltage [kVp]	60 | 66 | 77	60 | 66 | 77	60 | 66	57 | 77
Corresponding tube current · exposure time [mAs]	0.80 | 0.63 | 0.50	0.80 | 0.63 | 0.50	1.60 | 1.25	1.60 | 0.80
Calculated dose values
Thyroid dose [mSv]	0.026 | 0.028 | 0.029	0.029 | 0.031 | 0.032	0.031 | 0.032	0.026 | 0.028
Thymus dose [mSv]	0.025 | 0.026 | 0.028	0.028 | 0.029 | 0.031	0.031 | 0.032	0.026 | 0.028
Lung dose [mSv]	0.014 | 0.016 | 0.017	0.016 | 0.017 | 0.019	0.019 | 0.021	0.016 | 0.018
Breast dose [mSv]	0.033 | 0.034 | 0.035	0.037 | 0.038 | 0.039	0.036 | 0.036	0.030 | 0.031
Liver dose [mSv]	0.009 | 0.009 | 0.010	0.010 | 0.010 | 0.012	0.012 | 0.013	0.010 | 0.011
WBD [mSv]	0.005 | 0.005 | 0.006	0.005 | 0.006 | 0.006	0.006 | 0.007	0.005 | 0.006
E-ICRP103 [mSv]	0.010 | 0.011 | 0.011	0.011 | 0.011 | 0.012	0.012 | 0.012	0.010 | 0.011

Out of the 24 segments for which pixel information was automatically extracted, it was possible to evaluate 17. This was due to errors in the automatic information extraction, where data for the seven non-usable segments consisted for the most part of null values, probably because the segments were too thin. Out of the 17 usable segments, 15 segments had a Spearman’s correlation coefficient (ρ) to VGAS lower than 0.50 for both detectors, one segment had a Spearman’s ρ higher than 0.50 for one detector (signal profile curve of pericardium-lung border ρ = 0.56 (P < 0.0001) for DR 14s). Only the CNR of bone-to-soft tissue had a good correlation to the radiology raters’ VGAS with a Spearman’s correlation coefficient of 0.65 (P < 0.0001) for CR needle CsBr, and 0.54 (P < 0.0001) for DR CsI.

Discussion

In this study, we showed a comparable DAP, similar organ doses, and almost similar results for the assessability of images obtained via DR CsI detector and CR needle CsBr detector in neonatal chest radiography. In this first detailed analysis, it was possible to identify optimized exposure parameters suitable for clinical translation.

As expected, the CR needle CsBr detector has the highest absolute IQFinv values, and thus, a higher physical resolution capability without image post-processing due to the smaller pixel width (100 µm) and consequently higher spatial resolution. However, it should also be considered that a larger pixel width, as in the DR 14s detector (148 µm), does not necessarily translate to less image information. The reason is that a larger pixel can gather more X-ray photons per pixel and, therefore, provide a signal with lower noise [3].

The CR needle technology shows a more pronounced voltage dependency of IQ (represented by IQFinv) than the DR CsI technology, which we could evaluate at the two voltages of 66 kVp and 70 kVp ([Fig. 4]).

In contrast to the CDRAD 2.0 phantom results, the results of the neonatal chest phantom involved images that were all post-processed. The image post-processing, the latest generation from the vendor, is the most likely cause for the lack of a relevant difference between the two detector technologies. In addition, standard pediatric total filtering of 3.5 mm Al + 0.1 mm Cu showed no reduction in IQ (VGAS) compared to minimal filtering of 2.5 mm Al at the same DAP. Thus, the two detector technologies appeared to generate comparable results in a clinical setting of post-processed chest radiography despite their physical differences.

In addition, we showed how dose optimization could be applied in neonatal settings, particularly with the newer DR CsI technology. In our case, we used the VGAS_LUNGS as it best represented the difference in IQ among different image acquisition parameters, considering that some European criteria included in the VGC, e.g., vertebrae, were very well represented in all images of this study and are, therefore, less helpful in representing differences in IQ. The suggested kVp and mAs values ([Table 4]), used with the standard pediatric total filtering of 3.5 mm Al + 0.1 mm Cu, should provide an ideal compromise between ionizing radiation dose and IQ in the clinical setting.

Dose optimization, especially in the NICU, could significantly reduce the dose applied to infants. In our study, among the four different image acquisition settings that provide good IQ with CR needle CsBr and DR CsI with a total filter of 3.5 mm Al + 0.1 mm Cu, proposed in [Table 4], two had the lowest DAP (2.2 mGy · cm²): 57 kVp & 1.60 mAs; and 77 kVp & 0.80 mAs. We regard the latter, 77 kVp with 0.80 mAs, as the better settings.

The results of the VGAS of post-processed images, which show no significant difference in IQ between the CR needle CsBr technology and the DR CsI technology, are comparable to the previous results of Smet et al. [4]. We expanded this insight by providing a deeper understanding of the physical resolution capabilities of the detectors via contrast detail phantom analysis. We presented explicit image acquisition parameter recommendations for an ideal compromise between good IQ and radiation dose.

The one segment that provided a good Spearman’s correlation to the VGAS, namely the CNR of bone-to-soft tissue ([Fig. 6]), shows a graph trend relatively comparable to the VGAS’ trend. Here a plateau in IQ is also reached approximately at the same DAP level as per VGAS and this shows promise of being implementable for future studies, when addressing the described limitations. If successfully implemented, this kind of automated image quality evaluation by objective parameters could provide the time-sparing benefit of not having to have radiologists personally perform image quality evaluations.

Fig. 6 Objectively determined contrast-to-noise ratio (CNR) of bone-to-soft tissue. This parameter has good correlation to the radiology raters’ visual grading analysis score (VGAS) with a Spearman’s correlation coefficient of 0.65 (P < 0.0001) for computed radiography (CR) needle CsBr, and 0.54 (P < 0.0001) for digital radiography (DR) needle CsI. The band represents the 95% confidence interval (CI). Four images for each detector with missing DAP values not included.

Comparing the objective parameters to the raters’ VGAS could have been more valuable for generating usable data to replace the radiologists’ evaluation. We assume the limitation is due to minor, millimetric differences in the configuration of the neonatal chest phantom, which could not be fully compensated when using rigid registration. Due to the miniscule movements of the table, especially in the CR modality when extracting the detector cassette, the parts inside the neonatal phantom moved and lost their exact original positioning. As a suggestion, secure the neonatal phantom and the parts inside it with tape. This should prevent slight positioning imprecisions, which rigid registration cannot compensate for afterward. Another suggestion for improving this methodology is to introduce fixed physical reference points inside the phantom prior to imaging.

We have implemented these findings in clinical practice. [Fig. 7] shows a clinical image acquired with a DR detector using 3.5 mm Al + 0.1 mm Cu total filtering, 77 kVp and 0.63 mAs, thus an even lower current than the 0.80 mAs suggested from our calculations, resulting in an effective dose of 0.006 mSv. The dose area product was 2.1 mGy · cm². It is apparent that very good radiography is obtained, with very low dose.

Fig. 7 Clinical chest radiography of a 5-week-old neonate, with a weight of 3.7 kg, using a DR detector. Image acquisition parameters: 77 kVp, 0.63 mAs, total filtering of 3.5 mm Al plus 0.1 mm Cu. The dose area product was 2.1 mGy · cm². The calculated effective dose is 0.006 mSv. A very good clinical radiography is obtained with settings very close to those suggested by our calculations.

Although limited by being a phantom study, our study delivers usable data and provides a good starting point for clinical studies and implementation.

Conclusion and clinical relevance

We report a higher absolute image quality of computed radiography needle CsBr without image post-processing, which is voltage-dependent and more so than the DR detector. However, we show comparable clinical image quality of computed radiography needle CsBr and digital radiography CsI in neonatal phantom chest radiography with image post-processing. This is important for clinical practice, because the increasingly implemented DR technology provides comparable clinical image quality to the CR needle technology. As per the ALARA principle, we have also proposed different image acquisition parameters, of which we regard 77 kVp with 0.80 mAs as the ideal compromise between good image quality and effective dose. According to our study, with a standard pediatric total filtering of 3.5 mm Al + 0.1 mm Cu, an effective dose of approximately 0.010 mSv allows for good lung imaging in neonatal chest radiography. In first clinical implementations of DR imaging, even slightly lower current settings than those suggested by calculations, e.g. 0.63 mAs at 77 kVp, provide very good image quality, as shown in [Fig. 7]. This results in an effective dose of 0.006 mSv (3.5 mm Al + 0.1 Cu). As shown here, other institutions could implement these recommendations while adapting them to their own experience and expertise.

Conflict of Interest

Friedrich Wanninger is an employee of Agfa-Gevaert HealthCare GmbH, Peißenberg, Germany. Bernd Hoberg is a former employee of Agfa-Gevaert HealthCare GmbH, Peißenberg, Germany. Nothing to disclose for the other authors.

Acknowledgement

Agfa-Gevaert HealthCare GmbH procured the phantoms used in this study and helped in the process of image acquisition. We thank Prof. Sergios Gatidis for the help in the automated transfer of segments to the manually registered images, and for the automated pixel values extraction from the segments. The authors would like to thank Johann Jacoby for the statistical advice, as well as Georg Fehrenbacher and Michael Seidenbusch for the advice regarding the calculation of dose values. Part of the work presented in this manuscript has been presented as an oral presentation at ESPR Marseille 2022 and at GPR/KKJ Düsseldorf 2022. The work presented in this manuscript is part of the doctoral dissertation of the first author, Kevin Rama.

• References

• 1 Smans K, Struelens L, Smet M. et al. Patient dose in neonatal units. Radiation Protection Dosimetry 2008; 131: 143-147
• 2 Puch-Kapst K, Juran R, Stoever B. et al. Radiation Exposure in 212 Very Low and Extremely Low Birth Weight Infants. Pediatrics 2009; 124: 1556-1564
• 3 Apgar BK, Curley G, Vandenbroucke DAN. White Paper – Moving from CR to DR, Optimizing Image Quality and Dose. Agfa HealthCare N.V., Mortsel, Belgium, 2017. Downloaded on the 6th of July 2022. https://medimg.agfa.com/main/new-white-paper-moving-from-cr-to-dr/
• 4 Smet MH, Breysem L, Mussen E. et al. Visual grading analysis of digital neonatal chest phantom X-ray images: Impact of detector type, dose and image processing on image quality. Eur Radiol 2018; 28: 2951-2959
• 5 Martin L, Ruddlesden R, Makepeace C. et al. Paediatric x-ray radiation dose reduction and image quality analysis. J Radiol Prot 2013; 33: 621-633
• 6 Båth M, Månsson LG. Visual grading characteristics (VGC) analysis: a non-parametric rank-invariant statistical method for image quality evaluation. BJR 2007; 80: 169-176
• 7 European Commission. European guidelines on quality criteria for diagnostic radiographic images: In paediatrics. EUR 16261. Luxembourg: European Commission; 1996
• 8 Bundesärztekammer. Leitlinie der Bundesärztekammer zur Qualitätssicherung in der Röntgendiagnostik. Deutsches Ärzteblatt 2023; 85
• 9 Bildgebende Diagnostik bei Kindern. Empfehlung der Strahlenschutzkommission. 2022. Strahlenschutzkommissionm 321. Sitzung der Strahlenschutzkommission: 37.
• 10 Al-Murshedi S, Hogg P, England A. An investigation into the validity of utilising the CDRAD 2.0 phantom for optimisation studies in digital radiography. BJR 2018; 91: 20180317
• 11 van der Burght RJM, Floor M, Thijssen MAO. et al. CDRAD 2.0 Phantom & Analyser Software Manual Version 2.1. 2014
• 12 Hakansson M, Svensson S, Zachrisson S. et al. VIEWDEX: an efficient and easy-to-use software for observer performance studies. Radiation Protection Dosimetry 2010; 139: 42-51
• 13 Svalkvist A, Svensson S, Håkansson M. et al. VIEWDEX: A STATUS REPORT. Radiat Prot Dosimetry 2016; 169: 38-45
• 14 Seidenbusch M, Rösenberger V, Schneider K. Imaging Practice and Radiation Protection in Pediatric Radiology: Conventional Radiography. Cham, Switzerland: Springer International Publishing; 2019
• 15 ICRP. The 2007 Recommendations of the International Commission on Radiological Protection. ICRP Publication 103. 2007;
• 16 Fedorov A, Beichel R, Kalpathy-Cramer J. et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magnetic Resonance Imaging 2012; 30: 1323-1341

Publication History

Received: 11 July 2024

Accepted after revision: 21 January 2025

Article published online:
18 February 2025