Artificial Intelligence–Assisted Evaluation of Alara Compliance in Pediatric Chest Radiography

Artificial Intelligence–Assisted Evaluation of Alara Compliance in Pediatric Chest Radiography | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Artificial Intelligence–Assisted Evaluation of Alara Compliance in Pediatric Chest Radiography Gürbüz AKÇAY, Bünyamin GÜNEY This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9008392/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 14 You are reading this latest preprint version Abstract Background Pediatric radiology requires strict adherence to the ALARA principle (As Low As Reasonably Achievable), which refers to minimizing radiation exposure as much as possible, due to children’s high sensitivity to ionizing radiation. Objective This study aimed to develop an artificial intelligence–assisted system to evaluate compliance of pediatric chest radiographs with radiation safety principles. Material and Methods A total of 600 pediatric chest radiographs (aged 0–18 years) were collected and used to train a YOLO11n-based deep learning model for anatomical region and artifact detection. For ALARA compliance assessment, a separate test set of 4,461 chest radiographs was analyzed in batch mode. Image processing and metadata parsing were performed using the Pydicom and OpenCV libraries. Compliance with radiation safety principles was quantified on a 0–100 scale according to the BASICS criteria (Beam, Artifacts, Shielding, Immobilization & Indicators, Collimation, Structures). Results The model achieved a mean average precision (mAP50) of 0.995 for the chest class and 0.860 for artifacts. In the test set, the mean ALARA compliance score was 64.95. The highest compliance was observed in Structures (100% for necessary anatomy and 98.36% for devices/lines/tubes), along with the presence of side markers (99.69%) and immobilization (95.71%). In contrast, major deficiencies were identified in the tube angle score (0) and exposure indicator metrics (EI = 12.71 and DI = 12.71), followed by device alignment (41.13%) and collimation before exposure (56.91%). Conclusion The system provides objective and reproducible quality analysis, enabling continuous monitoring of pediatric chest radiography standards to support radiation safety. Physical sciences/Engineering Health sciences/Health care Health sciences/Medical research Radiation Safety Pediatric Radiology Artificial Intelligence Deep Learning Quality Control ALARA BASICS Figures Figure 1 Figure 2 Introduction X-ray imaging has been used in disease diagnosis for over a century (1). However, the harmful effects of X-rays on tissues have been known since their early medical use (2). These effects are classified as "deterministic" (short-term) and "stochastic" (long-term) (3). To minimize radiation exposure, the "As Low As Reasonably Achievable (ALARA)" principle is recommended (4, 5). BASICS standards (Beam, Artifact, Shielding, Immobilization/Indicators, Collimation, Structures) were developed by the "Image Gently" initiative to define radiographic safety (3). Previous BASICS-based evaluations in pediatric clinics have shown that chest radiographs are the most frequent direct radiographic examination (6). Artificial intelligence is used effectively in many medical fields today (7). Radiology is one such field. Recently, AI applications in chest radiographs have expanded beyond diagnostic decision support. They now address imaging quality, technical competence, and radiation safety. Schalekamp et al. described this as a three-stage transformation. The first stage covers disease classification and diagnostic algorithms. The second relates to triage and reporting support. The third, most recent stage includes systems for imaging quality and optimization (8). The ALARA-based system in this study fits this third phase. This study aimed to automate direct X-ray quality assessment using AI-assisted software. This approach enables radiologists, radiology professionals, and hospital managers to evaluate direct radiographs more efficiently and objectively, facilitating the identification of areas requiring improvement (9). Chest radiography is the most frequently performed examination in pediatric patients at the study institution; therefore, the research focused on these images (6). The developed method can be adapted for other types of direct radiography, such as limb or abdominal studies, in future applications. Highlighting these potential extensions demonstrates the method's scalability and encourages collaboration to adapt the technology for diverse imaging requirements. Materials and Methods Ethics Approval: This study was conducted in accordance with institutional and national ethical standards. Ethical approval was obtained from the local ethics committee prior to data collection (Date: June 24, 2025; Decision No: 12). This retrospective study used chest radiographs obtained from the institutional PACS archive of our hospital. Ethical approval for the study was granted by the Institutional Ethics Committee. Due to the retrospective nature of the study and the use of previously acquired imaging data, the requirement for informed consent was waived by the ethics committee. All metadata used for model training were anonymized prior to analysis. Data Sources and Study Cohorts The study dataset consisted of two distinct image groups serving different purposes within the proposed system. The first group included 600 pediatric chest radiographs (ages 0–18 years) available only in PNG format, which were used exclusively for training and validation of the YOLO11n-based object detection model. The second group comprised 4,461 pediatric chest PA radiographs in DICOM format, acquired from routine clinical imaging. This dataset contained complete acquisition metadata and was used for system-level automated quality evaluation and ALARA compliance analysis. Images with missing pixel data or unreadable DICOM headers were excluded from the analysis. To enhance system robustness, the software was designed to handle potential DICOM data omissions and model-level errors in an automated and fault-tolerant manner. The SpecificCharacterSet tag was automatically corrected to the ISO_IR 6 format in cases of incompatibility. When one or more exposure-related tags (EI, EIT, or DI) were missing, the corresponding criterion was evaluated with 0 points and explicitly documented in the report using the “obs” (observation) label. If TransferSyntaxUID was missing, decoding was attempted by sequentially trying common uncompressed transfer syntaxes. In cases where decoding was unsuccessful, an error message was generated and recorded. Pixel-based image quality metrics were not calculated for images lacking PixelData; instead, the result was added to the report with the descriptive status “image_status”. The YOLO11n detection model was executed in a fault-protected configuration, such that in the event of a model-related error, only the affected submodule was disabled while all other metric calculations continued uninterrupted. During dataset evaluation, data omissions—most commonly missing exposure index information—were observed in approximately 8% of cases, resulting in an estimated 5% reduction in overall system scoring robustness. This behavior demonstrates the high resilience of the proposed system in maintaining automated workflow continuity despite minor data loss. All images and DICOM metadata were processed locally only, and no patient identification information was transferred to external systems at any stage. Data processing procedures were fully compliant with the principles of the Personal Data Protection Law (KVKK) and the General Data Protection Regulation (GDPR). Software Environment and Processing Pipeline All analyses were performed using a Python-based software environment built on Miniconda infrastructure, including Python, pydicom, OpenCV, and the Ultralytics YOLO framework, running on a workstation equipped with an Intel i7-7700K processor, 32 GB RAM, and an NVIDIA GeForce GTX 1080 graphics card (10). The automated processing pipeline consisted of the following sequential steps: DICOM image loading and metadata extraction; pixel normalization and grayscale preprocessing; object detection using a YOLO11n-based model; feature extraction for BASICS criteria; score normalization and aggregation; and automated report generation in text, image, and spreadsheet formats. The software is compatible with Python version 3.9 and above. The core libraries required for execution include pydicom, NumPy, OpenCV-python, Ultralytics (YOLO11n), and OpenPyXL. All output files generated by the system (PNG, XLSX, and TXT formats) are archived in secure directories in accordance with institutional hospital policies, and access control mechanisms are implemented to ensure data security. Model weights "best.pt" and the complete software code base are maintained under a version control system, with all changes documented in a traceable manner. This versioning strategy supports both reproducibility and long-term quality monitoring. The developed system consists of three main components: alara.py Performs analysis of a single DICOM image and generates sample_report.txt (text-based evaluation report) and sample_annotated.png (annotated visual output). alara_folder.py : Performs batch evaluation of multiple images within a directory and saves the results to SAMPLES/alara_results.xlsx, which includes two worksheets: Results (detailed scores for each image) and Summary (average scores aggregated by folder). best.pt A YOLO11n-based deep learning model weight file containing two classes (0 = chest, 1 = artifact). This standardized file structure enables all output generated during single-image or batch analyses to be archived consistently and reused for longitudinal quality monitoring processes. best.pt A weight file of a YOLO11n-based deep learning model; it contains classes 0 = chest and 1 = artifact. Object Detection Model Development An object detection model was developed to identify key radiographic elements, including the chest region and obstructing artifacts. Manual annotations were created using the LabelMe software in accordance with the criteria defined by the European Commission (11, 12) (Fig. 1 ). Annotation files were converted from JSON format to YOLO-compatible TXT format, and the dataset was randomly split into training (80%), validation (10%), and test (10%) subsets. Model training was performed using YOLO11n architecture. The YOLO11n model "best.pt" used in the study was trained in two classes. Class 0 (Chest) The region representing chest anatomy was defined as a Region of Interest (ROI) and constituted the basis for centering, collimation, and structural coverage analyses within the BASICS framework. Class 1 (Artifact) This class aimed to identify artifacts (e.g., buttons, zippers, clothing fasteners, cables) that may adversely affect the diagnostic quality of the image. Prior to annotation, pixel data preprocessing was applied. RescaleSlope and RescaleIntercept corrections were used to convert raw DICOM pixel values into interpretable grayscale intensities. When required, the VOI LUT (Value of Interest Look-Up Table) function was applied for intensity mapping. Subsequently, images were converted to an 8-bit grayscale scale using a NumPy (version 2.x compatible) min–max normalization formula (out = (arr − min) / ptp × 255). Images in MONOCHROME1 format—where higher pixel values represent darker intensities—were inverted. All images were evaluated in this normalized PNG format prior to annotation. The basic DICOM metadata used in the study consisted of fields defining image type, orientation, and exposure parameters. The examination site was verified as CHEST/LUNG using the BodyPartExamined (0018,0015) field. A SpecificCharacterSet correction ('ISO IR 6' → 'ISO_IR 6') was applied when necessary. When Transfer Syntax UID (0002,0010) was missing or pixel decoding failed, the software applied a safe fallback strategy for pixel decoding; if decoding could not be completed (e.g., due to missing codes for compressed transfer syntaxes), pixel-based metrics were skipped, and the case was reported with an explicit image availability/status message. Technical acquisition parameters, including kVp (0018,0060), Exposure (mAs) (0018,1152), SID (0018,1110), Image Orientation Patient (0020,0037), and ViewPosition (0018,5101), were extracted and added to the reports together with relevant Study, Series, and Patient metadata fields. After model convergence, the best-performing weights were saved and used for all subsequent inference tasks within the BASICS-based scoring pipeline. BASICS-Based Methodological Framework The developed system produces normalized quality scores ranging from 0 to 100 across a total of 15 applicable sub-items, in accordance with the BASICS criteria. Although the system is capable of detecting shielding applications (gonadal and thyroid protection), this criterion was excluded from quantitative scoring and marked as not applicable (N/A) for chest radiographs, in alignment with the current “no-shielding” guideline (20)(13). The evaluated BASICS categories and their corresponding sub-items are summarized as follows: Beam: Anatomical centering, central beam alignment, tube angle, device alignment, and compatibility of exposure parameters (kVp and mAs). Artifacts: Presence of obstructive artifacts within the image and accuracy of the laterality (side) marker. Shielding: Detection of gonadal and thyroid protection (reported but excluded from scoring). Immobilization & Indicators: Motion control assessment, Exposure Indicator (EI), and Deviation Index (DI) compliance. Collimation was assessed using a proxy area-based metric: the ideal field was defined as a 10% expanded YOLO chest ROI, and the score was computed from the proportion of the image area outside this ideal region (no explicit segmentation of collimation borders). Structures: Complete visibility of required anatomical structures and medical devices, when applicable. The overall ALARA compliance score was calculated as the arithmetic mean of all applicable BASICS sub-criteria, excluding the Shielding category. To enhance system robustness, the software was designed to automatically manage potential DICOM metadata omissions and model-level errors. The SpecificCharacterSet tag was automatically corrected to the ISO_IR 6 format in cases of incompatibility. When exposure-related tags (EI, EIT, or DI) were missing, the corresponding criterion was assigned 0 points and explicitly documented using the “obs” (observation) label in the report. If the Transfer Syntax UID field was unavailable, the system attempted decoding using an Explicit VR Little Endian assumption; in the event of failure, an error message was generated and recorded. Pixel-based metrics were not calculated for images lacking PixelData, and these cases were reported with the descriptive status “image_status”. The YOLO11n detection model was executed in a fault-protected configuration, such that only the affected submodule was disabled in error states, while the computation of all other metrics continued uninterrupted. During dataset evaluation, metadata omissions—most commonly missing Exposure Index values—were observed in approximately 8% of cases, resulting in an estimated 5% reduction in overall system scoring robustness. This behavior demonstrates the high resilience of the proposed system in maintaining automated workflow continuity and functional integrity in the presence of minor data loss. Beam-related image quality assessment focused on centering accuracy, projection geometry, tube angulation, and exposure technique parameters. Anatomical centering was evaluated by calculating the normalized distance between the center of the detected chest Region of Interest (ROI) and the geometric center of the image. The centering score increased as the measured distance decreased; 100 points were awarded for perfect centering, while 0 points corresponded to the farthest deviation from the image center. Projection alignment and device orientation were assessed using DICOM Image Orientation Patient (IOP) vectors. Horizontal and vertical axis deviations were calculated from these vectors, and left–right (LR) and top–bottom (TB) edge symmetry were measured to quantify rotational misalignment. Device alignment analysis was performed using orientation axis vectors derived from the Image Orientation (Patient) (IOP) tag. Tube angulation was assessed using DICOM metadata to infer whether the acquisition corresponded to a standard or lordotic projection. In cases where angulation-related DICOM tags were absent, the tube angle criterion was assigned a score of 0. When relevant angulation metadata were available, tube angle appropriateness was evaluated; accordingly, otherwise, the parameter was recorded as unavailable. The central ray score was computed as a weighted composite of geometric proxies derived from DICOM angulation metadata (when available), SID proximity to a target value, left–right symmetry of the chest ROI margins, and the centering score. When angle/SID metadata were unavailable, the score relied on symmetry and centering only, and the report explicitly noted the missing tags. Exposure technique was assessed using DICOM metadata for tube voltage (kVp) and exposure (mAs). These parameters were compared with expected reference values based on patient age and projection type, and deviations resulted in proportional score reductions. A- Artifacts Artifact assessment included the detection of obstructing foreign objects and verification of laterality (side) markers. Obstructing artifacts such as clothing, monitoring cables, or external objects were detected using the trained YOLO11n model. Artifacts were penalized as 10 points per detection inside the chest ROI and 5 points per detection outside the ROI (floor at 0). Biological foreign objects were not annotated separately, as the current model focuses on removable non-biological artifacts that directly affect acquisition quality. An expected device list (e.g., endotracheal tube, nasogastric tube, central venous catheter, chest tube) was derived from DICOM free-text fields. Device/line visibility was then quantified using an image-based proxy: within the YOLO chest ROI, Canny edge detection followed by HoughLinesP was applied to detect line-like structures. The total detected line length was normalized by chest ROI height (norm_len), and the score was scaled on a 0–100 range based on this normalized length. If no device was expected from DICOM metadata, the criterion was scored as 100 to avoid unnecessary penalization. Laterality markers (L/R) were evaluated using a template matching approach applied to the four corner regions of the image (14). If a similarity score exceeding the predefined threshold was detected, the image was assigned 100 points for correct side marker presence; otherwise, a score of 0 points was assigned. To prevent false detections, the vertices corresponding to the chest bounding box were excluded from the template matching analysis. S- Shielding The system included the technical capability to detect protective shielding (gonadal and thyroid protection). However, shielding was not incorporated into the quantitative quality scoring process. In accordance with current pediatric radiography recommendations, this criterion was reported as not applicable (N/A) and excluded from the overall ALARA compliance calculations for chest x-rays (13). I- Immobilization & Indicators Image sharpness and motion artifacts were assessed using multiple image-based metrics , including Laplacian variance , Sobel-based edge detection , and frequency-domain analysis . Image sharpness was evaluated using a combined sharpness score derived from Laplacian variance , Tenengrad (Sobel size) , and edge density parameters. Each metric was normalized and averaged, and the resulting score was expressed on a 0–100 scale . Images with a score of ≥ 60 were classified as clear , scores between 40 and 60 as borderline , and scores < 40 as blurry. (15). This approach enables objective quantification of image blur. Motion analysis was performed by measuring the anisotropy of the angular profile obtained from the 2D Fast Fourier Transform (FFT) log-spectrum of the image. This metric quantitatively describes the directional dispersion of motion . A normalized no-motion score in the range of 0–100 was calculated by combining the anisotropy value with the acuity score and the gradient direction–orthogonal ratio . Exposure indicators were evaluated using the Deviation Index (DI) , calculated from Exposure Index (EI) values when available. If the DI tag was present in the DICOM data, it was used directly. In the absence of EI and Exposure Index Target (EIT) values, DI was calculated using the formula: DI = 10 × log₁₀(EI / EIT) The DI-based scoring function was defined as: Score = 100 − 20 × |DI| Accordingly, |DI| ≤ 1 was classified as ideal exposure , 1 3 as improper exposure . Images lacking exposure indicator metadata were assigned a zero score for this sub-criterion. This standardized approach minimizes variability in dose indication across different imaging device manufacturers. C- Collimation, Cropping Collimation evaluation was structured in accordance with the standards defined in the European Commission’s European Guidelines on Quality Criteria for Diagnostic Radiographic Images in Paediatrics (1996) . This approach is based on the concepts of minimum field size (MinFS) and maximum field size (MaxFS) described by Tschauner et al. In the present study, these concepts were digitized using a YOLO11n-based chest bounding box , allowing collimation excess, out-of-area irradiation, and centering to be converted into fully automated quantitative measures without the need for manual assessment (12, 16). The system considered a 10% expanded boundary around the YOLO11n-detected chest region as the recommended collimation zone (16). A score deduction was applied proportionally to the area of unnecessary irradiation extending beyond these limits, and the resulting value was normalized on a 0–100 scale . Collimation quality was further assessed by analyzing the spatial relationship between the detected chest region and the overall image field boundaries. Excessive irradiation outside the expected anatomical limits resulted in additional score penalties. Electronic clipping (cropping) was evaluated through analysis of pixel intensity distributions near the image borders and by reviewing relevant DICOM shutter-related metadata when available. Boundary regions of the image (default ± 5% band) exhibiting low standard deviation or high saturation were considered potential indicators of post-processing crops. When shutter or partial-view tags were present in the DICOM data, an additional point deduction was applied. The final collimation and cropping score was reported as a normalized value between 0 and 100 and visually indicated using a colored overlay on the output images. S- Structures Structural evaluation focused on anatomical coverage , assessed exclusively based on the chest region detected by the YOLO11n model . The system evaluated whether the essential thoracic anatomy was fully included within the image field by analyzing the spatial extent and position of the YOLO11n-detected chest Region of Interest (ROI) . Images in which the chest ROI was completely and appropriately contained within the image boundaries received full scores , while partial coverage resulted in proportional score reductions. Overall ALARA Compliance Score Each BASICS sub-criterion was normalized to a score between 0 and 100. The overall ALARA compliance score was calculated as the arithmetic mean of all applicable BASICS scores, excluding the shielding criterion. This structured and modular methodology enabled objective, reproducible, and automated assessment of pediatric chest radiograph quality. All metrics are normalized on a 0–100 scale according to the BASICS criteria. The system then calculates an average ALARA compliance score. In the final stage, the system generates the sample_report.txt file and the sample_annotated.png image for a single image. In aggregate analysis, the file alara_results.xlsx includes individual results and a general summary page. Limitations and Development Opportunities The system developed in this study is based on the YOLO11n model, which includes only the "chest" and "artifact" classes. For future development, expanding the model to incorporate additional subclasses, such as marker, tube, or line, is recommended to increase its applicability and improve its clinical relevance. Additionally, sharpness and motion assessments may be enhanced by integrating unreferenced quality metrics, such as BRISQUE or NIQA, which provide more comprehensive and objective evaluations of image quality. Additionally, EI/DI target ranges can be calibrated according to the device manufacturer's specifications and harmonized with dynamic criteria. Integrating anatomical landmark-based segmentation techniques into collimation analysis can also enhance the system's accuracy. Finally, it is planned to customize kVp/mAs target tables specific to enterprise protocols over configurable JSON or YAML files. Results Demographics A total of 600 chest X-rays belonging to 555 children were included in the study. More than one radiograph of some cases was evaluated. The ages of the participants ranged from 0 to 18 years, with an average age of 5.40 ± 5.02 years. Of the cases, 53.8% were male, and 46.2% were female. AI Metrics Model training was performed in an environment with CUDA acceleration enabled for 100 epochs at a resolution of 640 × 640 pixels. 80% (480 images) of the training data is allocated for training, 10% (60 images) for validation, and 10% for testing. The training process was completed in approximately 17 minutes. The model's test performance metrics are summarized in Table 1 . Table 1 Model Performance Class Precision Recall mAP50 mAP@50–95 Chest 0.978 1.000 0.995 0.870 Artifact 0.861 0.738 0.860 0.525 Average 0.919 0.869 0.927 0.697 Ultralytics 8.3.96, Python-3.12.9 torch-2.6.0 + cu118 CUDA:0 (NVIDIA GeForce GTX 1080, 8192 MiB). YOLO11n summary (fused): 100 layers, 2,582,542 parameters, 0 gradients, 6.3 GFLOPs. Speed: 0.2 ms preprocess, 2.9 ms inference, 0.0 ms loss, 1.2 ms postprocess per image. The accuracy of the YOLO11n model in the chest class is remarkably high (mAP50 = 0.995), indicating that the anatomical area is determined almost flawlessly. The relatively low accuracy and sensitivity rates in the Artifact class (Precision = 0.861, Recall = 0.738) indicate limited performance in detecting clothing elements or small objects that are obscured by movement. Single image evaluation Using the obtained YOLO11n model and the methodology described above, the evaluation of a sample.dcm file was performed. In this sample analysis, patient identification information was hidden or altered. Figure 2 presents an example of the annotated image created by the software. The mean overall score was found to be 64.40 in the evaluation of single samples. The highest scores were obtained in the kVp suitability (100) and Structures (100) criteria; the lowest scores were recorded in mAs suitability (0). This suggests that the lack of exposure labels in the dataset and operator habits are key determinants of ALARA compliance (Table 2 ). Table 2 Summary of Automated BASICS Assessment Results for a Single Image === BASICS (CHEST) Automatic Assessment === Overall score (mean of items): 64.40 [Beam] - Anatomy is centered in the image: 58.94 - Central ray appropriate for projection: 48.26 • OBS: View = NA; angP = NA, angS = NA, SID = 0 mm; sym = 82.3, center = 58.9 - Tube angle appropriate: 0.00 • OBS: mode=standard; target = 0 ± 10°; no angle tags - Device alignment appropriate: 41.17 • OBS: no IOP tag; LR = 82.3, TB = 0.0 - kVp appropriate for projection/patient: 100.00 • OBS: profile=pediatric/AP/PA [60–90] kVp; value = 90.0 (in-range); age=1y - mAs appropriate for projection/patient: 0.00 • OBS: profile=pediatric/AP/PA [0.5-3.0] mAs; value = 0.00 (below); age=1y [Artifacts] - No obstructing artifacts present (e.g., lead, clothing, ECG cables): 90.00 • OBS: in = 0, out = 2 - Side marker present and correct: 100.00 • OBS: found = L @ ML (score = 0.46) [Shielding] - Gonadal/thyroid shielding used appropriately (when indicated): N/A • OBS: N/A (Shielding not required for Chest) [Immobilization & Indicators] - Patient cooperation/immobilization adequate (no motion): 81.17 • OBS: aniso = 0.22, dir=0deg; terms: aniso = 0.29, blur = 0.05, dirR = 1.68 - Exposure Indicator (EI) within target range: 56.00 • OBS: EI = 170.0, EIT = 280.0, DI=-2.20 (acceptable; DI tag) - Deviation Index (DI) close to 0 (ideal): 56.00 • OBS: EI = 170.0, EIT = 280.0, DI=-2.20 (acceptable; DI tag) [Collimation] - Field is collimated to area of interest BEFORE exposure: 34.38 • OBS: unnecessary = 65.6%, useful = 34.4% - No reliance on electronic cropping after exposure: 100.00 • OBS: tags=[none]; suspicious_border = 0.0% [Structures] - Necessary anatomy fully demonstrated: 100.00 • OBS: chest bbox fully inside image - Devices/lines/tubes (if any) properly demonstrated: 100.00 • OBS: no device expected by tags; lines_detected = 1, norm_len = 0.14 --- Özet --- --- Summary --- PNG: sample.png Annotasyon: sample_annotated.png Assessment time: 3.54 seconds Batch Evaluation. For clinical validation of the system, a test set of 4461 images was initially prepared. The automatic score was calculated for each image, and the average results are presented in Table 3 . The automated evaluation process took approximately 5.6 hours and produced continuous, uninterrupted reporting throughout the run. The strongest performance was observed in the Structures criteria—Necessary anatomy fully demonstrated (100) and Devices/lines/tubes properly demonstrated (98.36)—together with Side Marker present and correct (99.69) and Patient cooperation/immobilization adequate (95.71). The Shielding criterion was excluded from the calculation (N/A) in accordance with updated radiation safety guidelines. In contrast, the lowest values were recorded for Tube Angle appropriate (0) and for the Exposure Indicator (EI) within target range (12.71), and Deviation Index (DI) close to 0 (12.71), followed by Device Alignment (41.13) and Collimation before exposure (56.91). Overall, the markedly low EI/DI scores indicate that missing parameters and technical/metadata inconsistencies in DICOM headers can directly reduce ALARA-based scoring performance. The zero tube angle score reflects the absence of angulation-related metadata rather than incorrect patient positioning. Table 3 Batch analysis scores (n = 4461). Metric Average score Beam | Anatomy is centered in the image (score) 82.12 Beam | Central ray appropriate for projection (score) 67.15 Beam | Tube angle appropriate (score) 0 Beam | Device alignment appropriate (score) 41.13 Beam | kVp appropriate for projection/patient (score) 63.45 Beam | mAs appropriate for projection/patient (score) 60.83 Artifacts | No obstructing artifacts present (e.g. lead. clothing. ECG cables) (score) 92.52 Artifacts | Side marker present and correct (score) 99.69 Shielding | Gonadal/thyroid shielding used appropriately (when indicated) (score) N/A Immobilization & Indicators | Patient cooperation/immobilization adequate (no motion) (score) 95.71 Immobilization & Indicators | Exposure Indicator (EI) within target range (score) 12.71 Immobilization & Indicators | Deviation Index (DI) close to 0 (ideal) (score) 12.71 Collimation | Field is collimated to area of interest BEFORE exposure (score) 56.91 Collimation | No reliance on electronic cropping after exposure (score) 90.95 Structures | Necessary anatomy fully demonstrated (score) 100 Structures | Devices/lines/tubes (if any) properly demonstrated (score) 98.36 Overall Score 64.95 Discussion Overall rating The ALARA-based artificial intelligence system developed in this study is an end-to-end solution that performs automatic quality assessment of pediatric chest X-rays. Based on the BASICS criteria, the system evaluates each quality indicator with quantitative scores and generates an overall ALARA compliance score. The most important stage before a chest X-ray is the proper determination of the indication; The next stage is optimization (12). Thus, it may be possible to achieve the principle of "As Low As Reasonably Achievable" (ALARA) without compromising diagnostic accuracy (13). For this purpose, necessary regulations have been established by the legislators of many countries (17). In the study by Abubakr et al., it was demonstrated that expert inspections based on nine different acquisition criteria resulted in a significant improvement in subsequent imaging (18). However, the increasing number of examinations and intensive workflow in modern healthcare make manual examination of these examinations time-consuming and subjective. Granata et al. reported that 20–50% of the examinations performed in pediatric radiology had indication eligibility problems (13). This finding underscores the need for automated inspection mechanisms to optimize radiation. The ALARA-based software proposed in our study represents an innovative approach that addresses this requirement. The system objectively applies the optimization principle by numerically evaluating collimation and anatomical coverage rates. The European Commission's Radiation Protection No. 162 defines the parameters of "anatomical coverage, collimation, centering, and acuity" for evaluating diagnostic quality in pediatric chest X-rays (12). These standards establish the minimum quality requirements for applying ALARA principles in the field. In our study, these criteria were automatically analyzed through artificial intelligence-assisted software. Recent artificial intelligence studies show that routine chest X-rays can be used not only for diagnostic purposes, but also for opportunistic analysis (e.g., osteoporosis risk prediction) (19). Similarly, this study expanded the role of artificial intelligence in radiography to a "beyond diagnosis" field by using pediatric chest X-rays in automated quality and radiation safety assessment. Thus, the proposed system reduces the manual inspection burden on radiologists and provides an objective and repeatable quality control mechanism. Although there are more than forty licensed artificial intelligence software programs for chest X-ray and tomography analysis today, most of them focus on the adult population (8). B – Beam (Dosing and Centering) Dosing parameters are one of the most critical quality determinants in pediatric radiology. In our study, it was observed that kVp values were mostly in the recommended range of 60–90, but mAs values were understated or under-recorded. This is thought to be related to the lack of DICOM tags on some devices. Beam centering score of 82.12 indicates that positioning is more challenging in pediatric patients than in adults (5). A – Artifacts (Foreign Objects and Side Signs) The presence of biological foreign objects (BFO) or non-biological foreign objects (NBFO) on chest X-rays may complicate the diagnosis of pathologies such as fluid accumulation, tuberculosis, or cysts (20). Santosh et al. (24)(21) achieved precision, recall, and F1-score values of 0.85, 0.93, and 0.89, respectively, on 400 images using the BFO/NBFO detection system they developed with the YOLOv4 algorithm. Zohora et al. proposed a method for detecting foreign circular objects in chest X-ray images using Sobel, Canny, Prewitt, and Roberts edge detectors, followed by morphological operations and a circular Hough transform, and reported high detection accuracy and computational efficiency compared to existing methods (22). In our model, only artifacts that fall into the NBFO class are labeled. The performance criteria of the model for this class were calculated as follows: precision, 0.861; recall, 0.738; mAP@50, 0.860; and mAP@50–95, 0.525. The results show that the model identifies artifacts with a large percentage of accuracy, but has limited sensitivity in small, low-contrast objects. Side marker signs (“L”/“R”) achieved an excellent score (99.69), suggesting strong operator compliance with laterality labeling (8). S – Shielding In our study, this item was excluded from the quantitative scoring and marked as 'N/A' (Not Applicable). As Granata et al. point out, the routine use of contact shielding in pediatric radiology has become controversial today due to its limited benefits and potential risks (13). Consensus reports from AAPM and European radiological societies (ESR, ESPR) emphasize that protectors carry the risk of covering anatomical details, are difficult to place correctly, and can paradoxically increase the dose to which the patient is exposed by misleading Automatic Exposure Control (AEC) systems. In this context, although the software we have developed scores the presence of shielding as a 'positive' quality criterion with its current algorithm, due to this paradigm shift in radiation safety culture, it would be a more clinically correct approach to revise this criterion according to the principle of 'no-shielding' in future versions of the system and to accept the non-use of shielding as a standard. Our aim in maintaining this scoring can be explained as the adequacy of our study for use in areas that require protection, such as the pelvis. I – Immobilization & Indicators The high score of 95.71 points in the motion analysis indicates that the children were immobilized with appropriate fixation methods during the shoot. Moore et al. emphasized that digital radiography may mask overexposure, resulting in unnecessary dose escalation (23). To prevent this, it is recommended to use the EI and DI index values (5). The fact that EI and DI labels were largely missing in our study was the main reason for the low scores (23). This deficiency is attributed to the non-standard data recording of the devices, highlighting a limitation in data quality. C – Collimation The automatic collimation assessment module is structured based on the field definitions in the European Commission's Radiation Protection No.162 guideline. The 10±% safety margin added to the chest area detected by YOLO11n represents the recommended MinFS–MaxFS (Minimum-Maximum Area) range for pediatric chest X-rays. In their study analyzing the European guidelines, Tschauner et al. stated that the ideal area should cover the T2–L1 vertebral spacing and defined age-related tolerance limits of 10–20 mm. In the same study, the average overexposure exceeding the guideline limits was reported as 45.1% (16). Using a similar methodology, Pedersen et al. demonstrated that U-Net-based AI models can detect lung segmentation and collimation boundaries in high agreement with radiologists (Dice score > 0.93) (24). In our study, the average fitness score for collimation remained as low as 56.91 in the test set of 4,461 images. These findings demonstrate that manual collimation is insufficient in clinical practice, and our automated assessment module makes a significant contribution to ensuring sustainable radiation safety in pediatric radiography by reducing this risk. S – Structures (Displayed Anatomical Structures) Pedersen and his team, utilizing the U-Net architecture, achieved automatic collimation, resulting in a Dice value of 0.95 in boundary segmentation. In contrast, our YOLO11n-based approach yielded similarly accurate results, facilitating faster integration into DICOM-based quality control lines (24). Mouton et al. achieved an accuracy of AUC = 0.782 in the classical CAD system (27)(25). These results demonstrate the level of accuracy of deep learning-based models. In our system, the value mAP@50 = 0.995 represents the transition from the classical CAD era to today's fully automated ALARA inspection systems. In addition, Kufel et al. in the YOLOv8-based study, a precision of 0.815 was reported in 112,120 chest X-rays (26). In our model, chest ROI was defined with 99% accuracy, with a score of 100 for the Structures criteria. This result shows that anatomical coverage inspection with artificial intelligence can be successfully integrated into clinical practice. Limitations This study enabled the development of a fully automated quality control process for pediatric chest X-rays. However, it is an important limitation that all criteria are given equal weight in the general scoring. In future studies, the criteria may be weighted based on their clinical significance, ensuring a fairer overall score. Conclusion This study integrated the DICOM infrastructure, image processing, and deep learning components to provide an objective and reproducible evaluation of pediatric chest radiographs within the framework of BASICS. The developed batch analysis flow enables monitoring, comparison, and continuous improvement of quality indicators throughout the organization. As a result, the system performed the evaluation based on ALARA principles in pediatric radiographs in an automated, standardized, and reproducible manner, in accordance with the principles outlined in European Guidelines RP 162 (1996) and the Image Gently initiative (2012) (5, 16). Declarations Funding No funding was received for this study. Author Contribution Author Contributions Statement:G.A. and B.G. jointly developed the study concept and design. G.A. was responsible for collection of pediatric chest radiographs, case assembly, development and coding of the artificial intelligence–based system, image processing workflows, and batch analysis. B.G. performed image annotation, reference labeling according to ALARA/BASICS criteria, radiological quality assessment, and clinical validation of the model outputs. G.A. and B.G. contributed to the interpretation of the results. All authors reviewed, revised, and approved the final manuscript. Data Availability The datasets generated and/or analysed during the current study are not publicly available due to patient privacy concerns and institutional data protection regulations. Some radiographic images contain patient identifiers embedded within the image. However, de-identified versions of the images may be made available from the corresponding author upon reasonable request for research purposes. References Rontgen WC. On a New Kind of Rays. Science. 1896;3(59):227-31. Daniel J. The X-rays. Science. 1896;3(67):562-3. Wagner LK, Eifel PJ, Geise RA. Potential biological effects following high X-ray dose interventional procedures. J Vasc Interv Radiol. 1994;5(1):71-84. Willis CE, Slovis TL. The ALARA concept in pediatric CR and DR: dose reduction in pediatric radiographic exams--a white paper conference executive summary. Pediatr Radiol. 2004;34 Suppl 3:S162-4. Don S, Goske MJ, John S, Whiting B, Willis CE. Image Gently pediatric digital radiography summit: executive summary. Pediatr Radiol. 2011;41(5):562-5. Akçay G, Güney B, Deveer M, Topal Y. A qualitative evaluation of direct radiography in pediatric clinics. Sağlık Akademisyenleri Dergisi. 2016;3(3):100-5. Gore JC. Artificial intelligence in medical imaging. Elsevier; 2020. p. A1-A4. Schalekamp S, Klein WM, van Leeuwen KG. Current and emerging artificial intelligence applications in chest imaging: a pediatric perspective. Pediatric Radiology. 2022;52(11):2120-30. Al-Naser YA. The impact of artificial intelligence on radiography as a profession: A narrative review. Journal of Medical Imaging and Radiation Sciences. 2023;54(1):162-6. Anaconda. Miniconda – Getting Started 2025 [Available from: https://www.anaconda.com/docs/getting-started/miniconda/main. Wada K. Labelme – Create your image dataset for vision AI 2025 [Available from: https://labelme.io/. European Commission. European guidelines on quality criteria for diagnostic radiographic images in paediatrics. Luxembourg: Office for Official Publications of the European Communities; 1996. Granata C, Sofia C, Francavilla M, Kardos M, Kasznia-Brown J, Nievelstein RA, et al. Let’s talk about radiation dose and radiation protection in children. Pediatric radiology. 2025;55(3):386-96. Arimura H, Ishida T, Katsuragawa S, Kawashita I, Doi K. [Development of a computerized method for identifying view position and orientation for chest radiographs by using a template matching technique]. Nihon Hoshasen Gijutsu Gakkai Zasshi. 2002;58(8):1047-54. Kim C-M, Hong EJ, Park RC. Chest X-ray outlier detection model using dimension reduction and edge detection. IEEE Access. 2021;9:86096-106. Tschauner S, Marterer R, Gübitz M, Kalmar PI, Talakic E, Weissensteiner S, et al. European Guidelines for AP/PA chest X-rays: routinely satisfiable in a paediatric radiology division? Eur Radiol. 2016;26(2):495-505. Council of the European Union. Council Directive 2013/59/Euratom of 5 December 2013 laying down basic safety standards for protection against the dangers arising from exposure to ionising radiation. Directive ed: European Union; 2013. p. 1-73. Muhammed AEM, Awad MSA, Saeed MM, Alkheder MA, Babiker MAM, Alzain MAA, et al. Adapted Anatomical Image Criteria for PA Chest Radiographs at Managil Teaching Hospital, Sudan 2023. Clinical Audit. 2024:63-8. Smith AD, Rothenberg SA. AI and Chest Radiographs: A Dawning Era in Osteoporosis Screening. Radiological Society of North America; 2024. p. e241339. Roy S, Santosh KC. Analyzing Overlaid Foreign Objects in Chest X-rays-Clinical Significance and Artificial Intelligence Tools. Healthcare (Basel). 2023;11(3):308. Santosh K, Roy S, Allu S. Generic Foreign Object Detection in Chest X-rays. International Conference on Recent Trends in Image Processing and Pattern Recognition; Cham: Springer International Publishing; 2022. p. 93-104. Zohora FT, Santosh K. Circular foreign object detection in chest x-ray images. International Conference on Recent Trends in Image Processing and Pattern Recognition. 2016:391-401. Moore QT, Don S, Goske MJ, Strauss KJ, Cohen M, Herrmann T, et al. Image gently: using exposure indicators to improve pediatric digital radiography. Radiol Technol. 2012;84(1). Pedersen A, Kusk MW, Knudsen G, Busk C, Lysdahlgaard S. Collimation border with U-Net segmentation on chest radiographs compared to radiologists. Radiography. 2023;29(3):647-52. Mouton A, Pitcher RD, Douglas TS. Computer-Aided Detection of Pulmonary Pathology in Pediatric Chest Radiographs. International Conference on Medical Image Computing and Computer-Assisted Intervention. 2010:619-25. Kufel J, Bargieł-Łączek K, Koźlik M, Czogalik Ł, Dudek P, Magiera M, et al. Chest X-ray foreign objects detection using artificial intelligence. Journal of Clinical Medicine. 2023;12(18):5841. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 14 May, 2026 Reviews received at journal 14 May, 2026 Reviews received at journal 10 May, 2026 Reviewers agreed at journal 09 May, 2026 Reviewers agreed at journal 04 May, 2026 Reviews received at journal 04 May, 2026 Reviewers agreed at journal 17 Apr, 2026 Reviewers agreed at journal 12 Apr, 2026 Reviewers agreed at journal 12 Apr, 2026 Reviewers invited by journal 10 Apr, 2026 Editor assigned by journal 10 Apr, 2026 Editor invited by journal 18 Mar, 2026 Submission checks completed at journal 14 Mar, 2026 First submitted to journal 13 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9008392","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":623408351,"identity":"fbbc6c64-08a4-4e5c-a3b7-f659fff28afd","order_by":0,"name":"Gürbüz AKÇAY","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYHACNhBhAMSMDyACCcRrYTaAazlApBY2CaK0yEfkPnvM88fOmL/9jFl1Qc1hBn72HAPmj3twazG8kW5uzMOTbCZxJsfs9oxjhxkke94YMBx4hkfLjDQ2aR4JZhsDBqAWHrbDDAY3coBa8LgMosWg3saA/41ZMc+/wwz2hLTIS4C0JBw2M5DIMWPmbQPaIkFAiwHPM3bDOQeOG0vceFYsPbMvnUfizLOCA2fw2dKexvbgzZ9qw/7+5I2fC75Zy/G3J298UIHPFoQchwEzkOQBMfFoANrSAGeyP2DGp3IUjIJRMApGLgAADdVNUexJq/kAAAAASUVORK5CYII=","orcid":"","institution":"Pamukkale University","correspondingAuthor":true,"prefix":"","firstName":"Gürbüz","middleName":"","lastName":"AKÇAY","suffix":""},{"id":623408352,"identity":"dba85196-afc3-4359-a292-e07c08bb9795","order_by":1,"name":"Bünyamin GÜNEY","email":"","orcid":"","institution":"Muğla University","correspondingAuthor":false,"prefix":"","firstName":"Bünyamin","middleName":"","lastName":"GÜNEY","suffix":""}],"badges":[],"createdAt":"2026-03-02 09:39:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9008392/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9008392/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107257805,"identity":"9c06e99d-0b5b-4241-97f8-033829bce9f7","added_by":"auto","created_at":"2026-04-19 12:35:03","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":651311,"visible":true,"origin":"","legend":"\u003cp\u003eLabeling of chest and artifact classes using LabelMe software.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-9008392/v1/33200c72602a8c88eb9d2062.png"},{"id":107257806,"identity":"400520d8-b180-43b8-aa47-01b2665e8d45","added_by":"auto","created_at":"2026-04-19 12:35:03","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2663735,"visible":true,"origin":"","legend":"\u003cp\u003eAn image automatically marked by software.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-9008392/v1/e4cca960c61bb670873281b3.png"},{"id":107484816,"identity":"a383c39d-5456-411c-b955-b69ae969a482","added_by":"auto","created_at":"2026-04-22 02:33:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4156953,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9008392/v1/020feff6-fb20-4f18-904d-c7690696b3a6.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eArtificial Intelligence–Assisted Evaluation of Alara Compliance in Pediatric Chest Radiography\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eX-ray imaging has been used in disease diagnosis for over a century (1). However, the harmful effects of X-rays on tissues have been known since their early medical use (2). These effects are classified as \"deterministic\" (short-term) and \"stochastic\" (long-term) (3). To minimize radiation exposure, the \"As Low As Reasonably Achievable (ALARA)\" principle is recommended (4, 5). BASICS standards (Beam, Artifact, Shielding, Immobilization/Indicators, Collimation, Structures) were developed by the \"Image Gently\" initiative to define radiographic safety (3). Previous BASICS-based evaluations in pediatric clinics have shown that chest radiographs are the most frequent direct radiographic examination (6).\u003c/p\u003e \u003cp\u003eArtificial intelligence is used effectively in many medical fields today (7). Radiology is one such field. Recently, AI applications in chest radiographs have expanded beyond diagnostic decision support. They now address imaging quality, technical competence, and radiation safety. Schalekamp et al. described this as a three-stage transformation. The first stage covers disease classification and diagnostic algorithms. The second relates to triage and reporting support. The third, most recent stage includes systems for imaging quality and optimization (8). The ALARA-based system in this study fits this third phase.\u003c/p\u003e \u003cp\u003eThis study aimed to automate direct X-ray quality assessment using AI-assisted software. This approach enables radiologists, radiology professionals, and hospital managers to evaluate direct radiographs more efficiently and objectively, facilitating the identification of areas requiring improvement (9). Chest radiography is the most frequently performed examination in pediatric patients at the study institution; therefore, the research focused on these images (6). The developed method can be adapted for other types of direct radiography, such as limb or abdominal studies, in future applications. Highlighting these potential extensions demonstrates the method's scalability and encourages collaboration to adapt the technology for diverse imaging requirements.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eEthics Approval:\u003c/h2\u003e \u003cp\u003e This study was conducted in accordance with institutional and national ethical standards. Ethical approval was obtained from the local ethics committee prior to data collection (Date: June 24, 2025; Decision No: 12).\u003c/p\u003e \u003cp\u003e This retrospective study used chest radiographs obtained from the institutional PACS archive of our hospital. Ethical approval for the study was granted by the Institutional Ethics Committee. Due to the retrospective nature of the study and the use of previously acquired imaging data, the requirement for informed consent was waived by the ethics committee. All metadata used for model training were anonymized prior to analysis.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Sources and Study Cohorts\u003c/h3\u003e\n\u003cp\u003eThe study dataset consisted of two distinct image groups serving different purposes within the proposed system. The first group included 600 pediatric chest radiographs (ages 0\u0026ndash;18 years) available only in PNG format, which were used exclusively for training and validation of the YOLO11n-based object detection model.\u003c/p\u003e \u003cp\u003eThe second group comprised 4,461 pediatric chest PA radiographs in DICOM format, acquired from routine clinical imaging. This dataset contained complete acquisition metadata and was used for system-level automated quality evaluation and ALARA compliance analysis. Images with missing pixel data or unreadable DICOM headers were excluded from the analysis.\u003c/p\u003e \u003cp\u003eTo enhance system robustness, the software was designed to handle potential DICOM data omissions and model-level errors in an automated and fault-tolerant manner. The SpecificCharacterSet tag was automatically corrected to the ISO_IR 6 format in cases of incompatibility. When one or more exposure-related tags (EI, EIT, or DI) were missing, the corresponding criterion was evaluated with 0 points and explicitly documented in the report using the \u0026ldquo;obs\u0026rdquo; (observation) label.\u003c/p\u003e \u003cp\u003eIf TransferSyntaxUID was missing, decoding was attempted by sequentially trying common uncompressed transfer syntaxes. In cases where decoding was unsuccessful, an error message was generated and recorded. Pixel-based image quality metrics were not calculated for images lacking PixelData; instead, the result was added to the report with the descriptive status \u0026ldquo;image_status\u0026rdquo;.\u003c/p\u003e \u003cp\u003eThe YOLO11n detection model was executed in a fault-protected configuration, such that in the event of a model-related error, only the affected submodule was disabled while all other metric calculations continued uninterrupted. During dataset evaluation, data omissions\u0026mdash;most commonly missing exposure index information\u0026mdash;were observed in approximately 8% of cases, resulting in an estimated 5% reduction in overall system scoring robustness. This behavior demonstrates the high resilience of the proposed system in maintaining automated workflow continuity despite minor data loss.\u003c/p\u003e \u003cp\u003eAll images and DICOM metadata were processed locally only, and no patient identification information was transferred to external systems at any stage. Data processing procedures were fully compliant with the principles of the Personal Data Protection Law (KVKK) and the General Data Protection Regulation (GDPR).\u003c/p\u003e\n\u003ch3\u003eSoftware Environment and Processing Pipeline\u003c/h3\u003e\n\u003cp\u003eAll analyses were performed using a Python-based software environment built on Miniconda infrastructure, including Python, pydicom, OpenCV, and the Ultralytics YOLO framework, running on a workstation equipped with an Intel i7-7700K processor, 32 GB RAM, and an NVIDIA GeForce GTX 1080 graphics card (10). The automated processing pipeline consisted of the following sequential steps:\u003c/p\u003e \u003cp\u003eDICOM image loading and metadata extraction; pixel normalization and grayscale preprocessing; object detection using a YOLO11n-based model; feature extraction for BASICS criteria; score normalization and aggregation; and automated report generation in text, image, and spreadsheet formats. The software is compatible with Python version 3.9 and above. The core libraries required for execution include pydicom, NumPy, OpenCV-python, Ultralytics (YOLO11n), and OpenPyXL. All output files generated by the system (PNG, XLSX, and TXT formats) are archived in secure directories in accordance with institutional hospital policies, and access control mechanisms are implemented to ensure data security.\u003c/p\u003e \u003cp\u003eModel weights \"best.pt\" and the complete software code base are maintained under a version control system, with all changes documented in a traceable manner. This versioning strategy supports both reproducibility and long-term quality monitoring.\u003c/p\u003e \u003cp\u003eThe developed system consists of three main components:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ealara.py\u003c/strong\u003e \u003cp\u003ePerforms analysis of a single DICOM image and generates sample_report.txt (text-based evaluation report) and sample_annotated.png (annotated visual output).\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ealara_folder.py\u003c/b\u003e: Performs batch evaluation of multiple images within a directory and saves the results to SAMPLES/alara_results.xlsx, which includes two worksheets: Results (detailed scores for each image) and Summary (average scores aggregated by folder).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ebest.pt\u003c/strong\u003e \u003cp\u003eA YOLO11n-based deep learning model weight file containing two classes (0\u0026thinsp;=\u0026thinsp;chest, 1\u0026thinsp;=\u0026thinsp;artifact). This standardized file structure enables all output generated during single-image or batch analyses to be archived consistently and reused for longitudinal quality monitoring processes.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003ebest.pt\u003c/strong\u003e \u003cp\u003eA weight file of a YOLO11n-based deep learning model; it contains classes 0\u0026thinsp;=\u0026thinsp;chest and 1\u0026thinsp;=\u0026thinsp;artifact.\u003c/p\u003e \u003c/p\u003e\n\u003ch3\u003eObject Detection Model Development\u003c/h3\u003e\n\u003cp\u003eAn object detection model was developed to identify key radiographic elements, including the chest region and obstructing artifacts. Manual annotations were created using the LabelMe software in accordance with the criteria defined by the European Commission (11, 12) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Annotation files were converted from JSON format to YOLO-compatible TXT format, and the dataset was randomly split into training (80%), validation (10%), and test (10%) subsets.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eModel training was performed using YOLO11n architecture. The YOLO11n model \"best.pt\" used in the study was trained in two classes.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eClass 0 (Chest)\u003c/strong\u003e \u003cp\u003eThe region representing chest anatomy was defined as a Region of Interest (ROI) and constituted the basis for centering, collimation, and structural coverage analyses within the BASICS framework.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eClass 1 (Artifact)\u003c/strong\u003e \u003cp\u003eThis class aimed to identify artifacts (e.g., buttons, zippers, clothing fasteners, cables) that may adversely affect the diagnostic quality of the image.\u003c/p\u003e \u003c/p\u003e \u003cp\u003ePrior to annotation, pixel data preprocessing was applied. RescaleSlope and RescaleIntercept corrections were used to convert raw DICOM pixel values into interpretable grayscale intensities. When required, the VOI LUT (Value of Interest Look-Up Table) function was applied for intensity mapping. Subsequently, images were converted to an 8-bit grayscale scale using a NumPy (version 2.x compatible) min\u0026ndash;max normalization formula (out = (arr\u0026thinsp;\u0026minus;\u0026thinsp;min) / ptp \u0026times; 255).\u003c/p\u003e \u003cp\u003eImages in MONOCHROME1 format\u0026mdash;where higher pixel values represent darker intensities\u0026mdash;were inverted. All images were evaluated in this normalized PNG format prior to annotation.\u003c/p\u003e \u003cp\u003eThe basic DICOM metadata used in the study consisted of fields defining image type, orientation, and exposure parameters. The examination site was verified as CHEST/LUNG using the BodyPartExamined (0018,0015) field. A SpecificCharacterSet correction ('ISO IR 6' \u0026rarr; 'ISO_IR 6') was applied when necessary. When Transfer Syntax UID (0002,0010) was missing or pixel decoding failed, the software applied a safe fallback strategy for pixel decoding; if decoding could not be completed (e.g., due to missing codes for compressed transfer syntaxes), pixel-based metrics were skipped, and the case was reported with an explicit image availability/status message.\u003c/p\u003e \u003cp\u003eTechnical acquisition parameters, including kVp (0018,0060), Exposure (mAs) (0018,1152), SID (0018,1110), Image Orientation Patient (0020,0037), and ViewPosition (0018,5101), were extracted and added to the reports together with relevant Study, Series, and Patient metadata fields.\u003c/p\u003e \u003cp\u003eAfter model convergence, the best-performing weights were saved and used for all subsequent inference tasks within the BASICS-based scoring pipeline.\u003c/p\u003e\n\u003ch3\u003eBASICS-Based Methodological Framework\u003c/h3\u003e\n\u003cp\u003e The developed system produces normalized quality scores ranging from 0 to 100 across a total of 15 applicable sub-items, in accordance with the BASICS criteria. Although the system is capable of detecting shielding applications (gonadal and thyroid protection), this criterion was excluded from quantitative scoring and marked as not applicable (N/A) for chest radiographs, in alignment with the current \u0026ldquo;no-shielding\u0026rdquo; guideline (20)(13).\u003c/p\u003e \u003cp\u003eThe evaluated BASICS categories and their corresponding sub-items are summarized as follows:\u003c/p\u003e \u003cp\u003eBeam: Anatomical centering, central beam alignment, tube angle, device alignment, and compatibility of exposure parameters (kVp and mAs).\u003c/p\u003e \u003cp\u003eArtifacts: Presence of obstructive artifacts within the image and accuracy of the laterality (side) marker.\u003c/p\u003e \u003cp\u003eShielding: Detection of gonadal and thyroid protection (reported but excluded from scoring).\u003c/p\u003e \u003cp\u003eImmobilization \u0026amp; Indicators: Motion control assessment, Exposure Indicator (EI), and Deviation Index (DI) compliance. Collimation was assessed using a proxy area-based metric: the ideal field was defined as a 10% expanded YOLO chest ROI, and the score was computed from the proportion of the image area outside this ideal region (no explicit segmentation of collimation borders). Structures: Complete visibility of required anatomical structures and medical devices, when applicable. The overall ALARA compliance score was calculated as the arithmetic mean of all applicable BASICS sub-criteria, excluding the Shielding category. To enhance system robustness, the software was designed to automatically manage potential DICOM metadata omissions and model-level errors. The SpecificCharacterSet tag was automatically corrected to the ISO_IR 6 format in cases of incompatibility. When exposure-related tags (EI, EIT, or DI) were missing, the corresponding criterion was assigned 0 points and explicitly documented using the \u0026ldquo;obs\u0026rdquo; (observation) label in the report. If the Transfer Syntax UID field was unavailable, the system attempted decoding using an Explicit VR Little Endian assumption; in the event of failure, an error message was generated and recorded. Pixel-based metrics were not calculated for images lacking PixelData, and these cases were reported with the descriptive status \u0026ldquo;image_status\u0026rdquo;. The YOLO11n detection model was executed in a fault-protected configuration, such that only the affected submodule was disabled in error states, while the computation of all other metrics continued uninterrupted. During dataset evaluation, metadata omissions\u0026mdash;most commonly missing Exposure Index values\u0026mdash;were observed in approximately 8% of cases, resulting in an estimated 5% reduction in overall system scoring robustness. This behavior demonstrates the high resilience of the proposed system in maintaining automated workflow continuity and functional integrity in the presence of minor data loss.\u003c/p\u003e \u003cp\u003eBeam-related image quality assessment focused on centering accuracy, projection geometry, tube angulation, and exposure technique parameters.\u003c/p\u003e \u003cp\u003eAnatomical centering was evaluated by calculating the normalized distance between the center of the detected chest Region of Interest (ROI) and the geometric center of the image. The centering score increased as the measured distance decreased; 100 points were awarded for perfect centering, while 0 points corresponded to the farthest deviation from the image center.\u003c/p\u003e \u003cp\u003eProjection alignment and device orientation were assessed using DICOM Image Orientation Patient (IOP) vectors. Horizontal and vertical axis deviations were calculated from these vectors, and left\u0026ndash;right (LR) and top\u0026ndash;bottom (TB) edge symmetry were measured to quantify rotational misalignment.\u003c/p\u003e \u003cp\u003eDevice alignment analysis was performed using orientation axis vectors derived from the Image Orientation (Patient) (IOP) tag. Tube angulation was assessed using DICOM metadata to infer whether the acquisition corresponded to a standard or lordotic projection. In cases where angulation-related DICOM tags were absent, the tube angle criterion was assigned a score of 0. When relevant angulation metadata were available, tube angle appropriateness was evaluated; accordingly, otherwise, the parameter was recorded as unavailable.\u003c/p\u003e \u003cp\u003eThe central ray score was computed as a weighted composite of geometric proxies derived from DICOM angulation metadata (when available), SID proximity to a target value, left\u0026ndash;right symmetry of the chest ROI margins, and the centering score. When angle/SID metadata were unavailable, the score relied on symmetry and centering only, and the report explicitly noted the missing tags.\u003c/p\u003e \u003cp\u003eExposure technique was assessed using DICOM metadata for tube voltage (kVp) and exposure (mAs). These parameters were compared with expected reference values based on patient age and projection type, and deviations resulted in proportional score reductions.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eA- Artifacts\u003c/h2\u003e \u003cp\u003eArtifact assessment included the detection of obstructing foreign objects and verification of laterality (side) markers. Obstructing artifacts such as clothing, monitoring cables, or external objects were detected using the trained YOLO11n model. Artifacts were penalized as 10 points per detection inside the chest ROI and 5 points per detection outside the ROI (floor at 0). Biological foreign objects were not annotated separately, as the current model focuses on removable non-biological artifacts that directly affect acquisition quality.\u003c/p\u003e \u003cp\u003eAn expected device list (e.g., endotracheal tube, nasogastric tube, central venous catheter, chest tube) was derived from DICOM free-text fields. Device/line visibility was then quantified using an image-based proxy: within the YOLO chest ROI, Canny edge detection followed by HoughLinesP was applied to detect line-like structures. The total detected line length was normalized by chest ROI height (norm_len), and the score was scaled on a 0\u0026ndash;100 range based on this normalized length. If no device was expected from DICOM metadata, the criterion was scored as 100 to avoid unnecessary penalization.\u003c/p\u003e \u003cp\u003eLaterality markers (L/R) were evaluated using a template matching approach applied to the four corner regions of the image (14). If a similarity score exceeding the predefined threshold was detected, the image was assigned 100 points for correct side marker presence; otherwise, a score of 0 points was assigned. To prevent false detections, the vertices corresponding to the chest bounding box were excluded from the template matching analysis.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eS- Shielding\u003c/h3\u003e\n\u003cp\u003eThe system included the technical capability to detect protective shielding (gonadal and thyroid protection). However, shielding was not incorporated into the quantitative quality scoring process. In accordance with current pediatric radiography recommendations, this criterion was reported as not applicable (N/A) and excluded from the overall ALARA compliance calculations for chest x-rays (13).\u003c/p\u003e\n\u003ch3\u003eI- Immobilization \u0026 Indicators\u003c/h3\u003e\n\u003cp\u003eImage sharpness and motion artifacts were assessed using multiple \u003cb\u003eimage-based metrics\u003c/b\u003e, including \u003cb\u003eLaplacian variance\u003c/b\u003e, \u003cb\u003eSobel-based edge detection\u003c/b\u003e, and \u003cb\u003efrequency-domain analysis\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eImage sharpness was evaluated using a \u003cb\u003ecombined sharpness score\u003c/b\u003e derived from \u003cb\u003eLaplacian variance\u003c/b\u003e, \u003cb\u003eTenengrad (Sobel size)\u003c/b\u003e, and \u003cb\u003eedge density\u003c/b\u003e parameters. Each metric was normalized and averaged, and the resulting score was expressed on a \u003cb\u003e0\u0026ndash;100 scale\u003c/b\u003e. Images with a score of \u003cb\u003e\u0026ge;\u0026thinsp;60\u003c/b\u003e were classified as \u003cem\u003eclear\u003c/em\u003e, scores between \u003cb\u003e40 and 60\u003c/b\u003e as \u003cem\u003eborderline\u003c/em\u003e, and scores\u0026thinsp;\u003cb\u003e\u0026lt;\u0026thinsp;40\u003c/b\u003e as \u003cem\u003eblurry.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e(15). This approach enables objective quantification of image blur.\u003c/p\u003e \u003cp\u003eMotion analysis was performed by measuring the \u003cb\u003eanisotropy of the angular profile\u003c/b\u003e obtained from the \u003cb\u003e2D Fast Fourier Transform (FFT) log-spectrum\u003c/b\u003e of the image. This metric quantitatively describes the \u003cb\u003edirectional dispersion of motion\u003c/b\u003e. A normalized \u003cb\u003eno-motion score\u003c/b\u003e in the range of \u003cb\u003e0\u0026ndash;100\u003c/b\u003e was calculated by combining the anisotropy value with the \u003cb\u003eacuity score\u003c/b\u003e and the \u003cb\u003egradient direction\u0026ndash;orthogonal ratio\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eExposure indicators were evaluated using the \u003cb\u003eDeviation Index (DI)\u003c/b\u003e, calculated from \u003cb\u003eExposure Index (EI)\u003c/b\u003e values when available. If the DI tag was present in the DICOM data, it was used directly. In the absence of EI and \u003cb\u003eExposure Index Target (EIT)\u003c/b\u003e values, DI was calculated using the formula: \u003cb\u003eDI\u0026thinsp;=\u0026thinsp;10 \u0026times; log₁₀(EI / EIT)\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe DI-based scoring function was defined as: \u003cb\u003eScore\u0026thinsp;=\u0026thinsp;100\u0026thinsp;\u0026minus;\u0026thinsp;20 \u0026times; |DI|\u003c/b\u003e\u003c/p\u003e \u003cp\u003eAccordingly, \u003cb\u003e|DI| \u0026le; 1\u003c/b\u003e was classified as \u003cem\u003eideal exposure\u003c/em\u003e, \u003cb\u003e1 \u0026lt; |DI| \u0026le; 3\u003c/b\u003e as \u003cem\u003eacceptable exposure\u003c/em\u003e, and \u003cb\u003e|DI| \u0026gt; 3\u003c/b\u003e as \u003cem\u003eimproper exposure\u003c/em\u003e. Images lacking exposure indicator metadata were assigned a \u003cb\u003ezero score\u003c/b\u003e for this sub-criterion. This standardized approach minimizes variability in dose indication across different imaging device manufacturers.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eC- Collimation, Cropping\u003c/h2\u003e \u003cp\u003eCollimation evaluation was structured in accordance with the standards defined in the \u003cb\u003eEuropean Commission\u0026rsquo;s European Guidelines on Quality Criteria for Diagnostic Radiographic Images in Paediatrics (1996)\u003c/b\u003e. This approach is based on the concepts of \u003cb\u003eminimum field size (MinFS)\u003c/b\u003e and \u003cb\u003emaximum field size (MaxFS)\u003c/b\u003e described by \u003cb\u003eTschauner et al.\u003c/b\u003e In the present study, these concepts were digitized using a \u003cb\u003eYOLO11n-based chest bounding box\u003c/b\u003e, allowing collimation excess, out-of-area irradiation, and centering to be converted into \u003cb\u003efully automated quantitative measures\u003c/b\u003e without the need for manual assessment (12, 16). The system considered a \u003cb\u003e10% expanded boundary\u003c/b\u003e around the YOLO11n-detected chest region as the recommended \u003cb\u003ecollimation zone\u003c/b\u003e (16). A score deduction was applied proportionally to the area of unnecessary irradiation extending beyond these limits, and the resulting value was normalized on a \u003cb\u003e0\u0026ndash;100 scale\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eCollimation quality was further assessed by analyzing the spatial relationship between the detected chest region and the overall image field boundaries. Excessive irradiation outside the expected anatomical limits resulted in additional score penalties.\u003c/p\u003e \u003cp\u003e \u003cb\u003eElectronic clipping (cropping)\u003c/b\u003e was evaluated through analysis of \u003cb\u003epixel intensity distributions\u003c/b\u003e near the image borders and by reviewing relevant \u003cb\u003eDICOM shutter-related metadata\u003c/b\u003e when available. Boundary regions of the image (default\u0026thinsp;\u0026plusmn;\u0026thinsp;5% band) exhibiting \u003cb\u003elow standard deviation\u003c/b\u003e or \u003cb\u003ehigh saturation\u003c/b\u003e were considered potential indicators of post-processing crops. When shutter or partial-view tags were present in the DICOM data, an \u003cb\u003eadditional point deduction\u003c/b\u003e was applied.\u003c/p\u003e \u003cp\u003eThe final collimation and cropping score was reported as a \u003cb\u003enormalized value between 0 and 100\u003c/b\u003e and visually indicated using a \u003cb\u003ecolored overlay\u003c/b\u003e on the output images.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eS- Structures\u003c/h2\u003e \u003cp\u003eStructural evaluation focused on \u003cb\u003eanatomical coverage\u003c/b\u003e, assessed exclusively based on the \u003cb\u003echest region detected by the YOLO11n model\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eThe system evaluated whether the \u003cb\u003eessential thoracic anatomy\u003c/b\u003e was fully included within the image field by analyzing the spatial extent and position of the \u003cb\u003eYOLO11n-detected chest Region of Interest (ROI)\u003c/b\u003e. Images in which the chest ROI was completely and appropriately contained within the image boundaries received \u003cb\u003efull scores\u003c/b\u003e, while partial coverage resulted in proportional score reductions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eOverall ALARA Compliance Score\u003c/h2\u003e \u003cp\u003eEach BASICS sub-criterion was normalized to a score between 0 and 100. The overall ALARA compliance score was calculated as the arithmetic mean of all applicable BASICS scores, excluding the shielding criterion. This structured and modular methodology enabled objective, reproducible, and automated assessment of pediatric chest radiograph quality.\u003c/p\u003e \u003cp\u003eAll metrics are normalized on a 0\u0026ndash;100 scale according to the BASICS criteria. The system then calculates an average ALARA compliance score.\u003c/p\u003e \u003cp\u003eIn the final stage, the system generates the \u003cem\u003esample_report.txt\u003c/em\u003e file and the \u003cem\u003esample_annotated.png\u003c/em\u003e image for a single image. In aggregate analysis, the file \u003cem\u003ealara_results.xlsx\u003c/em\u003e includes individual results and a general summary page.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Development Opportunities\u003c/h2\u003e \u003cp\u003eThe system developed in this study is based on the YOLO11n model, which includes only the \"chest\" and \"artifact\" classes. For future development, expanding the model to incorporate additional subclasses, such as marker, tube, or line, is recommended to increase its applicability and improve its clinical relevance. Additionally, sharpness and motion assessments may be enhanced by integrating unreferenced quality metrics, such as BRISQUE or NIQA, which provide more comprehensive and objective evaluations of image quality.\u003c/p\u003e \u003cp\u003eAdditionally, EI/DI target ranges can be calibrated according to the device manufacturer's specifications and harmonized with dynamic criteria. Integrating anatomical landmark-based segmentation techniques into collimation analysis can also enhance the system's accuracy. Finally, it is planned to customize kVp/mAs target tables specific to enterprise protocols over configurable JSON or YAML files.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eDemographics\u003c/h2\u003e \u003cp\u003eA total of 600 chest X-rays belonging to 555 children were included in the study. More than one radiograph of some cases was evaluated. The ages of the participants ranged from 0 to 18 years, with an average age of 5.40\u0026thinsp;\u0026plusmn;\u0026thinsp;5.02 years. Of the cases, 53.8% were male, and 46.2% were female.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eAI Metrics\u003c/h2\u003e \u003cp\u003eModel training was performed in an environment with CUDA acceleration enabled for 100 epochs at a resolution of 640 \u0026times; 640 pixels. 80% (480 images) of the training data is allocated for training, 10% (60 images) for validation, and 10% for testing. The training process was completed in approximately 17 minutes. The model's test performance metrics are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel Performance\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClass\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003emAP50\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003emAP@50\u0026ndash;95\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.995\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.870\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArtifact\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.738\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.860\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.525\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAverage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.919\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.869\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.927\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.697\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eUltralytics 8.3.96, Python-3.12.9 torch-2.6.0\u0026thinsp;+\u0026thinsp;cu118 CUDA:0 (NVIDIA GeForce GTX 1080, 8192 MiB). YOLO11n summary (fused): 100 layers, 2,582,542 parameters, 0 gradients, 6.3 GFLOPs. Speed: 0.2 ms preprocess, 2.9 ms inference, 0.0 ms loss, 1.2 ms postprocess per image.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe accuracy of the YOLO11n model in the chest class is remarkably high (mAP50\u0026thinsp;=\u0026thinsp;0.995), indicating that the anatomical area is determined almost flawlessly. The relatively low accuracy and sensitivity rates in the Artifact class (Precision\u0026thinsp;=\u0026thinsp;0.861, Recall\u0026thinsp;=\u0026thinsp;0.738) indicate limited performance in detecting clothing elements or small objects that are obscured by movement.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eSingle image evaluation\u003c/h2\u003e \u003cp\u003eUsing the obtained YOLO11n model and the methodology described above, the evaluation of a \u003cb\u003esample.dcm\u003c/b\u003e file was performed. In this sample analysis, patient identification information was hidden or altered. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents an example of the annotated image created by the software. The mean overall score was found to be 64.40 in the evaluation of single samples. The highest scores were obtained in the kVp suitability (100) and Structures (100) criteria; the lowest scores were recorded in mAs suitability (0). This suggests that the lack of exposure labels in the dataset and operator habits are key determinants of ALARA compliance (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of Automated BASICS Assessment Results for a Single Image\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"1\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e=== BASICS (CHEST) Automatic Assessment ===\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverall score (mean of items): 64.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Beam]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Anatomy is centered in the image: 58.94\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Central ray appropriate for projection: 48.26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: View\u0026thinsp;=\u0026thinsp;NA; angP\u0026thinsp;=\u0026thinsp;NA, angS\u0026thinsp;=\u0026thinsp;NA, SID\u0026thinsp;=\u0026thinsp;0 mm; sym\u0026thinsp;=\u0026thinsp;82.3, center\u0026thinsp;=\u0026thinsp;58.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Tube angle appropriate: 0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: mode=standard; target\u0026thinsp;=\u0026thinsp;0\u0026thinsp;\u0026plusmn;\u0026thinsp;10\u0026deg;; no angle tags\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Device alignment appropriate: 41.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: no IOP tag; LR\u0026thinsp;=\u0026thinsp;82.3, TB\u0026thinsp;=\u0026thinsp;0.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- kVp appropriate for projection/patient: 100.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: profile=pediatric/AP/PA [60\u0026ndash;90] kVp; value\u0026thinsp;=\u0026thinsp;90.0 (in-range); age=1y\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- mAs appropriate for projection/patient: 0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: profile=pediatric/AP/PA [0.5-3.0] mAs; value\u0026thinsp;=\u0026thinsp;0.00 (below); age=1y\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Artifacts]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- No obstructing artifacts present (e.g., lead, clothing, ECG cables): 90.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: in =\u0026thinsp;0, out =\u0026thinsp;2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Side marker present and correct: 100.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: found\u0026thinsp;=\u0026thinsp;L @ ML (score\u0026thinsp;=\u0026thinsp;0.46)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Shielding]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Gonadal/thyroid shielding used appropriately (when indicated): N/A\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: N/A (Shielding not required for Chest)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Immobilization \u0026amp; Indicators]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Patient cooperation/immobilization adequate (no motion): 81.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: aniso\u0026thinsp;=\u0026thinsp;0.22, dir=0deg; terms: aniso\u0026thinsp;=\u0026thinsp;0.29, blur\u0026thinsp;=\u0026thinsp;0.05, dirR\u0026thinsp;=\u0026thinsp;1.68\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Exposure Indicator (EI) within target range: 56.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: EI\u0026thinsp;=\u0026thinsp;170.0, EIT\u0026thinsp;=\u0026thinsp;280.0, DI=-2.20 (acceptable; DI tag)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Deviation Index (DI) close to 0 (ideal): 56.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: EI\u0026thinsp;=\u0026thinsp;170.0, EIT\u0026thinsp;=\u0026thinsp;280.0, DI=-2.20 (acceptable; DI tag)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Collimation]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Field is collimated to area of interest BEFORE exposure: 34.38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: unnecessary\u0026thinsp;=\u0026thinsp;65.6%, useful\u0026thinsp;=\u0026thinsp;34.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- No reliance on electronic cropping after exposure: 100.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: tags=[none]; suspicious_border\u0026thinsp;=\u0026thinsp;0.0%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e[Structures]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Necessary anatomy fully demonstrated: 100.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: chest bbox fully inside image\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e- Devices/lines/tubes (if any) properly demonstrated: 100.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; OBS: no device expected by tags; lines_detected\u0026thinsp;=\u0026thinsp;1, norm_len\u0026thinsp;=\u0026thinsp;0.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e--- \u0026Ouml;zet ---\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e--- Summary ---\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePNG: sample.png\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnnotasyon: sample_annotated.png\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAssessment time: 3.54 seconds\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eBatch Evaluation.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eFor clinical validation of the system, a test set of 4461 images was initially prepared. The automatic score was calculated for each image, and the average results are presented in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The automated evaluation process took approximately 5.6 hours and produced continuous, uninterrupted reporting throughout the run. The strongest performance was observed in the Structures criteria\u0026mdash;Necessary anatomy fully demonstrated (100) and Devices/lines/tubes properly demonstrated (98.36)\u0026mdash;together with Side Marker present and correct (99.69) and Patient cooperation/immobilization adequate (95.71). The Shielding criterion was excluded from the calculation (N/A) in accordance with updated radiation safety guidelines. In contrast, the lowest values were recorded for Tube Angle appropriate (0) and for the Exposure Indicator (EI) within target range (12.71), and Deviation Index (DI) close to 0 (12.71), followed by Device Alignment (41.13) and Collimation before exposure (56.91). Overall, the markedly low EI/DI scores indicate that missing parameters and technical/metadata inconsistencies in DICOM headers can directly reduce ALARA-based scoring performance. The zero tube angle score reflects the absence of angulation-related metadata rather than incorrect patient positioning.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBatch analysis scores (n\u0026thinsp;=\u0026thinsp;4461).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAverage score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | Anatomy is centered in the image (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e82.12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | Central ray appropriate for projection (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e67.15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | Tube angle appropriate (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | Device alignment appropriate (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e41.13\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | kVp appropriate for projection/patient (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeam | mAs appropriate for projection/patient (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArtifacts | No obstructing artifacts present (e.g. lead. clothing. ECG cables) (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e92.52\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArtifacts | Side marker present and correct (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e99.69\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eShielding | Gonadal/thyroid shielding used appropriately (when indicated) (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImmobilization \u0026amp; Indicators | Patient cooperation/immobilization adequate (no motion) (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e95.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImmobilization \u0026amp; Indicators | Exposure Indicator (EI) within target range (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImmobilization \u0026amp; Indicators | Deviation Index (DI) close to 0 (ideal) (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCollimation | Field is collimated to area of interest BEFORE exposure (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56.91\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCollimation | No reliance on electronic cropping after exposure (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e90.95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStructures | Necessary anatomy fully demonstrated (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStructures | Devices/lines/tubes (if any) properly demonstrated (score)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e98.36\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOverall Score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e64.95\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eOverall rating\u003c/h2\u003e \u003cp\u003eThe ALARA-based artificial intelligence system developed in this study is an end-to-end solution that performs automatic quality assessment of pediatric chest X-rays. Based on the BASICS criteria, the system evaluates each quality indicator with quantitative scores and generates an overall ALARA compliance score.\u003c/p\u003e \u003cp\u003eThe most important stage before a chest X-ray is the proper determination of the indication; The next stage is optimization (12). Thus, it may be possible to achieve the principle of \"As Low As Reasonably Achievable\" (ALARA) without compromising diagnostic accuracy (13). For this purpose, necessary regulations have been established by the legislators of many countries (17).\u003c/p\u003e \u003cp\u003eIn the study by Abubakr et al., it was demonstrated that expert inspections based on nine different acquisition criteria resulted in a significant improvement in subsequent imaging (18). However, the increasing number of examinations and intensive workflow in modern healthcare make manual examination of these examinations time-consuming and subjective.\u003c/p\u003e \u003cp\u003eGranata et al. reported that 20\u0026ndash;50% of the examinations performed in pediatric radiology had indication eligibility problems (13). This finding underscores the need for automated inspection mechanisms to optimize radiation. The ALARA-based software proposed in our study represents an innovative approach that addresses this requirement. The system objectively applies the optimization principle by numerically evaluating collimation and anatomical coverage rates.\u003c/p\u003e \u003cp\u003eThe European Commission's Radiation Protection No. 162 defines the parameters of \"anatomical coverage, collimation, centering, and acuity\" for evaluating diagnostic quality in pediatric chest X-rays (12). These standards establish the minimum quality requirements for applying ALARA principles in the field. In our study, these criteria were automatically analyzed through artificial intelligence-assisted software.\u003c/p\u003e \u003cp\u003eRecent artificial intelligence studies show that routine chest X-rays can be used not only for diagnostic purposes, but also for opportunistic analysis (e.g., osteoporosis risk prediction) (19). Similarly, this study expanded the role of artificial intelligence in radiography to a \"beyond diagnosis\" field by using pediatric chest X-rays in automated quality and radiation safety assessment.\u003c/p\u003e \u003cp\u003eThus, the proposed system reduces the manual inspection burden on radiologists and provides an objective and repeatable quality control mechanism. Although there are more than forty licensed artificial intelligence software programs for chest X-ray and tomography analysis today, most of them focus on the adult population (8).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eB \u0026ndash; Beam (Dosing and Centering)\u003c/h2\u003e \u003cp\u003eDosing parameters are one of the most critical quality determinants in pediatric radiology. In our study, it was observed that kVp values were mostly in the recommended range of 60\u0026ndash;90, but mAs values were understated or under-recorded. This is thought to be related to the lack of DICOM tags on some devices. Beam centering score of 82.12 indicates that positioning is more challenging in pediatric patients than in adults (5).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eA \u0026ndash; Artifacts (Foreign Objects and Side Signs)\u003c/h2\u003e \u003cp\u003eThe presence of biological foreign objects (BFO) or non-biological foreign objects (NBFO) on chest X-rays may complicate the diagnosis of pathologies such as fluid accumulation, tuberculosis, or cysts (20). Santosh et al. (24)(21) achieved precision, recall, and F1-score values of 0.85, 0.93, and 0.89, respectively, on 400 images using the BFO/NBFO detection system they developed with the YOLOv4 algorithm.\u003c/p\u003e \u003cp\u003eZohora et al. proposed a method for detecting foreign circular objects in chest X-ray images using Sobel, Canny, Prewitt, and Roberts edge detectors, followed by morphological operations and a circular Hough transform, and reported high detection accuracy and computational efficiency compared to existing methods (22).\u003c/p\u003e \u003cp\u003eIn our model, only artifacts that fall into the NBFO class are labeled. The performance criteria of the model for this class were calculated as follows: precision, 0.861; recall, 0.738; mAP@50, 0.860; and mAP@50\u0026ndash;95, 0.525. The results show that the model identifies artifacts with a large percentage of accuracy, but has limited sensitivity in small, low-contrast objects.\u003c/p\u003e \u003cp\u003e Side marker signs (\u0026ldquo;L\u0026rdquo;/\u0026ldquo;R\u0026rdquo;) achieved an excellent score (99.69), suggesting strong operator compliance with laterality labeling (8).\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eS \u0026ndash; Shielding\u003c/h2\u003e \u003cp\u003eIn our study, this item was excluded from the quantitative scoring and marked as 'N/A' (Not Applicable). As Granata et al. point out, the routine use of contact shielding in pediatric radiology has become controversial today due to its limited benefits and potential risks (13). Consensus reports from AAPM and European radiological societies (ESR, ESPR) emphasize that protectors carry the risk of covering anatomical details, are difficult to place correctly, and can paradoxically increase the dose to which the patient is exposed by misleading Automatic Exposure Control (AEC) systems. In this context, although the software we have developed scores the presence of shielding as a 'positive' quality criterion with its current algorithm, due to this paradigm shift in radiation safety culture, it would be a more clinically correct approach to revise this criterion according to the principle of 'no-shielding' in future versions of the system and to accept the non-use of shielding as a standard. Our aim in maintaining this scoring can be explained as the adequacy of our study for use in areas that require protection, such as the pelvis.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eI \u0026ndash; Immobilization \u0026amp; Indicators\u003c/h2\u003e \u003cp\u003eThe high score of 95.71 points in the motion analysis indicates that the children were immobilized with appropriate fixation methods during the shoot.\u003c/p\u003e \u003cp\u003eMoore et al. emphasized that digital radiography may mask overexposure, resulting in unnecessary dose escalation (23). To prevent this, it is recommended to use the EI and DI index values (5). The fact that EI and DI labels were largely missing in our study was the main reason for the low scores (23). This deficiency is attributed to the non-standard data recording of the devices, highlighting a limitation in data quality.\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eC \u0026ndash; Collimation\u003c/h2\u003e \u003cp\u003e The automatic collimation assessment module is structured based on the field definitions in the European Commission's Radiation Protection No.162 guideline. The 10\u0026plusmn;% safety margin added to the chest area detected by YOLO11n represents the recommended MinFS\u0026ndash;MaxFS (Minimum-Maximum Area) range for pediatric chest X-rays. In their study analyzing the European guidelines, Tschauner et al. stated that the ideal area should cover the T2\u0026ndash;L1 vertebral spacing and defined age-related tolerance limits of 10\u0026ndash;20 mm. In the same study, the average overexposure exceeding the guideline limits was reported as 45.1% (16). Using a similar methodology, Pedersen et al. demonstrated that U-Net-based AI models can detect lung segmentation and collimation boundaries in high agreement with radiologists (Dice score\u0026thinsp;\u0026gt;\u0026thinsp;0.93) (24).\u003c/p\u003e \u003cp\u003eIn our study, the average fitness score for collimation remained as low as 56.91 in the test set of 4,461 images. These findings demonstrate that manual collimation is insufficient in clinical practice, and our automated assessment module makes a significant contribution to ensuring sustainable radiation safety in pediatric radiography by reducing this risk.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003eS \u0026ndash; Structures (Displayed Anatomical Structures)\u003c/h2\u003e \u003cp\u003ePedersen and his team, utilizing the U-Net architecture, achieved automatic collimation, resulting in a Dice value of 0.95 in boundary segmentation. In contrast, our YOLO11n-based approach yielded similarly accurate results, facilitating faster integration into DICOM-based quality control lines (24).\u003c/p\u003e \u003cp\u003eMouton et al. achieved an accuracy of AUC\u0026thinsp;=\u0026thinsp;0.782 in the classical CAD system (27)(25). These results demonstrate the level of accuracy of deep learning-based models. In our system, the value mAP@50\u0026thinsp;=\u0026thinsp;0.995 represents the transition from the classical CAD era to today's fully automated ALARA inspection systems.\u003c/p\u003e \u003cp\u003eIn addition, Kufel et al. in the YOLOv8-based study, a precision of 0.815 was reported in 112,120 chest X-rays (26). In our model, chest ROI was defined with 99% accuracy, with a score of 100 for the Structures criteria. This result shows that anatomical coverage inspection with artificial intelligence can be successfully integrated into clinical practice.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eThis study enabled the development of a fully automated quality control process for pediatric chest X-rays. However, it is an important limitation that all criteria are given equal weight in the general scoring. In future studies, the criteria may be weighted based on their clinical significance, ensuring a fairer overall score.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study integrated the DICOM infrastructure, image processing, and deep learning components to provide an objective and reproducible evaluation of pediatric chest radiographs within the framework of BASICS. The developed batch analysis flow enables monitoring, comparison, and continuous improvement of quality indicators throughout the organization. As a result, the system performed the evaluation based on ALARA principles in pediatric radiographs in an automated, standardized, and reproducible manner, in accordance with the principles outlined in European Guidelines RP 162 (1996) and the Image Gently initiative (2012) (5, 16).\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eNo funding was received for this study.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthor Contributions Statement:G.A. and B.G. jointly developed the study concept and design. G.A. was responsible for collection of pediatric chest radiographs, case assembly, development and coding of the artificial intelligence\u0026ndash;based system, image processing workflows, and batch analysis. B.G. performed image annotation, reference labeling according to ALARA/BASICS criteria, radiological quality assessment, and clinical validation of the model outputs. G.A. and B.G. contributed to the interpretation of the results. All authors reviewed, revised, and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and/or analysed during the current study are not publicly available due to patient privacy concerns and institutional data protection regulations. Some radiographic images contain patient identifiers embedded within the image. However, de-identified versions of the images may be made available from the corresponding author upon reasonable request for research purposes.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eRontgen WC. On a New Kind of Rays. Science. 1896;3(59):227-31.\u003c/li\u003e\n\u003cli\u003eDaniel J. The X-rays. Science. 1896;3(67):562-3.\u003c/li\u003e\n\u003cli\u003eWagner LK, Eifel PJ, Geise RA. Potential biological effects following high X-ray dose interventional procedures. J Vasc Interv Radiol. 1994;5(1):71-84.\u003c/li\u003e\n\u003cli\u003eWillis CE, Slovis TL. The ALARA concept in pediatric CR and DR: dose reduction in pediatric radiographic exams--a white paper conference executive summary. Pediatr Radiol. 2004;34 Suppl 3:S162-4.\u003c/li\u003e\n\u003cli\u003eDon S, Goske MJ, John S, Whiting B, Willis CE. Image Gently pediatric digital radiography summit: executive summary. Pediatr Radiol. 2011;41(5):562-5.\u003c/li\u003e\n\u003cli\u003eAk\u0026ccedil;ay G, G\u0026uuml;ney B, Deveer M, Topal Y. A qualitative evaluation of direct radiography in pediatric clinics. Sağlık Akademisyenleri Dergisi. 2016;3(3):100-5.\u003c/li\u003e\n\u003cli\u003eGore JC. Artificial intelligence in medical imaging. Elsevier; 2020. p. A1-A4.\u003c/li\u003e\n\u003cli\u003eSchalekamp S, Klein WM, van Leeuwen KG. Current and emerging artificial intelligence applications in chest imaging: a pediatric perspective. Pediatric Radiology. 2022;52(11):2120-30.\u003c/li\u003e\n\u003cli\u003eAl-Naser YA. The impact of artificial intelligence on radiography as a profession: A narrative review. Journal of Medical Imaging and Radiation Sciences. 2023;54(1):162-6.\u003c/li\u003e\n\u003cli\u003eAnaconda. Miniconda \u0026ndash; Getting Started 2025 [Available from: https://www.anaconda.com/docs/getting-started/miniconda/main.\u003c/li\u003e\n\u003cli\u003eWada K. Labelme \u0026ndash; Create your image dataset for vision AI 2025 [Available from: https://labelme.io/.\u003c/li\u003e\n\u003cli\u003eEuropean Commission. European guidelines on quality criteria for diagnostic radiographic images in paediatrics. Luxembourg: Office for Official Publications of the European Communities; 1996.\u003c/li\u003e\n\u003cli\u003eGranata C, Sofia C, Francavilla M, Kardos M, Kasznia-Brown J, Nievelstein RA, et al. Let\u0026rsquo;s talk about radiation dose and radiation protection in children. Pediatric radiology. 2025;55(3):386-96.\u003c/li\u003e\n\u003cli\u003eArimura H, Ishida T, Katsuragawa S, Kawashita I, Doi K. [Development of a computerized method for identifying view position and orientation for chest radiographs by using a template matching technique]. Nihon Hoshasen Gijutsu Gakkai Zasshi. 2002;58(8):1047-54.\u003c/li\u003e\n\u003cli\u003eKim C-M, Hong EJ, Park RC. Chest X-ray outlier detection model using dimension reduction and edge detection. IEEE Access. 2021;9:86096-106.\u003c/li\u003e\n\u003cli\u003eTschauner S, Marterer R, G\u0026uuml;bitz M, Kalmar PI, Talakic E, Weissensteiner S, et al. European Guidelines for AP/PA chest X-rays: routinely satisfiable in a paediatric radiology division? Eur Radiol. 2016;26(2):495-505.\u003c/li\u003e\n\u003cli\u003eCouncil of the European Union. Council Directive 2013/59/Euratom of 5 December 2013 laying down basic safety standards for protection against the dangers arising from exposure to ionising radiation. Directive ed: European Union; 2013. p. 1-73.\u003c/li\u003e\n\u003cli\u003eMuhammed AEM, Awad MSA, Saeed MM, Alkheder MA, Babiker MAM, Alzain MAA, et al. Adapted Anatomical Image Criteria for PA Chest Radiographs at Managil Teaching Hospital, Sudan 2023. Clinical Audit. 2024:63-8.\u003c/li\u003e\n\u003cli\u003eSmith AD, Rothenberg SA. AI and Chest Radiographs: A Dawning Era in Osteoporosis Screening. Radiological Society of North America; 2024. p. e241339.\u003c/li\u003e\n\u003cli\u003eRoy S, Santosh KC. Analyzing Overlaid Foreign Objects in Chest X-rays-Clinical Significance and Artificial Intelligence Tools. Healthcare (Basel). 2023;11(3):308.\u003c/li\u003e\n\u003cli\u003eSantosh K, Roy S, Allu S. Generic Foreign Object Detection in Chest X-rays. International Conference on Recent Trends in Image Processing and Pattern Recognition; Cham: Springer International Publishing; 2022. p. 93-104.\u003c/li\u003e\n\u003cli\u003eZohora FT, Santosh K. Circular foreign object detection in chest x-ray images. International Conference on Recent Trends in Image Processing and Pattern Recognition. 2016:391-401.\u003c/li\u003e\n\u003cli\u003eMoore QT, Don S, Goske MJ, Strauss KJ, Cohen M, Herrmann T, et al. Image gently: using exposure indicators to improve pediatric digital radiography. Radiol Technol. 2012;84(1).\u003c/li\u003e\n\u003cli\u003ePedersen A, Kusk MW, Knudsen G, Busk C, Lysdahlgaard S. Collimation border with U-Net segmentation on chest radiographs compared to radiologists. Radiography. 2023;29(3):647-52.\u003c/li\u003e\n\u003cli\u003eMouton A, Pitcher RD, Douglas TS. Computer-Aided Detection of Pulmonary Pathology in Pediatric Chest Radiographs. International Conference on Medical Image Computing and Computer-Assisted Intervention. 2010:619-25.\u003c/li\u003e\n\u003cli\u003eKufel J, Bargieł-Łączek K, Koźlik M, Czogalik Ł, Dudek P, Magiera M, et al. Chest X-ray foreign objects detection using artificial intelligence. Journal of Clinical Medicine. 2023;12(18):5841.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Radiation Safety, Pediatric Radiology, Artificial Intelligence, Deep Learning, Quality Control, ALARA, BASICS","lastPublishedDoi":"10.21203/rs.3.rs-9008392/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9008392/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePediatric radiology requires strict adherence to the ALARA principle (As Low As Reasonably Achievable), which refers to minimizing radiation exposure as much as possible, due to children\u0026rsquo;s high sensitivity to ionizing radiation.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003e This study aimed to develop an artificial intelligence\u0026ndash;assisted system to evaluate compliance of pediatric chest radiographs with radiation safety principles.\u003c/p\u003e\u003ch2\u003eMaterial and Methods\u003c/h2\u003e \u003cp\u003eA total of 600 pediatric chest radiographs (aged 0\u0026ndash;18 years) were collected and used to train a YOLO11n-based deep learning model for anatomical region and artifact detection. For ALARA compliance assessment, a separate test set of 4,461 chest radiographs was analyzed in batch mode. Image processing and metadata parsing were performed using the Pydicom and OpenCV libraries. Compliance with radiation safety principles was quantified on a 0\u0026ndash;100 scale according to the BASICS criteria (Beam, Artifacts, Shielding, Immobilization \u0026amp; Indicators, Collimation, Structures).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe model achieved a mean average precision (mAP50) of 0.995 for the chest class and 0.860 for artifacts. In the test set, the mean ALARA compliance score was 64.95. The highest compliance was observed in Structures (100% for necessary anatomy and 98.36% for devices/lines/tubes), along with the presence of side markers (99.69%) and immobilization (95.71%). In contrast, major deficiencies were identified in the tube angle score (0) and exposure indicator metrics (EI\u0026thinsp;=\u0026thinsp;12.71 and DI\u0026thinsp;=\u0026thinsp;12.71), followed by device alignment (41.13%) and collimation before exposure (56.91%).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe system provides objective and reproducible quality analysis, enabling continuous monitoring of pediatric chest radiography standards to support radiation safety.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence–Assisted Evaluation of Alara Compliance in Pediatric Chest Radiography","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-19 12:34:59","doi":"10.21203/rs.3.rs-9008392/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-14T17:09:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-14T10:26:35+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-10T13:15:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"105509733858005144653122821666205831933","date":"2026-05-09T14:00:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"11745834195279110685938649971242261404","date":"2026-05-04T13:51:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-04T10:44:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"74944548104340636396899882865027464915","date":"2026-04-17T19:31:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"11282410818092318009409111475967164527","date":"2026-04-12T11:00:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"177828368338814957185857975544782900406","date":"2026-04-12T08:27:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-10T06:49:07+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-10T06:42:45+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-18T04:58:29+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-14T06:13:07+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-03-13T21:24:33+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fcdf5da5-f477-45ec-929f-9acbf4ec80b1","owner":[],"postedDate":"April 19th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-14T17:09:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-14T10:26:35+00:00","index":81,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-10T13:15:28+00:00","index":80,"fulltext":""},{"type":"reviewerAgreed","content":"105509733858005144653122821666205831933","date":"2026-05-09T14:00:57+00:00","index":79,"fulltext":""},{"type":"reviewerAgreed","content":"11745834195279110685938649971242261404","date":"2026-05-04T13:51:32+00:00","index":78,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-04T10:44:07+00:00","index":75,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":66552709,"name":"Physical sciences/Engineering"},{"id":66552710,"name":"Health sciences/Health care"},{"id":66552711,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2026-05-17T15:39:03+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-19 12:34:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9008392","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9008392","identity":"rs-9008392","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}