From May 2023 to November 2024, our prospective, quantitative cross-sectional study obtained data from 41 X-ray systems in 40 hospitals. The hospitals were selected using a maximum variation sampling strategy to ensure representative inclusion from urban, peri-urban, and rural settings in 12 of Ghana’s 16 regions. Two modalities were evaluated: computed radiography and digital radiography. For each X-ray system, machine-specific parameters including type, model, and total filtration were recorded. Table 1 presents the year of manufacture of the selected X-ray systems. All systems were equipped with manual exposure control; only 33 had functional automatic exposure control (AEC). All the X-ray systems had adequate inherent filtration and anti-scatter grid configurations.

Data were prospectively collected from adult patients (≥18 years) who had been referred for chest posterior–anterior (PA), chest lateral (LAT), pelvis anterior–posterior (AP), skull AP/PA, skull LAT, lumbar spine (LS) AP, LS LAT, and abdomen AP examinations. The examination data were included only if the resulting radiographs had been deemed diagnostically acceptable by radiographers and subsequently used by clinicians for diagnosis. Examination data from critically ill and nonconsenting patients were excluded. Per technical report series 457 from the International Atomic Energy Agency [20], patient demographic details such as age, sex, weight, and height were recorded, with body mass index (BMI) being subsequently estimated. Ethics approval was obtained from the Ghana Health Service Ethics Review Committee, and the necessary permissions were secured from participating hospitals before the study began.

Before patient dosimetry was captured, quality control was conducted on the X-ray equipment. Image quality and performance tests were conducted to assess the reproducibility and constancy of exposure parameters: kilovoltage peak (kVp), tube current×time (mAs), exposure time, and half-value layer (HVL). Beam collimation and perpendicularity were also verified. All tests adhered to protocols outlined in the International Atomic Energy Agency’s Technical Cooperation Regional Project RAF6/053 radiologic quality control manual [21, 22]. Direct exposure measurements were obtained using a calibrated black Piranha solid-state dosimeter (type 657; RTI Group) interfaced with the Ocean software application (2014; RTI Group), capable of measuring tube voltage at outputs as low as 50 kV/0.05 mA at 50 cm [23].

Entrance surface dose (ESD) in milligrays was used to evaluate dosimetric quantity. The ESD was estimated indirectly using exposure parameters such as focus-to-detector distance (FDD) and focus-to-skin distance (FSD), together with technical settings (kVp, mAs), the X-ray system’s tube output, and examination-specific backscatter factors (BSFs) [24, 25], as follows:

Equation (1) was then modified to incorporate the tube output correlation equation as well as the kVp and mAs correction factors (K_kVp, K_mAs) for each X-ray system. Using the dosimeter, radiation exposure in milligrays was measured for each exposure factor setting to determine the X-ray tube output and correction factors. The dosimeter was placed on the patient couch with its radiation-sensitive active area perpendicular to the X-ray central beam axis at an FDD of 1 m.

The correction factors were calculated by determining the ratio between the set exposure factors and their corresponding measured values for three consecutive measurements with uniform increments. The averaged ratios were used to derive the correction factors for the exposure settings. To determine the tube output equation, the tube voltage was experimentally varied from 40 kVp to 110 kVp in 10 kVp increments, while maintaining a constant tube current of 5 mAs. The exposure (mGy·mAs^–1) against kVp² was plotted for each X-ray system, and the gradient (s) and intercept (C) of the linear correlation equation from the resulting graph were substituted into Equation (1). The general form of the modified Equation (1) is, as Equation (2):

The BSF for each radiographic procedure was adopted from Martin (2011) [26].

A systematic review of kVp, mAs, FDD, FSD, and ESD for data cleaning identified and corrected inconsistencies and removed records with missing values and screen outliers. FDD was retained only within the good-practice ranges recommended by the European Guidelines on Quality Criteria for Diagnostic Radiographic Images: chest AP and LAT, 140–200 cm; pelvis AP, skull AP/PA and LAT, LS AP and LAT, and abdomen AP, 100–150 cm [27]. Values outside those bands were cross-checked against originals and excluded if unsupported. FSD plausibility was verified from imaging geometry (FSD=FDD–[approximate patient-to-detector gap+typical body thickness]) with the constraint 0<FSD<FDD, and checks were conducted with input from an experienced radiographer (>4 years). Cross-field consistency checks ensured that, at comparable kVp and FDD, higher mAs did not cooccur with an implausibly low ESD. Demographics were cleaned around 70±10 kg to reflect typical national practice without suppressing genuine body-size variability. Obvious errors were queried (adult height <1.2 m or >2.1 m; weight <30 kg or >200 kg), and duplicates were removed. Statistical summaries by system and examination (mean, standard deviation [σ], minimum [Min], maximum [Max], and median), percentile-based outlier flags, and facility-level box-and-whisker plots were used to detect pattern deviations and highlight extreme outliers. Records with discrepancies unresolved after query were corrected where possible; otherwise, they were excluded.

The NDRL for each radiographic examination was estimated using the 75^th percentile of the facility-specific median ESD for each examination/projection, in line with the recommendations of ICRP 135 [1]. Results from this study are presented with descriptive statistics using Min–Max, mean, median, and σ. NDRLs from this study were compared with established and proposed NDRLs from other countries.

The 41 evaluated X-ray systems were the products of 11 manufacturers: Varian, Shimadzu, Varex, IMD Generators, GE Hualun, MEDEX, Shenzhen, Siemens, Philips, Mikasa, and Samsung. The most prevalent brands were Shimadzu and Philips, each constituting 24.4% of the fleet. Table 1 details the system distribution by year of manufacture. The largest subgroup consisted of systems manufactured during 2009–2012 (31.6% of the 38 units with available records). Notably, 57.9% of the systems were produced during 2013–2024, demonstrating the country’s commitment to modernizing its radiology infrastructure for improved patient dose optimization. Older systems produced during 1997–2012 (42.1%) might lack the advanced dose reduction and image-processing technologies found in more modern equipment, potentially leading to elevated patient doses and suboptimal image quality [28]. That possibility reinforces the importance of establishing Ghanan NDRLs. Quality control tests indicated the overall acceptable performance of the X-ray systems. However, 11 systems had measured HVLs ranging from 2.3 mmAl to 2.9 mmAl, with seven legacy systems (1997–2012) measuring <2.5 mmAl, below the minimum total filtration for X-ray systems at >70 kVp [29]. Those results were communicated to the respective facilities for system rectification. Nevertheless, examination-specific point-biserial correlations (r) were computed for HVL noncompliance (binary: 1=prior noncompliant, 0=compliant) with facility-level median ESD (continuous) to quantify the association between group status and dose. That analysis assessed whether doses from HVL noncompliant X-ray systems systematically differed from doses from compliant systems and from the national dose distribution.

Table 2 summarizes the patients’ sex, age, weight, height, and BMI for each examination/projection. The data were obtained from 7,643 patients (3,757 men, 49%; and 3,886 women, 51%), aged 18 to 88 years. ICRP recommends standardizing patient size for dosimetry surveys because patient size influences the X-ray dose required for the acquisition of radiographs of good diagnostic quality [1]. That standardization was achieved by defining patient weight as 40 kg to 100 kg to capture a wide spectrum of patients with varying body habitus. A mean weight of 69±7 kg was achieved, which resembles the standard-size adult of 70±0 kg suggested by the ICRP. The mean BMI of the study patients was 24.5±0.8 kg/m² which is similar to Ghana’s average BMI of 24.7 kg/m² [30].

In radiographic imaging, the selected radiographic exposure parameters directly influence the balance between diagnostic efficacy and patient safety. Imaging by the various systems surveyed here was conducted by qualified radiographers and radiology technicians following the standards of Ghana’s Allied Health Professions Council and the Ghana Health Service. These professionals undertake various continuous professional development activities as a prerequisite for the annual renewal of their professional license to practice.

In diagnostic radiography, increasing the kVp enhances beam penetration and allows for lower mAs settings, thereby reducing the patient dose; however, image contrast might be compromised, potentially obscuring subtle pathologies. Table 3 summarizes the exposure factors (kVp, mAs, FDD, and FSD) used in the nationwide dosimetry survey. Notable interhospital variability in kVp and mAs was attributable to differences in radiographer technique (AEC not always being used), equipment performance, and patient habitus. As expected, kVp and mAs were lower in chest examinations, whereas higher settings were used for LS imaging, consistent with clinical requirements and the existing literature [3, 7–13]. Nine FDD settings (100–180 cm) were recorded: chest PA and LAT were predominantly performed at 180 cm, whereas non-chest projections clustered near 100 cm. FSD distributions closely mirrored those of the FDDs. Sample skewness (g₁) and distributional shape metrics for non-excess kurtosis (κ) indicated that chest PA and LAT demonstrated positive skew and were platykurtic (g₁=0.51, κ=1.89, and g₁=0.31, κ=2.01, respectively), abdomen AP and pelvis AP were right-skewed and leptokurtic (g₁=1.90, κ=4.21, and g₁=2.81, κ=4.31, respectively), and LS AP and LAT were right-skewed with moderate tail heaviness (g₁=1.85, κ=2.15, and g₁=2.11, κ=2.09, respectively) and similar cranial patterns. The σ values in Table 3 highlight targets for optimization: The recommendation is that departments standardize protocols (for instance, fix default FDD by projection at the console, mark tube positions, lock stand/table geometry, etc.) and verify AEC time limits and mA/kVp to maintain short exposure times. Radiographers should annotate justified departures from the local standard FDD. Such annotations will assist interpreters in accounting for magnification and part-size changes when comparing serial images [31], and will support effective dose audit and optimization initiatives. Of the eight radiographic procedures evaluated, chest PA was the most frequent, being conducted by all the surveyed X-ray systems (n=41). Following, in order, were LS AP (n=40), LS LAT (n=40), pelvis AP (n=36), abdomen AP (n=33), skull AP/PA (n=32), skull LAT (n=21), and chest LAT (n=18).

Table 4 presents descriptive statistics for the ESD values across X-ray systems, including the 25^th, 50^th, and 75^th percentiles. The 50^th percentile (median) represents the achievable doses received by patients, which allows for optimization of doses at local radiology units. The 75^th percentile is used for setting NDRLs. The dose variations observed across the procedures of interest reveal the influence of anatomic region, patient habitus, and exposure parameters on radiation dose. Chest PA had the lowest median ESD (0.19 mGy), consistent with lower attenuation in the thoracic cavity, whereas LS LAT had the highest ESD (6.52 mGy), reflecting increased tissue density and the necessity for higher exposure settings.

Fig. 1 also depicts the ESD distributions across the surveyed X-ray systems for all projections, with box-and-whisker plots illustrating the interquartile ranges, which capture ESD dispersion and account for outliers.

High-dose variability was observed in the pelvis AP, skull AP and PA, abdomen AP, and LS AP and LAT examinations. LS LAT had the largest interquartile range, closely followed by LS AP. Although the variation in chest PA was substantial, its lower ESD values minimize its effect relative to the higher-dose examinations. Fig. 1 also highlights outliers in some examinations. The high-dose outliers observed in the chest PA and LAT, pelvis AP, and the LS LAT came from examinations that required higher exposure factors to account for increased patient habitus and greater tissue attenuation. Conversely, the low-dose outliers in the pelvis AP examinations were characteristic of systems using modern radiography equipment under effective QA frameworks. There, improved detector efficiency, use of AEC, and advanced image post-processing capabilities enabled diagnostically adequate imaging at reduced ESD values compared with the broader dataset. The inclusion of results from those outlying units created a broader picture reflecting current clinical practices in Ghana. Such variations further demonstrate the need for standardized exposure protocols and technologic advancements to optimize patient dose while maintaining diagnostic efficacy.

Table 5 presents a comparison of the NDRLs proposed for Ghana in this study with established NDRLs from Italy [32], the United Kingdom [33], Japan [34], and Korea [35], as well as proposed NDRLs from Kenya [36], Iran [10], and Sudan [37].

The proposed Ghanan NDRL for chest PA is, respectively, 10% lower and 20% higher than those for Japan at <100 kV and ≥100 kV; 16.7% higher than that for Italy; 47.1% lower than that for Korea; and 140% higher than that for the United Kingdom. For pelvis AP, the proposed Ghanan NDRL is 9.7% higher than that for Japan, but approximately 30% lower than that for both Korea and the United Kingdom. In LS imaging, the proposed Ghanan NDRL for the AP projection is 3.8% higher than that for Japan, 24.6% lower than that for Korea, and 36.1% lower than that for the United Kingdom. The proposed Ghanan NDRL for the LS LAT projection is even lower—specifically, 20.8%, 16.0%, and 28.8% lower than those for Japan, Korea, the United Kingdom respectively. A similar pattern emerged in comparisons with proposed NDRLs for other countries, with the Ghanan NDRL proposed in this study for chest PA being 40% lower than that proposed by Ahmed et al. [37] (Sudan), 42.9% lower than that proposed by Mohsenzadeh et al. [10] (Iran), and 28% lower than that proposed by Korir et al. [36] (Kenya). For pelvis AP, the proposed Ghanan NDRL is 41.5% higher than that proposed for Iran, but 44.6% lower than that proposed for Kenya. Furthermore, the Ghanan NDRL proposed in this study for LS LAT is 11% lower than that proposed for Sudan, 40.7% higher than that proposed for Iran, but 35.3% lower than that proposed for Kenya.

The disparity in proposed NDRLs across countries might be attributable to variations in imaging protocols, patient demographics, and the intrinsic technologic capabilities of the equipment in use (such as detector efficiency and AEC use). The lower institutional DRLs reported by Gyan et al. [25] in Ghana for computed radiography (chest PA, 0.4 mGy; LS LAT, 3.4 mGy) and digital radiography systems (chest PA, 0.3 mGy; LS LAT, 3.1 mGy) reveal that the integration of advanced dose management techniques, implementation of effective QA programs, and optimization of protocols could result in significant dose reductions at individual units in the country. In procedures such as chest PA and LS AP, standardization could be more complex relative to skull AP and PA because of the relative invariability of skull size in different patients. Nevertheless, optimization of programs should equally target high-frequency procedures such as chest PA, high-dose examinations such as LS AP and LAT, and anatomic regions with highly radiosensitive organs (such as in pelvis AP).

Comparisons between previously HVL noncompliant X-ray systems and the remaining X-ray systems indicated that performance was generally consistent with national distributions. For chest PA, the noncompliant group had a median ESD of 0.28 mGy, 47.4% above the national pooled median and yet 22.2% below the NDRL, with an exceedance rate of 18.2% compared to 23.3% among compliant X-ray systems and a negligible point-biserial correlation (r=0.076). Abdomen AP was observed to be 2.28 mGy (1.8% above the national median), with 25% of noncompliant systems compared with 20% of compliant systems exceeding the proposed NDRL (r=0.002). In the noncompliant group, doses were higher for pelvis and LS examinations, although the medians remained below the respective NDRLs. In the noncompliant group, pelvis AP was observed to be 2.61 mGy, 10.1% above the national pooled median and 5.8% below the NDRL, with an exceedance rate of 50.0% compared with 15.4% for compliant systems (r=0.084). LS LAT was observed to be 6.78 mGy, 3.9% above the national pooled median and 4.8% below the NDRL, with 40.0% of noncompliant systems compared with 20.0% of compliant systems being above the NDRL, for a small point-biserial correlation (r=0.380). Observations for chest LAT, skull, and LS AP examinations were similar. Overall, the small-to-negligible point-biserial correlations (|r|=0) reveal no significant linear association between HVL noncompliance and facility-level median ESD. Thus, a system noncompliant status for HVL does not explain the variability in median ESD across facilities (r²=0%) and indicates less than 1% variance. That is, median ESDs are not systematically higher or lower in the previously noncompliant groups of X-ray systems.

X-ray systems have finite service lives, and units 10 years and older are no longer considered state of the art [28]. In the present study, X-ray system manufacture dates spanned approximately 25 years, a range that could introduce systematic variation in ESD attributable to equipment age. As highlighted in Table 6, analysis of system-level median ESDs revealed that modern systems (2021–2024) had 75^th percentile values below the proposed NDRLs: approximately 24.6% for abdomen AP, 13.9% for chest PA, 11.9% for pelvis AP, 9.7% for LS LAT, 6.6% for LS AP, and about 5.1% for skull LAT, with skull AP and PA approaching the NDRL (approximately 0.5% below). By contrast, legacy systems (1999–2012) exceeded the proposed NDRL in higher-dose projections: for example, by approximately 14.8% for LS AP, 15.4% for skull LAT, 9.4% for abdomen AP, and 6.0% for LS LAT. Intermediate systems (2013–2020) were generally close to or modestly below the NDRL, with exceptions such as chest LAT at approximately 10.3% above the proposed NDRL. Because the NDRLs were defined as the 75^th percentile of facility medians within the latter group, the higher central tendency and wider dispersion seen with legacy systems inflate the pooled reference levels.

Given the foregoing observations, optimization efforts for legacy systems should prioritize patient-specific selection of exposure parameters; judicious collimation use; supplemental filtration where indicated; effective preventive maintenance and calibration schedules; routine dose auditing with concurrent radiographic quality review; strengthened periodic QA/control programs, with integration of LDRLs and NDRLs; detector recalibration; and scheduled updates to examination protocols and image-processing software. Those measures should run in parallel with a structured replacement strategy: ideally, a rolling, forward-looking plan with a minimum 5 years horizon and annual review as recommended by the European Society of Radiology [28]. As fleet composition shifts toward newer technologies, NDRLs are expected subsequently to decline, further enhancing patient dose optimization.

The findings in this study further reinforce the necessity for establishing country-specific NDRLs reflective of a country’s patient demographics and clinical practices. Given the influence of population-specific characteristics on radiation dose distribution, a one-size-fits-all approach is inadequate for effective dose optimization. We recommend that regulatory authorities take steps to begin establishing NDRLs in radiography by promoting collaboration between researchers, healthcare institutions, and governmental health agencies. In Ghana, the Nuclear Regulatory Authority might consider assessing existing national dose data and consolidating them into a unified set of NDRLs to serve as an initial benchmark. A structured framework for implementation that incorporates standardized dose tracking systems and periodic review mechanisms is essential to ensure that the NDRLs remain dynamic and responsive to technological advancements and evolving clinical protocols, thereby enhancing patient radiation protection while maintaining diagnostic accuracy.