© M. Alvarez et al., Published by EDP Sciences 2025

1 Introduction

In recent decades, there has been a significant increase in the number of interventional radiology procedures (Kim et al., 2012). These procedures require special attention in terms of the harmful effects of ionizing radiation, since the complication of some procedures, the long exposure time and the number of exposures increase the dose absorbed by patients (Triedman and Newburger, 2016). In view of the growing number of procedures, an in-depth study of the Dose levels used in the procedures is essential.

One of the ways to evaluate clinical practice in interventional radiology sectors is to determine Diagnostic Reference Levels (DRLs). These levels serve as an indication of whether, for a given procedure performed, the Dose value received by a patient is higher or lower than a typical Dose value for this procedure (ICRP Publication 103, 2007; Talbi et al., 2022). The first DRLs were implemented in conventional radiography procedures in the 1980s and were developed for other modalities in the 1990s (Vañó Carruana et al., 2013; Semghouli et al., 2024). The determination of DRLs is associated with dose optimization (Damilakis and Vassileva, 2021) and considers dozens of institutions and equipment to values determination, established by the International Commission on Radiological Protection (ICRP) as the third quartile value of the distribution dosimetric magnitude evaluated for a given clinical diagnostic procedure (Vañó Carruana et al., 2013). However, the study of the typical Dose value for a procedure can be carried out within a single institution or a single piece of equipment, through the determination of Local Diagnostic Reference Levels (LDRLs). LDRLs indicate whether, for a given procedure performed in specific equipment and institution, the dose value received by a patient is higher or lower than normal for this procedure (Semghouli et al., 2024).

The ICRP enforces that, in the absence of national DRLs, LDRLs can be introduced to assist in the Dose optimization process (Vañó et al., 2017). The advantages of this approach are its relative ease in being calculated and updated according to the needs of the evaluated clinic. Furthermore, its determination is more flexible than that of national DRLs (Semghouli et al., 2024). Therefore, it is essential to create methodologies for collecting dosimetric quantities that allow the calculation of Typical Dose and the determination of LDRLs.

In the case of Interventional Radiology procedures, the Air Kerma (or input Kerma at the interventional reference point) or Kerma-Area Product (KAP) quantities are commonly provided by the angiography equipment. This methodology, however, requires mobility and time from the institution’s team of medical physicists. It is therefore necessary to create innovative methodologies that allow the collection of this information quickly and on a large scale. In the evaluated institution, the ArtisZee Ceiling ® (Siemens Healthineers, Germany) angiographic equipment, is used in hemodynamic procedures.

Another interest is the estimation of the biological risk to which the patient population is exposed during invasive procedures. The PCXMC 2.0 software is a Monte Carlo simulation software that receives dosimetric parameters (air Kerma or KAP) as input to calculate the history of x-ray photons and allows customization regarding the projection used, additional filtration, focus-detector distance, skin focus, irradiated region, among others (Tapiovaara and Siiskonen 2008) and provides the calculation of Effective Dose (E) according to ICRP 103 weighting factors. In this way, once the dosimetric value for the evaluated procedure is known, the estimated value for that patient and procedure can be calculated. Thus, the objective of this work was, based on data collection using a semiautomatic software, to determine LDRLs for the main hemodynamic procedures of a large hospital, compare with typical values from other institutions and use the LDRLs as input in PCXMC 2.0 Monte Carlo simulation software for E estimation.

2 Methods

The study was carried out in a large Brazilian hospital, which has a dedicated hemodynamics sector with a monoplane Artis Zee® (Siemens Healthineers, Germany) C-arm. At the end of each procedure, this equipment provides a report in the form of an image containing, in addition to information regarding the type of procedure in the Digital Imaging and Communications in Medicine (DICOM) header, technique used, KAP and nominal Air Kerma, the number of exposures in cine mode and the total exposure time. Unfortunately, information regarding dosimetry is stored in the DICOM image and not in the DICOM reader, which makes it difficult the access to this information in an automatic way. As in Radiation Dose Management Systems (RDMS) when the DICOM dosimetric data is not available the RDMS analyses the screenshot dose report by Optical Character Recognition (OCR) (Riccardi et al., 2018; Loose et al., 2021) This way, we developed a software to extract these information’s using OCR technology (TulioGSMarques, 2024). This software receives as input the final report of the procedure in DICOM format, as shown in Figure 1.

The software then converts the DICOM file into a PNG file, in which it reads the pixel arrangement and associates it into letters and numbers, extracting the information about Air Kerma (K) in the equipment’s reference point, Kerma-Area product (KAP) in the equipment’s reference point, number of exposures in cineangiography mode (N), and procedure time (T). The software received all performed procedures on the C-arm equipment of a large hospital performed from January 2021 to September 2023. The workflow of the authoring software for data collection is illustrated on Figure 2.

Fig 1 Example of a report provided by the equipment. Quantities such as kV, Air Kerma, KAP, number of exposures, exposure time and date of examination are made available in the form of an image.

Fig. 2 Flowchart representing the operation of the software for extracting quantities from the reports provided by the Artis Zee Ceiling equipment. (I) The first step is to determine the evaluation period for the procedures. (II) The reports provided by the equipment are extracted, added to the program’s image base and additional information is removed from the DICOM header associated with the procedure. (III) The Pytesseract library, in Python, is used to transcribe the data into the report. Information provided in the DICOM header, such as procedure name or patient ID, was extracted. (IV) The information extracted is reviewed by the authors. (V) The verified data is added to the database.

2.1 Inclusion and exclusion criteria

When patients underwent multiple procedures on the same day, identified by the same record in the PACS system (such as two arteriography procedures on the lower limbs for the same patient), the dosimetric data from these procedures were aggregated and considered as a single procedure. During the validation phase, the OCR-generated data were scrutinized, and any entries with data inconsistencies were eliminated, resulting in 8 exclusions. Consequently, 1334 entries were deemed valid and successfully extracted by the algorithm, achieving a data extraction accuracy rate of 95%.

2.2 Data analysis

Procedures were grouped according to the most relevant classification as: Coronary percutaneous transluminal angioplasty, stent placement, or radiofrequency ablation; Lower limb arteriography; Head and/or neck angiography; Coronary angiography (diagnostic); Pelvic vein embolization; Thoracic angiography of pulmonary artery or aorta; Abdominal angiography or aortography and Transjugular intrahepatic portosystemic shunt placement.

The LDRLs were established at the third quartile (Q3) for the values of K, KAP, T, and N. The median was determined to showcase the Typical Dose Values, in accordance with the guidelines recommended in ICRP 135. Even though dosimetric data typically do not follow a normal distribution, mean values were also computed due to their frequent reference in ICRP 135. For the statistical analysis of each variable, the Python library Numpy was employed for its data analysis capabilities.

2.3 Determination of effective dose and conversion factor

With the LDRLs obtained, the PCXMC 2.0 software was used to calculate the Effective Dose (E) in patients, considering the weighting factors of the ICRP103 available in the Monte Carlo software. This software allows for detailed customization of projection angles, additional filtration, and focus-detector distance. We used the PCXMC Rotation feature to simulate the various projections and accurately calculate the Effective Dose. Monte Carlo beam simulation and projections were matched with the protocol used in the routine and are specified in Table 1, including C-arm projections for each procedure type.

The kV values and additional filtration were determined using the median values from our clinical data. The standard dose and the most used Field of View (FOV) were selected for the simulations. These scenarios were established by analysing the reports provided by the equipment, which detailed the kV values and the percentage contributions of each projection to the total KAP. These projections include adult cardiac catheterization, bile drainage, catheter implant and lower limb arteriography.

All equipment underwent rigorous quality control procedures in accordance with Brazilian regulatory standards for fluoroscopy protocols. The tolerance for the percentage difference between displayed values and measured values for Air Kerma and KAP is ≤20% This ensures the accuracy and reliability of the dose measurements used in our study (Ministério da Saúde, 2021).

In each simulation, the median values for kVp and added filtration were utilized. The most frequently used projection and field of view (FOV) settings were also applied. The stories of one million photons were applied, focus-detector distance of 120 centimetres, focus-skin distance of 80 cm. The simulated patient is 178.2 cm tall, 73.2 kg, and 30 years old. The ratio between the E values and the 3rd quartile value for K were used to calculate the conversion factor (Cf), as described by equation (1): $$cf=\frac{E}{K},$$(1)

where E is in mSv, K is measured in mGy, and Cf in Sv/Gy. Subsequently, the results were compared with previous studies carried out for dose surveys in interventional radiology procedures.

3 Results

Table 2 presents an overview of interventional radiology procedures, highlighting number of exams, air kerma (K, measured in mGy), Kerma Area Product (KAP, measured in μGym²), procedure duration (T, measured in minutes), and the number of exposures taken (N). The first column lists the procedures, as “Coronary percutaneous transluminal angioplasty, stent placement, or radiofrequency ablation” had the most exams (526), followed by “Lower limb arteriography” (386) and “Head and/or neck angiography” (168). The median values for K (air kerma) vary across procedures, with the highest observed in “Head and/or neck angiography” at 645.5 mGy, while the lowest is seen in “Coronary angiography (diagnostic)” at 28.0 mGy.

The KAP median values reveal a broader range, with the “Transjugular intrahepatic portosystemic shunt placement” recording the highest median at 18,794 μGym², and the “Coronary angiography (diagnostic)” showing the lowest median at 625 μGym². Similarly, the procedure duration (T) median spans from 2.6 minutes for “Lower limb arteriography” to 20.2 minutes for “Thoracic angiography of pulmonary artery or aorta”.

Figure 3 presents a series of boxplots summarizing the distribution of radiation metrics across various fluoroscopically guided interventional procedures. Panel A displays the range of doses (in mGy) absorbed by patients, illustrating substantial variability across procedures with some, like Thoracic Angiography and Abdominal Aortography, reaching up to 2000 mGy. Panel B shows the Kerma-Area Product (KAP) measured in μGy · m², where procedures involving higher complexity or longer duration, such as Abdominal Aortography, again demonstrate elevated values, some exceeding 60,000 μGy · m². Panel C details the procedure duration in minutes, highlighting significant procedural duration for Hepatic Artery Embolization and Thoracic Angiography, with times occasionally extending beyond 50 minutes. Finally, Panel D illustrates the number of exposures per procedure, with certain procedures like Thoracic Angiography necessitating upwards of 30 exposures, indicating a higher complexity and possibly greater patient risk.

Table 3 provides a detailed comparison of radiation exposure metrics across various interventional radiology procedures, juxtaposing findings from the present study against reference values from the literature and established standards. The Kerma Area Product (KAP) values and Effective Dose (E) are presented in columns for both the current study and referenced studies, highlighting the variance between them.

For instance, the KAP for coronary percutaneous transluminal angioplasty, stent placement, or radiofrequency ablation from this study is reported at 29.8 Gy · cm², which is slightly higher than the reference value of 26.0 Gy · cm². Conversely, the KAP for head and/or neck angiography shows a significant increase in the present study, 163.9 Gy · cm², compared to the referenced 14.1 Gy · cm². This may be due the routine use of 3D Cone Beam acquisitions, which were not cited in the reference literature (Aroua et al., 2007).

Effective dose (E) measurements in the present study also show substantial agreement from referenced data, with the most notable disparity observed in abdominal angiography or aortography, where the present study’s value is 30.5 mSv compared to 73.9 mSv from the literature, which may be due the lower number of CINE acquisitions used in our institution.

The associated error (%) presented in Table 3 was calculated based on the percentage difference between the value from the Present Study and the NCRP 160 reference value using the following equation (2): $$AssociatedError(\%)=(\frac{PresentStudy(\mathrm{DCCE})-\mathrm{NCRP}160(\mathrm{DCCE})}{\mathrm{NCRP}160(\mathrm{DCCE})})\times 100$$(2)

where:

• Present Study (DCCE) refers to the value of the Dose Cumulative Corresponding Exposure (DCCE) determined in the current study.

• NCRP 160 (DCCE) refers to the value of the Dose Cumulative Corresponding Exposure (DCCE) as reported in the NCRP 160 report.

Fig. 3 Boxplots represent four measurements from different medical procedures: (A) Dose (in mGy), (B) Kerma-Area Product (KAP in μGym2), (C) Time (min), and (D) Number of Exposures. The procedures are grouped along the x-axis with each boxplot illustrating the distribution of data for each variable.

4 Discussion

The original OCR software met the development proposal, accurately extracting information from PNG images and being a valuable tool for controlling doses in equipment whose value is not available in DICOM headers. The use of Monte Carlo simulations to estimate the Effective Dose in interventional radiology procedures has already been discussed by Aroua et al. (2007). The authors determined through a study with a polymethyl methacrylate (PMMA) simulator and Monte Carlo simulations that a simple conversion factor can be used for estimates of E through KAP values. The advantage of the method proposed in the present study is the semi-automatic collection of data, without the need for on-site presence, providing agility to the radiology safety officer staff.

The data presented in Table 1 provide insight into the typical radiation doses patients might receive during such procedures and serve as a potential benchmark for future studies and dose management.

As expected, our findings indicate that the median Kerma-area product (KAP) and median exposure time vary significantly depending on the complexity and duration of the procedures. For instance, procedures such as thoracic angiography and abdominal aortography demonstrate higher median KAP values, which is reflective of their intricate nature and the necessity for prolonged fluoroscopy time.

The mean KAP values reported here can be instrumental in setting dose thresholds within radiation management programs, especially when considering the 75th percentile as a reference point. This is aligned with the Joint Commission’s requirements for radiation dose management and can contribute to establishing standards that ensure patient safety without compromising diagnostic efficacy.

Notably, the majority of procedures have median effective doses that exceed the annual background radiation, indicating the importance of judicious use of fluoroscopy to minimize patient radiation exposure. Despite this, the median values for many procedures remain below dose levels associated with increased risk for radiation injury, which underscores the effectiveness of current radiation safety practices in interventional radiology.

Regarding radiation injury, the ICRP suggests that doses to the skin above 2 Gy can lead to deterministic effects such as erythema and hair removal, with higher doses potentially causing more severe injuries such as necrosis. Our study shows that procedures such as thoracic angiography and abdominal aortography, which have higher KAP values and exposure times, are more likely to reach these dose levels if not carefully monitored. This demonstrates the importance of maintaining rigorous radiological safety protocols and continuous monitoring to ensure that doses are kept as low as reasonably achievable (ALARA) while achieving the desired clinical outcomes, reaffirming the effectiveness of our current radiological practices.

The study also sheds light on the necessity for patient-specific dose considerations, as evidenced by cases with significantly higher KAP values, such as those involving bilateral uterine artery embolization. Such instances highlight the need for tailored radiation management strategies to further optimize patient safety.

Comparisons with data from the NCRP Report 160 and 172 reveal that the 75th percentile KAP values from our study are generally lower, suggesting that advances in technology and technique have potentially led to reductions in patient radiation exposure over time.

It is important to mention that, in clinical practice, the focus-skin distance and projection angle may change, and the kV may not remain constant, as Automatic Exposure Control may change the technique used. It is also essential to mention that the quantities used for simulations with PCXMC 2.0 represent a generalized average of the techniques, distances, projections, and filtrations most used in these procedures. Consequently, these values may vary when considering the wide range of patient ages and thicknesses.

The calculated conversion factors are consistent with the KAP values, and the number of radiosensitive organs irradiated in the simulation, serving as an important tool for risk estimations. Olcay et al. (2015) measured an approximately four-fold difference in KAP values in procedures performed with cine mode imaging compared to those performed in “last fluoro hold mode”, in which the last image obtained in fluoro mode can be saved. This technology is present in the equipment evaluated in this work and the reduction in the number of pulses is essential for the optimization of doses without a decline in image quality (Pyne et al., 2014). The evaluated equipment also has automatic filtration control, which plays an important role in reducing scattered radiation, but requires a greater operating voltage range (Martin 2007).

It’s important to acknowledge that the LDRL methodology employed only accounts for a segment of the regional population, and the E values should be viewed as prudent estimates, influenced by individual patient factors such as age, sex, and Body Mass Index (BMI) (Wambani et al., 2014). For a more thorough evaluation, it’s necessary to calculate E considering the specifics of various procedures and patient biotypes, to prevent LDRLs from becoming outdated and disconnected from the values found in clinical practice.

Wambani et al. (2014) contend that using E as a generic measure to estimate the occurrence of stochastic effects can result in inappropriate oversimplifications of the complex biological processes associated with the development of such effects, highlighting that the quantity assessed for determining the biological risk associated with medical exposures is heavily dependent on the specific context. Conversely, the ICRP strongly advocates the use of LDRLs as an initial tool for the monitoring and optimization of doses at a particular institution (Vañó et al., 2017). The authors agree that merely determining these values is not enough and that LDRLs should serve as indicators for actions and measures that the medical team must implement in relation to radiation protection in the hospital environment.

5 Conclusion

The developed software was efficient in extracting data from the PACs image sent by the equipment evaluated, and the determined LDRLs for the main hemodynamic procedures in this work were coherent with literature. Determining the risk of stochastic effects is complex, given the varied geometries and complexities of the procedures. Thus, utilizing LDRLs emerges as an initial step in dose optimization. We concluded the methodology applied with the software is robust, and the tool has been successful in monitoring and optimizing doses at the evaluated institution. Notably, the Effective Dose calculated using the collected data agrees with the values described in previous studies. It is recognized that there is a continual need to update procedure data, ensuring that the institution’s reference values accurately represent current practices.

Acknowledgments

The authors would like to thank the Ministry of Health for the financial support for the Medical Physics Residency scholarships.

Funding

This research was funded by São Paulo Research Foundation (Process number: 2023/01156-6) and by Brazilian National Council for Scientific and Technological Development (Process number: 304992/2022-4).

Conflicts of interest

The authors declare no conficts of interest in regards to this article.

Data Availability Statement

All data generated or analyzed during this study are included in this published article. Additional data may be made available by the corresponding author upon reasonable request.

Author contribution statement

The authors of this work have contributed equally to its completion. Each member has brought unique expertise, dedication, and effort to the project, resulting in a collaborative and balanced effort.

Ethics approval

Approved by the Research Ethics Committee (CEP) under protocol: CAAE 16932513.5.0000.5411.

Informed Consent

In this study, we exclusively utilize DICOM images for analysis and research purposes. It is important to note that no personally identifiable information of the patients has been included or utilized in our work. Therefore, we consider informed consent unnecessary, given the strictly anonymous and non-invasive nature of our approach. Our focus lies in aggregate data analysis rather than individual participant identification. Nonetheless, we acknowledge the importance of privacy and ethics in medical research and adhere to all relevant regulations and guidelines in this domain.

References

All Tables

All Figures

	Fig 1 Example of a report provided by the equipment. Quantities such as kV, Air Kerma, KAP, number of exposures, exposure time and date of examination are made available in the form of an image.
In the text

Fig. 2 Flowchart representing the operation of the software for extracting quantities from the reports provided by the Artis Zee Ceiling equipment. (I) The first step is to determine the evaluation period for the procedures. (II) The reports provided by the equipment are extracted, added to the program’s image base and additional information is removed from the DICOM header associated with the procedure. (III) The Pytesseract library, in Python, is used to transcribe the data into the report. Information provided in the DICOM header, such as procedure name or patient ID, was extracted. (IV) The information extracted is reviewed by the authors. (V) The verified data is added to the database.

In the text

	Fig. 3 Boxplots represent four measurements from different medical procedures: (A) Dose (in mGy), (B) Kerma-Area Product (KAP in μGym2), (C) Time (min), and (D) Number of Exposures. The procedures are grouped along the x-axis with each boxplot illustrating the distribution of data for each variable.
In the text