Combined Lung and Cranial Ultrasound for Early Identification and Severity Stratification of Neonatal Respiratory Distress Syndrome

Summary

This protocol describes a standardized bedside approach combining lung and cranial ultrasound for early identification and severity stratification of neonatal respiratory distress syndrome. The method integrates lung aeration scoring with cerebral hemodynamic assessment to support reproducible risk evaluation, treatment-escalation decisions, and dynamic monitoring in neonatal intensive care.

Abstract

Early identification and severity stratification of neonatal respiratory distress syndrome (NRDS) are essential for timely respiratory support and prevention of complications. This prospective observational study evaluated the clinical utility of a combined lung ultrasound (LUS) and cranial ultrasound (CrUS) framework for multidimensional bedside risk assessment in 137 neonates with respiratory distress. Baseline LUS and CrUS examinations were performed within 6 h of admission, followed by dynamic LUS monitoring at 24 h and 72 h. The integrated model (Clinical + LUS + CrUS) demonstrated superior diagnostic performance (AUC = 0.941). Optimal neonatal LUS (nLUS) thresholds were 8.5 for NRDS diagnosis, 9.5 for surfactant therapy, and 12.0 for prediction of invasive mechanical ventilation. In addition, improvement in nLUS at 24 h was significantly correlated with oxygenation response (r = 0.52, p < 0.001). These findings indicate that combined LUS-CrUS assessment provides a non-invasive and dynamic strategy for early NRDS identification and management optimization while enabling concurrent neurological risk monitoring.

Introduction

Neonatal respiratory distress syndrome (NRDS) is a major cause of respiratory failure and early neonatal mortality, particularly in preterm infants¹. The condition is primarily driven by pulmonary surfactant deficiency and structural lung immaturity², and its incidence remains high despite advances in perinatal care and respiratory support strategies³. Although pulmonary surfactant replacement therapy and non-invasive ventilation have improved survival in very low birth weight infants⁴, NRDS continues to impose substantial short-term and long-term burdens. In the acute phase, severe hypoxemia and acidosis are common, while persistent alveolar collapse and inflammatory injury may contribute to later complications, including bronchopulmonary dysplasia and neurodevelopmental impairment⁵. Therefore, accurate early identification and severity stratification within the first postnatal day are critical for improving outcomes and guiding individualized management⁶.

Traditional assessment of NRDS relies on clinical signs, arterial blood gas analysis, and chest radiography. However, these methods have important limitations in neonatal intensive care practice. Arterial blood gas analysis is invasive, and repeated sampling may increase the risk of anemia and infection in vulnerable neonates⁷. Chest radiography remains widely used, but radiographic findings may not fully synchronize with early clinical deterioration, and repeated exposure raises concerns about cumulative radiation, especially in preterm infants. In addition, radiography is a static imaging method and cannot provide real-time bedside monitoring to support rapid adjustment of respiratory support or the timing of surfactant administration⁸. These limitations create a clear need for a non-invasive, repeatable, and dynamic bedside assessment pathway.

Bedside lung ultrasound (LUS) has emerged as an effective tool for neonatal respiratory evaluation because it can detect pulmonary aeration loss, interstitial syndrome, consolidation, and pleural-line abnormalities with high sensitivity and specificity⁹. Standardized lung ultrasound scoring further enables semi-quantitative assessment of pulmonary injury severity and dynamic follow-up of treatment response¹⁰. At the same time, cranial ultrasound (CrUS) plays an essential role in monitoring cerebral status in preterm and critically ill neonates. Respiratory dysfunction can impair cerebral autoregulation through hypoxemia, hypercapnia, and hemodynamic instability, thereby increasing the risk of intracranial pressure fluctuation and intraventricular hemorrhage¹¹. Doppler-based CrUS can provide real-time information on cerebral blood flow parameters, including resistance index (RI) and pulsatility index (PI), while also allowing bedside evaluation of ventricular morphology¹².

However, pulmonary and cerebral ultrasound assessments are often performed separately in routine practice, which limits early integrated risk interpretation in neonates with respiratory distress¹³. A combined framework that links pulmonary structural severity with cerebral hemodynamic response may provide a more clinically useful and physiologically complete assessment of early NRDS progression¹⁴. To improve reproducibility and clinical transferability, the present study applies a predefined dual-axis bedside assessment workflow with fixed examination time points (within 6 h, 24 h, and 72 h), standardized ultrasound scoring output, and unified outcome-oriented risk evaluation. This study therefore proposes a dual-axis stratification model integrating lung ultrasound severity scoring with cranial Doppler parameters and examines whether this combined approach improves early identification and severity assessment of NRDS. Specifically, the study evaluates correlations between neonatal lung ultrasound score (nLUS) and cerebral Doppler indices and assesses the predictive value of the combined model for surfactant administration and escalation to invasive mechanical ventilation. Through dynamic bedside monitoring, this work aims to provide a practical and clinically applicable pathway for early risk stratification and precision management in neonates with respiratory distress.

Protocol

Ethics Statement

All procedures were performed in accordance with the institutional neonatal intensive care unit (NICU) guidelines and approved clinical protocols for neonatal imaging and monitoring. Ethical approval for this study was obtained from the Institutional Ethics Committee of Ganzhou Maternal and Child Health Hospital before study initiation (Approval No.: (2023) 101). Written informed consent was obtained from the parents or legal guardians of all enrolled neonates before ultrasound examination and data collection. Consent status was recorded in the case report form (CRF) before any study-specific imaging or data extraction was performed.

1. Study Design and Case Enrollment

1. Screen and enroll eligible neonates
   1. Conduct the study in the NICU using a single-center observational design.
   2. Screen neonates admitted with respiratory distress consecutively during the study period.
   3. Review the admission record, respiratory symptoms, oxygen requirement, and initial clinical diagnosis before enrollment.
   4. Confirm eligibility using the predefined inclusion and exclusion criteria before the first ultrasound examination.
   5. Assign a unique study identification number to each enrolled neonate.
   6. Record screening status, enrollment time, gestational age, birth weight, respiratory support mode, treatment interventions, and planned ultrasound time points in a unified CRF.
2. Perform baseline and follow-up ultrasound examinations
   1. Perform baseline lung ultrasound (LUS) and cranial ultrasound (CrUS) within 6 h of NICU admission.
   2. Perform baseline LUS before or as close as possible to major treatment escalation, including surfactant administration or invasive mechanical ventilation, when clinically feasible.
   3. Perform follow-up LUS examinations at 24 ± 4 h and 72 ± 8 h after baseline imaging.
   4. Confirm that the neonate is clinically stable enough for bedside ultrasound before each scan.
   5. Keep the neonate in a supine position for anterior lung and cranial scanning and use gentle lateral repositioning for lateral or posterior lung zones when clinically tolerated.
   6. Delay non-urgent imaging if the neonate has severe instability, emergency intervention, or unsafe handling conditions.
   7. Record the actual scan time, respiratory support mode, oxygen requirement, and major treatment events at each ultrasound time point.
   8. Document the reason for any delayed or incomplete examination if scheduled imaging cannot be completed within the target time window.
   9. Follow the standardized workflow shown in Figure 1 for study screening, diagnostic grouping, ultrasound timing, and follow-up assessment.

2. Diagnostic Reference and Clinical Definitions

1. Classify NRDS and non-NRDS cases
   1. Diagnose NRDS using a composite clinical reference standard based on respiratory distress manifestations, oxygen or respiratory support requirement, oxygenation impairment, compatible imaging findings, and final clinical diagnosis.
   2. Differentiate non-NRDS respiratory disorders including transient tachypnea of the newborn, infection-related pneumonia, meconium aspiration syndrome, pneumothorax, and congenital cardiopulmonary abnormalities.
   3. Re-evaluate uncertain cases using follow-up clinical course, repeat imaging, oxygenation response, and treatment response.
   4. Adjudicate discrepant or uncertain cases through consensus review by two senior neonatologists who are not involved in ultrasound scoring.
2. Define severity categories
   1. Classify cases as severe NRDS if invasive mechanical ventilation, escalation of respiratory support, or repeat surfactant administration occurs during early hospitalization.
   2. Categorize NRDS severity into mild, moderate, and severe strata using respiratory support intensity, oxygen requirement, oxygenation impairment, and treatment burden.
   3. Generate binary severity comparisons for ROC-based cutoff analyses, including mild versus moderate, moderate versus severe, and non-severe versus severe NRDS.
3. Categorize cranial ultrasound findings
   1. Classify CrUS findings as normal or abnormal for neurologic risk stratification.
   2. Record intraventricular hemorrhage-related abnormalities, ventricular enlargement, and abnormal periventricular white matter echogenicity using the predefined CrUS reporting form.
   3. Measure Doppler-derived cerebral hemodynamic parameters in the anterior cerebral artery when a stable waveform is obtained.
   4. Calculate the resistance index (RI) using the formula: RI = (peak systolic velocity − end-diastolic velocity) / peak systolic velocity.
   5. Calculate the pulsatility index (PI) using the formula: PI = (peak systolic velocity − end-diastolic velocity) / mean flow velocity.
   6. Repeat each Doppler measurement three times during the same examination and record the mean RI and PI values in the dataset.
   7. Treat RI and PI as supplemental Doppler-derived variables for neurologic risk stratification rather than as standalone diagnostic criteria for NRDS.

3. Ultrasound Acquisition, Timing, and Image Storage

1. Acquire ultrasound images
   1. Perform all ultrasound examinations at the bedside using a standardized acquisition workflow.
   2. Record study ID, imaging time point, actual scan time, respiratory support mode, and major clinical interventions during each scan.
2. Configure ultrasound settings
   1. Perform LUS using a high-frequency linear probe with a frequency range of 10–14 MHz.
   2. Perform CrUS using a micro-convex or sector probe with a frequency range of 5–8 MHz.
   3. Set LUS imaging depth to 3–5 cm and position the pleural line in the upper third of the image.
   4. Set CrUS imaging depth to 6–8 cm to include ventricular and periventricular structures.
   5. Adjust gain and focus to optimize image visualization while avoiding over-gain.
   6. Adjust pulsed-wave Doppler settings to obtain stable waveforms without aliasing.
   7. Maintain the insonation angle as low as feasible during Doppler acquisition.
3. Store ultrasound data
   1. Store all images and clips in a de-identified digital archive.
   2. Use standardized file naming including study ID, modality, time point, and operator code.

4. Lung Ultrasound Acquisition and Scoring

1. Perform lung ultrasound scanning
   1. Perform LUS using a fixed 12-region scanning scheme including anterior, lateral, and posterior lung zones.
   2. Scan upper and lower regions bilaterally using a consistent sequence across cases, starting from the right anterior zones, followed by the right lateral and posterior zones, and then repeating the same sequence on the left side.
   3. Perform anterior scanning in the supine position and lateral/posterior scanning using gentle lateral repositioning when clinically feasible. Keep the neonate in the incubator or radiant warmer during scanning whenever possible and avoid unnecessary handling or prolonged repositioning.
   4. For unstable neonates, complete anterior and lateral scanning first and perform posterior scanning only when gentle repositioning is clinically tolerated. Document any unscanned region and the reason for incomplete acquisition.
   5. Record pleural-line abnormalities, B-line pattern, confluent B-lines, consolidation, and atelectatic change.
2. Calculate neonatal lung ultrasound score
   1. Assign regional scores from 0 to 3 using predefined scoring criteria.
   2. Assign score 0 for normal aeration with A-lines.
   3. Assign score 1 for focal or sparse B-lines.
   4. Assign score 2 for confluent B-lines or white lung.
   5. Assign score 3 for consolidation with or without air bronchograms.
   6. Calculate total neonatal lung ultrasound score (nLUS) as the sum of all regional scores.
   7. Record regional scores and total nLUS at baseline, 24 h, and 72 h.
3. Handle incomplete posterior scanning
   1. Document missing posterior regions and reasons for incomplete scanning if repositioning is contraindicated or clinical instability is present.
   2. Retain incomplete cases for sensitivity analyses rather than excluding them.
   3. Use the standardized scoring rubric provided in Supplementary Table S1 during operator training and reviewer calibration.

5. Cranial Ultrasound Acquisition and Neurologic Risk Assessment

1. Acquire cranial ultrasound images
   1. Perform CrUS primarily through the anterior fontanelle with the neonate in the supine position whenever clinically feasible.
   2. Maintain minimal handling and keep the head in a stable neutral position during CrUS acquisition to reduce motion artifacts and physiologic fluctuation.
   3. Acquire standard coronal and sagittal imaging planes through the anterior fontanelle using a consistent scanning sequence.
   4. Assess ventricular morphology, parenchymal echogenicity, and hemorrhage-related findings.
   5. Record structural findings using a predefined reporting format.
2. Assess neurologic abnormalities
   1. Assess IVH-related abnormalities, ventricular enlargement, and abnormal periventricular white matter echogenicity.
   2. Classify IVH findings using the predefined neonatal cranial ultrasound grading framework.
   3. Categorize ventricular enlargement and periventricular echogenicity using predefined reporting criteria.
3. Perform Doppler measurements
   1. Perform pulsed-wave Doppler assessment in the anterior cerebral artery when clinically feasible.
   2. Obtain Doppler measurements during clinically stable periods with minimal handling and stable cardiorespiratory status.
   3. Avoid Doppler acquisition during crying, marked movement, oxygen desaturation, bradycardia, or immediately after major respiratory support adjustment.
   4. Measure RI and PI three times during the same examination.
   5. Enter the mean value into the dataset.

6. Quality Control, Blinding, and Interobserver Consistency

1. Standardize operator training
   1. Complete standardized operator training before study initiation.
   2. Train operators in LUS scanning sequence, nLUS scoring, CrUS structural assessment, and Doppler measurements.
   3. Use a small set of representative training images to calibrate regional LUS scoring, CrUS structural interpretation, and Doppler measurement procedures before formal image review.
2. Perform blinded image interpretation
   1. Interpret images in a blinded manner.
   2. Restrict assessor access to clinical outcomes, severity grouping, surfactant administration status, respiratory support escalation, invasive ventilation status, and integrated model results during scoring.
   3. Separate research image scoring from bedside treatment decision-making. Ultrasound images used for study scoring are reviewed independently by trained assessors who are not involved in clinical treatment decisions whenever feasible.
   4. Base clinical treatment decisions on routine NICU assessment, including respiratory status, oxygen requirement, blood gas results, and attending neonatologist judgment, rather than on the blinded research scoring results.
   5. Release research scoring results for analysis only after clinical grouping, treatment data, and follow-up outcomes have been recorded in the study database.
3. Evaluate image quality
   1. Classify LUS images as evaluable only when pleural-line and vertical artifacts are interpretable.
   2. Classify CrUS images as evaluable only when ventricular and periventricular structures are sufficiently visualized.
   3. Code non-evaluable parameters as missing and document the reason.
   4. Document the reason for non-evaluable images, including motion artifact, incomplete lung-zone acquisition, poor acoustic window, unstable clinical status, or interrupted examination.
4. Assess reproducibility
   1. Perform repeat blinded review using a randomly selected subset of scans.
   2. Perform repeat scoring after a washout interval for intraobserver assessment.
   3. Evaluate agreement for total nLUS using intraclass correlation coefficient (ICC).
   4. Evaluate agreement for categorical CrUS findings using Cohen’s kappa.

7. Variables and Outcomes

1. Record baseline and treatment variables
   1. Record gestational age, birth weight, sex, mode of delivery, antenatal corticosteroid exposure, Apgar scores, and age at NICU admission.
   2. Record oxygen requirement, respiratory support mode, oxygenation-related parameters, and escalation of respiratory support.
   3. Record ultrasound variables including total nLUS, regional findings, dynamic nLUS changes, CrUS findings, RI, and PI.
   4. Record surfactant administration, repeat surfactant dosing, invasive mechanical ventilation, respiratory support duration, and hospital stay duration.
2. Define study outcomes
   1. Define primary outcomes as early NRDS identification and severity progression.
   2. Assess primary severity progression outcomes within the first 72 h after baseline imaging.
   3. Define secondary outcomes as repeat surfactant administration, ventilation-related outcomes, respiratory support duration, hospital stay duration, and progression of abnormal CrUS findings.
   4. Analyze dynamic response using changes in nLUS at 24 h and 72 h.

8. Statistical Analysis

1. Summarize study variables
   1. Summarize normally distributed continuous variables as mean ± standard deviation.
   2. Summarize non-normally distributed variables as median and interquartile range.
   3. Summarize categorical variables as counts and percentages.
2. Perform diagnostic and predictive analyses
   1. Perform receiver operating characteristic (ROC) analysis for diagnostic and predictive evaluation.
   2. Calculate AUC, sensitivity, specificity, positive predictive value, and negative predictive value.
   3. Determine nLUS cutoff values using the Youden index.
   4. Evaluate cutoff performance for NRDS identification, surfactant administration, severity stratification, escalation of respiratory support, and invasive mechanical ventilation prediction.
3. Construct multivariable models
   1. Construct logistic regression models using stepwise analysis.
   2. Include gestational age, birth weight, and baseline respiratory support mode as prespecified confounders.
   3. Compare model discrimination and calibration across models.
   4. Perform decision curve analysis to estimate net clinical benefit.
4. Perform sensitivity and subgroup analyses
   1. Perform complete-case analysis for all primary endpoints.
   2. Perform sensitivity analyses for alternative severity definitions, incomplete regional LUS data, and imaging time-window deviations.
   3. Perform subgroup analyses according to gestational age strata and ventilation strategy when sample size permits.
   4. Use two-tailed statistical testing and define statistical significance as p < 0.05.
   5. Perform all analyses using validated statistical analysis software, with software names and version information provided in the Table of Materials.

Results

Baseline characteristics:

A total of 137 neonates with respiratory distress met the screening criteria and were included in the final analysis. Based on the composite diagnostic reference standard, 92 neonates were classified as NRDS and 45 as non-NRDS respiratory distress. The study population was predominantly preterm, consistent with the NICU admission profile during the study period.

Compared with the non-NRDS group, neonates in the NRDS group had lower gestational age and birth weight, a higher proportion of cesarean delivery and antenatal corticosteroid exposure, and lower 1-min and 5-min Apgar scores. At admission, the NRDS group also showed greater respiratory burden, including higher FiO₂ requirement, lower PaO₂, lower arterial pH, and higher PaCO₂. Respiratory support distribution was consistent with this pattern, with more high-level non-invasive support and more invasive mechanical ventilation in the NRDS group.

Treatment intensity was higher in the NRDS group, with greater use of surfactant therapy within 24 h and more frequent repeat surfactant administration. Abnormal cranial ultrasound findings were more common in the NRDS group, although the between-group difference was not statistically significant at baseline; this variable was therefore retained as a neurologic risk-axis component for subsequent integrated analyses.

These findings support the clinical validity of the study grouping and are summarized in Table 1.

Univariable screening and predictor landscape

Univariable screening was performed for the study’s main analytic tasks: early NRDS identification and early severity/risk stratification. Candidate variables included perinatal baseline characteristics, early respiratory support and blood gas parameters, and ultrasound-derived variables from LUS and CrUS. Variables were ranked by statistical significance and effect strength to support multivariable model construction.

Across early NRDS identification and severity-related analyses, pulmonary ultrasound severity (total nLUS) showed the strongest discriminative signal. Gestational age and birth weight were also consistently ranked among the leading predictors, followed by early respiratory load indicators (FiO₂ requirement and blood gas abnormalities). This pattern indicates that early NRDS discrimination was driven primarily by pulmonary imaging severity, developmental maturity, and gas-exchange impairment.

For neurologic risk-axis analyses based on abnormal CrUS findings, variables reflecting physiologic instability (including oxygenation- and ventilation-related parameters) showed stronger univariable associations than demographic factors alone. Gestational age and nLUS also remained important contributors, supporting the coupling of pulmonary disease burden, developmental immaturity, and early neurologic vulnerability. The ranked predictor landscape is shown in Figure 2.

Early identification of NRDS and incremental value of the combined model

Baseline LUS performed within the admission window (0–6 h) showed strong discriminative ability for NRDS. The nLUS-only model achieved an AUC of 0.884 (95% CI 0.823–0.931), with an optimal cutoff of 8.5, sensitivity of 84.8%, and specificity of 82.2%.

CrUS abnormality alone showed limited discriminative capacity for NRDS diagnosis (AUC 0.577, 95% CI 0.487–0.664), consistent with its role as a neurologic risk-axis variable rather than a primary respiratory diagnostic marker. However, adding CrUS to LUS improved performance (Combined model A: LUS + CrUS, AUC 0.906, 95% CI 0.852–0.946). Further addition of clinical baseline variables improved performance again (Combined model B: Clinical + LUS, AUC 0.922, 95% CI 0.871–0.956), and the fully integrated model (Combined model C: Clinical + LUS + CrUS) showed the best performance (AUC 0.941, 95% CI 0.897–0.972).

The integrated model also provided the best overall classification profile, with sensitivity 90.2%, specificity 88.9%, PPV 94.3%, NPV 82.1%, and accuracy 89.8%. Likelihood ratios improved in parallel (LR+ 8.12; LR− 0.11), indicating stronger rule-in and rule-out utility than single-modality approaches. Comparative model performance is presented in Table 2.

Within the NRDS cohort, the same dual-axis framework was also applied to predict escalation of respiratory support. LUS severity remained the dominant respiratory signal, and CrUS-derived neurologic risk information improved identification of neonates requiring higher-intensity support or intensified management. The added value of CrUS was most apparent in neonates with greater respiratory severity and higher ventilatory support requirements. The integrated assessment framework is illustrated in Figure 3.

Severity grading and threshold determination based on LUS

Within the NRDS cohort (n = 92), baseline nLUS showed a clear gradient across clinical severity strata and escalation of respiratory support categories. Compared with the non-severe NRDS group (n = 58), the severe NRDS group (n = 34) had higher baseline nLUS (12.7 ± 2.9 vs 8.9 ± 2.4, p < 0.001), and the same pattern was observed for median nLUS values (12.8 [10.7–14.9] vs 8.6 [7.3–10.1], p < 0.001). The proportions of neonates with baseline nLUS > 8.5 and > 10.5 were also higher in the severe group (both p < 0.001).

Severity-associated LUS patterns were more frequent in severe NRDS. Confluent B-lines/white lung, consolidation with air bronchograms, and bilateral involvement were all more common in the severe subgroup. These imaging differences were accompanied by higher initial FiO₂ requirement and greater treatment intensity, including more invasive mechanical ventilation at admission, more frequent escalation of respiratory support within 72 h, and greater surfactant use and repeat surfactant administration.

Baseline nLUS was also higher in neonates who later required escalation of respiratory support than in those who did not (11.9 ± 2.9 vs 9.1 ± 2.6; mean difference 2.8, 95% CI 1.7–3.9; p < 0.001). Detailed distributions across severity strata and escalation categories are shown in Table 3.

Dynamic monitoring results were consistent with baseline severity. Improvement in nLUS was smaller in the severe NRDS group than in the non-severe group at both follow-up time points (24 h: −1.3 ± 1.8 vs −2.6 ± 1.7, p = 0.002; 72 h: −2.8 ± 2.5 vs −4.7 ± 2.4, p = 0.001), indicating slower imaging recovery in neonates with greater initial respiratory burden.

After confirming these distribution-level associations, ROC-based threshold analyses were performed to derive clinically usable nLUS cutoffs for severity grading and respiratory support-related outcomes (Figure 4). For severity grading, the optimal cutoffs were 8.6 for mild vs moderate (AUC 0.881), 10.6 for moderate vs severe (AUC 0.833), and 9.7 for mild vs severe (AUC 0.921). For management-relevant outcomes, the optimal cutoffs were 9.5 for surfactant therapy within 24 h (AUC 0.863), 10.5 for escalation of respiratory support within 72 h (AUC 0.892), 12.0 for invasive mechanical ventilation within 72 h (AUC 0.909), and 11.4 for repeat surfactant dosing (AUC 0.846). These results are summarized in Table 4.

Cranial ultrasound–integrated IVH risk stratification

After pulmonary phenotypic stratification and escalation prediction, CrUS variables were incorporated into the early risk framework to support bedside IVH risk stratification. An integrated CrUS IVH risk model was constructed by combining early CrUS findings with clinical variables related to perfusion and hypoxic stress. The model complemented structural CrUS interpretation by generating a continuous risk score for early neurologic risk ranking.

The integrated CrUS model showed better discriminative performance than single-source approaches, indicating that combined CrUS and clinical information provided a more stable neurologic risk signal than imaging or clinical variables alone. Risk scores increased with IVH severity, and higher-score regions contained a greater proportion of neonates with advanced IVH findings, whereas lower-score regions were enriched for neonates without IVH.

Using prespecified score thresholds, the model also provided an interpretable high-risk/low-risk bedside classification that supported decisions on CrUS re-examination timing, monitoring intensity, and prioritization of neuroprotective management. Model performance and risk stratification patterns are shown in Figure 5.

Decision utility for surfactant therapy and management optimization

Because surfactant (PS) initiation is a key early management decision in NRDS, decision performance was compared across routine clinical criteria, an nLUS-guided strategy, and combined LUS+CrUS-guided strategies using the same composite reference standard for “PS indicated” (PS indicated, n = 54; PS not indicated, n = 38).

Conventional clinical criteria identified most PS-indicated neonates but showed a higher false-negative burden (10 false negatives). The nLUS-guided strategy (baseline nLUS ≥ 9.5) improved sensitivity (85.2%) and reduced false negatives (n = 8) while maintaining acceptable specificity (76.3%).

The combined LUS+CrUS-guided strategy (nLUS ≥ 9.5, or nLUS 8.6–9.4 with high IVH-risk score) showed the best overall balance, with 48 true positives and 6 false negatives, corresponding to sensitivity 88.9%, specificity 78.9%, PPV 85.7%, NPV 83.3%, and accuracy 84.8%. This strategy also showed the best rule-out performance among the non-conservative strategies (LR− 0.14).

A more conservative combined strategy (nLUS ≥ 10.5 and no low-risk CrUS pattern) increased specificity (84.2%) and PPV (87.8%) but reduced sensitivity (79.6%) and increased false negatives (n = 11), reflecting a more selective rule-in profile.

Overall, CrUS-derived neurologic risk information was most useful in borderline nLUS intervals, where it improved decision consistency and reduced under-treatment risk without substantially increasing overtreatment. Comparative decision performance is summarized in Table 5.

Dynamic monitoring after surfactant therapy: 24 h and 72 h response patterns

To evaluate the role of LUS in early response assessment after surfactant (PS) administration, longitudinal follow-up was performed in PS-treated neonates at prespecified time points (0 h before PS, 24 h post-PS, and 72 h post-PS). The analysis focused on early response within 24 h and short-term trajectory evolution through 72 h.

At the population level, most PS-treated neonates showed measurable improvement within 24 h, reflected by a decline in nLUS together with improved oxygenation. The magnitude of nLUS reduction at 24 h was significantly associated with improvement in oxygenation-related indices (r = 0.52, p < 0.001), supporting nLUS change as an early imaging response marker after PS. This coupling pattern is shown in Figure 6.

Within the 72 h observation window, nLUS trajectories showed heterogeneity across respiratory severity and CrUS-derived neurologic risk strata. Overall, the decline in nLUS was greater during the first 24 h and slower between 24 h and 72 h. Neonates with higher respiratory severity and/or higher neurologic risk showed smaller reductions in nLUS and greater variability over time, indicating slower or less stable recovery. Stratified trajectory patterns are shown in Figure 7.

Serial imaging changes were accompanied by parallel physiologic improvement. From pre-PS to 24 h and 72 h post-PS, nLUS decreased, FiO₂ requirement declined, and oxygenation improved, including increases in SpO₂/FiO₂ ratio and PaO₂, together with lower PaCO₂ and higher arterial pH. Oxygenation index, respiratory rate, Silverman–Andersen score, and lactate also decreased over time. Mean arterial pressure increased modestly, and heart rate decreased slightly. Longitudinal changes in imaging and clinical variables are summarized in Table 6.

A subset of neonates required escalation of respiratory support or repeat PS within 72 h, and a smaller subset required invasive mechanical ventilation. These events indicate that early improvement at 24 h did not uniformly predict stable short-term recovery and support repeated LUS-based monitoring within the first 72 h after PS.

Figure 1: Study flowchart and ultrasound assessment timeline (A) Flowchart of patient screening, enrollment, diagnostic grouping (NRDS vs non-NRDS), severity stratification within the NRDS cohort, and follow-up for outcomes. (B) The figure also shows the prespecified ultrasound time points, including baseline LUS/CrUS (0–6 h) and repeat LUS at 24 ± 4 h and 72 ± 8 h. Please click here to view a larger version of this figure.

Figure 2: Standardized bedside combined lung and cranial ultrasound acquisition workflow. (A) Bedside ultrasound setup and standardized infant positioning during image acquisition. (B) Twelve-region lung ultrasound scanning map used for neonatal lung ultrasound score (nLUS) calculation. (C) Standard cranial ultrasound imaging planes obtained through the anterior fontanelle. (D) Cerebral Doppler ultrasound acquisition for measurement of cerebral hemodynamic parameters when feasible. (E) Integrated data-output pathway combining LUS, CrUS, Doppler, and clinical variables for diagnostic and severity assessment. Please click here to view a larger version of this figure.

Figure 3: Comparative model performance for early NRDS identification and escalation-risk assessment. (A) ROC curve analysis of LUS alone for early NRDS identification. (B) ROC curve analysis of CrUS abnormalities alone for early NRDS identification. (C) Comparative ROC analysis of combined LUS + CrUS and clinical + LUS models. (D) ROC comparison of the integrated clinical + LUS + CrUS model for prediction of respiratory support escalation risk. Please click here to view a larger version of this figure.

Figure 4: Threshold determination of neonatal lung ultrasound score for severity grading and respiratory support-related outcomes. (A) ROC-based nLUS threshold analysis for NRDS severity grading, including mild versus moderate, moderate versus severe, and mild versus severe NRDS comparisons. (B) ROC-based nLUS threshold analysis for management-related outcomes, including surfactant therapy within 24 h, respiratory support escalation within 72 h, invasive mechanical ventilation within 72 h, and repeat surfactant dosing. (C) Summary schematic of clinically relevant nLUS cutoff values, showing progressively higher nLUS thresholds for NRDS identification, surfactant therapy, respiratory support escalation, repeat surfactant dosing, and invasive mechanical ventilation. Please click here to view a larger version of this figure.

Figure 5: Cranial ultrasound-integrated neurologic risk stratification. (A) Representative CrUS image showing intraventricular hemorrhage-related abnormalities, CrUS image demonstrating ventricular enlargement and periventricular echogenicity changes. (B) Cerebral Doppler ultrasound measurements of resistive index (RI) and (C) pulsatility index (PI). (D) Integrated neurologic risk classification based on CrUS structural and Doppler findings. Please click here to view a larger version of this figure.

Figure 6: Early response of neonatal lung ultrasound score within 24 h after surfactant therapy. Representative LUS images and quantitative nLUS changes (A) representative lung ultrasound image before surfactant therapy (Pre-PS, 0 h); (B) representative lung ultrasound image at 24 h after surfactant therapy (Post-PS, 24 h); (C) correlation between the change in neonatal lung ultrasound score (ΔnLUS, 0 h to 24 h) and oxygenation improvement; and (D) paired change in nLUS from Pre-PS to 24 h Post-PS. Please click here to view a larger version of this figure.

Figure 7: Longitudinal neonatal lung ultrasound score trajectories up to 72 h after surfactant therapy. (A) representative lung ultrasound image pattern of a non-responder/slow recovery case; (B) representative lung ultrasound image pattern of a responder/rapid recovery case; (C) the overall longitudinal trajectory of nLUS at 0 h, 24 h, and 72 h after surfactant therapy; and (D) stratified longitudinal nLUS trajectories according to respiratory severity and cranial ultrasound risk strata. Please click here to view a larger version of this figure.

Characteristics	Total n = 137	NRDS n = 92	Non-NRDS n = 45	P value
Gestational age, weeks	32.8 ± 3.4	31.9 ± 3.2	34.5 ± 2.9	<0.001
Birth weight, g	1926 ± 642	1758 ± 598	2278 ± 621	<0.001
Male sex, n (%)	79 (57.7)	55 (59.8)	24 (53.3)	0.463
Cesarean delivery, n (%)	83 (60.6)	61 (66.3)	22 (48.9)	0.049
Antenatal corticosteroids, n (%)	69 (50.4)	54 (58.7)	15 (33.3)	0.006
1-min Apgar score	6.8 ± 1.5	6.4 ± 1.4	7.5 ± 1.3	<0.001
5-min Apgar score	8.1 ± 1.2	7.8 ± 1.1	8.7 ± 1.0	<0.001
Small for gestational age, n (%)	21 (15.3)	17 (18.5)	4 (8.9)	0.138
Maternal hypertension, n (%)	29 (21.2)	22 (23.9)	7 (15.6)	0.266
Maternal diabetes, n (%)	18 (13.1)	11 (12.0)	7 (15.6)	0.553
Initial FiO ₂ requirement	0.38 ± 0.12	0.42 ± 0.11	0.31 ± 0.10	<0.001
Initial arterial pH	7.29 ± 0.07	7.27 ± 0.06	7.33 ± 0.07	<0.001
PaO ₂ , mmHg	61.4 ± 12.8	58.2 ± 11.9	68.1 ± 12.3	<0.001
PaCO ₂ , mmHg	46.9 ± 8.7	49.2 ± 8.3	42.3 ± 7.6	<0.001
Nasal oxygen, n (%)	34 (24.8)	12 (13.0)	22 (48.9)	<0.001
CPAP, n (%)	56 (40.9)	42 (45.7)	14 (31.1)	0.095
NIPPV, n (%)	31 (22.6)	25 (27.2)	6 (13.3)	0.071
Invasive mechanical ventilation, n (%)	16 (11.7)	13 (14.1)	3 (6.7)	0.215
Surfactant therapy within 24 h, n (%)	58 (42.3)	54 (58.7)	4 (8.9)	<0.001
Repeat surfactant administration, n (%)	19 (13.9)	18 (19.6)	1 (2.2)	0.004
Cranial ultrasound abnormality, n (%)	25 (18.2)	20 (21.7)	5 (11.1)	0.129

Table 1: Baseline characteristics of enrolled neonates and initial respiratory support profiles. Comparison of baseline demographic, perinatal, physiologic, and treatment variables between the NRDS and non-NRDS groups.

Model / test	AUC	Optimal cutoff	Sensitivity, %	Specificity, %	PPV, %	NPV, %	Accuracy, %	LR+	LR−
	(95% CI)		(95% CI)	(95% CI)	(95% CI)	(95% CI)
LUS score alone (0–6 h)	0.884 (0.823–0.931)	≥ 8.5	84.8 (75.9–91.2)	82.2 (68.0–91.7)	90.3 (82.1–95.1)	73.5 (60.1–83.6)	84	4.77	0.19
CrUS abnormality alone	0.577 (0.487–0.664)	Abnormal vs normal	21.7 (13.9–31.5)	88.9 (75.9–96.3)	80.0 (59.3–93.2)	36.1 (27.8–45.0)	43.8	1.95	0.88
Combined model A: LUS + CrUS	0.906 (0.852–0.946)	≥ 0.66	87.0 (78.5–93.0)	84.4 (70.5–93.5)	92.0 (84.4–96.5)	76.0 (62.8–86.0)	86.1	5.58	0.15
Combined model B: Clinical + LUS	0.922 (0.871–0.956)	≥ 0.63	88.0 (79.7–93.7)	86.7 (73.2–94.9)	93.2 (86.0–97.2)	78.0 (65.1–87.9)	87.6	6.6	0.14
Combined model C: Clinical + LUS + CrUS	0.941 (0.897–0.972)	≥ 0.60	90.2 (82.4–95.2)	88.9 (75.9–96.3)	94.3 (87.8–97.8)	82.1 (69.6–91.1)	89.8	8.12	0.11

Table 2: Diagnostic performance for early NRDS identification: LUS alone versus combined LUS+CrUS and clinical models Comparative diagnostic metrics for early NRDS identification across single-modality and integrated models, including AUC, sensitivity, specificity, PPV, NPV, and likelihood ratios. The table quantifies the incremental value of adding CrUS and baseline clinical variables to LUS.

Metric/Category	Overall NRDS n = 92	Non-severe NRDS n = 58	Severe NRDS n = 34	P value
Baseline nLUS score (0–6 h), mean ± SD	10.3 ± 3.2	8.9 ± 2.4	12.7 ± 2.9	<0.001
Baseline nLUS score, median (IQR)	10.2 (8.1–12.4)	8.6 (7.3–10.1)	12.8 (10.7–14.9)	<0.001
Baseline nLUS > 8.5, n (%)	64 (69.6)	33 (56.9)	31 (91.2)	<0.001
Baseline nLUS > 10.5, n (%)	43 (46.7)	14 (24.1)	29 (85.3)	<0.001
Confluent B-lines / white lung, n (%)	57 (62.0)	28 (48.3)	29 (85.3)	<0.001
Consolidation with air bronchograms, n (%)	31 (33.7)	14 (24.1)	17 (50.0)	0.012
Pleural line abnormalities, n (%)	71 (77.2)	42 (72.4)	29 (85.3)	0.162
Bilateral involvement, n (%)	68 (73.9)	38 (65.5)	30 (88.2)	0.018
Initial FiO₂ requirement	0.44 ± 0.10	0.40 ± 0.09	0.50 ± 0.09	<0.001
CPAP, n (%)	42 (45.7)	31 (53.4)	11 (32.4)	0.049
NIPPV, n (%)	25 (27.2)	18 (31.0)	7 (20.6)	0.278
Invasive mechanical ventilation at admission, n (%)	13 (14.1)	2 (3.4)	11 (32.4)	<0.001
Respiratory support escalation within 72 h, n (%)	39 (42.4)	12 (20.7)	27 (79.4)	<0.001
Time to escalation, h, median (IQR)	14.6 (7.8–26.9)	22.4 (14.2–33.7)	9.7 (5.6–17.9)	<0.001
Surfactant therapy within 24 h, n (%)	54 (58.7)	25 (43.1)	29 (85.3)	<0.001
Repeat surfactant dose, n (%)	18 (19.6)	3 (5.2)	15 (44.1)	<0.001
nLUS change at 24 h, Δ score	−2.1 ± 1.8	−2.6 ± 1.7	−1.3 ± 1.8	0.002
nLUS change at 72 h, Δ score	−4.0 ± 2.6	−4.7 ± 2.4	−2.8 ± 2.5	0.001

Table 3: Distribution of lung ultrasound severity and clinical management indicators across NRDS severity strata and escalation categories Baseline nLUS values, key LUS patterns, and treatment-related indicators across non-severe vs severe NRDS and escalation vs no-escalation groups. The table supports the role of nLUS as a severity marker and links imaging burden to treatment intensity and short-term progression.

Prediction task / outcome	AUC	Optimal nLUS cutoff	Sensitivity, %	Specificity, %	PPV, %	NPV, %	Youden index
	(95% CI)
Mild vs moderate NRDS	0.881 (0.792–0.942)	≥ 8.6	83.3	81.8	78.9	85.7	0.651
Moderate vs severe NRDS	0.833 (0.737–0.906)	≥ 10.6	79.4	76.5	73	82.5	0.559
Mild vs severe NRDS	0.921 (0.849–0.965)	≥ 9.7	88.2	84.8	81.1	90.9	0.73
Surfactant therapy within 24 h	0.863 (0.777–0.927)	≥ 9.5	85.2	76.3	83.6	78.4	0.615
Respiratory support escalation within 72 h	0.892 (0.811–0.947)	≥ 10.5	82.1	84.9	80	86.5	0.67
Invasive mechanical ventilation within 72 h	0.909 (0.832–0.960)	≥ 12.0	81.3	88.2	64	95.2	0.695
Repeat surfactant dosing	0.846 (0.756–0.913)	≥ 11.4	77.8	81.1	50	93.8	0.589

Table 4: Optimal nLUS thresholds for severity grading and prediction of respiratory support-related outcomes ROC-derived cutoff values and diagnostic indices for nLUS across severity grading and management-relevant tasks (e.g., surfactant use, escalation of support, invasive mechanical ventilation). The table provides the threshold hierarchy used for bedside stratification and decision support.

Decision strategy		Decision rule		True positives	False positives	True negatives	False negatives	Sensitivity, %	Specificity, %	PPV, %	NPV, %	Accuracy, %	LR+	LR−
Routine clinical criteria		Clinical indication based on respiratory support and oxygenation status		44	7	31	10	81.5	81.6	86.3	75.6	81.5	4.43	0.23
nLUS-guided strategy		Baseline nLUS ≥ 9.5		46	9	29	8	85.2	76.3	83.6	78.4	81.5	3.59	0.19
Combined LUS + CrUS-guided strategy		nLUS ≥ 9.5, or nLUS 8.6–9.4 with high IVH-risk score		48	8	30	6	88.9	78.9	85.7	83.3	84.8	4.21	0.14
Conservative combined strategy		nLUS ≥ 10.5 and no low-risk CrUS pattern		43	6	32	11	79.6	84.2	87.8	74.4	81.5	5.04	0.24

Table 5: Decision performance for surfactant therapy: comparison of routine clinical criteria, LUS-guided strategy, and LUS+CrUS-guided strategies Comparison of decision classification results (true/false positive/negative) and diagnostic metrics for PS initiation strategies. The table shows how adding LUS, and then CrUS in borderline nLUS ranges, affects sensitivity-specificity balance and overall decision consistency.

Variable	Pre-surfactant baseline	24 h post-surfactant	72 h post-surfactant	Overall P value
nLUS score	11.4 ± 2.8	9.3 ± 2.7	7.4 ± 2.8	<0.001
Δ nLUS from baseline	Reference	− 2.1 ± 1.8	− 4.0 ± 2.6	<0.001
FiO ₂ requirement	0.46 ± 0.11	0.35 ± 0.10	0.29 ± 0.08	<0.001
SpO ₂ /FiO ₂ ratio	207 ± 52	270 ± 64	327 ± 71	<0.001
PaO ₂ , mmHg	57.8 ± 11.5	66.9 ± 12.3	74.2 ± 13.1	<0.001
PaCO ₂ , mmHg	50.1 ± 8.4	45.8 ± 7.6	42.1 ± 7.2	<0.001
Arterial pH	7.27 ± 0.06	7.32 ± 0.06	7.35 ± 0.05	<0.001
Oxygenation index	7.8 ± 2.6	5.6 ± 2.1	4.2 ± 1.7	<0.001
Respiratory rate, breaths/min	68 ± 12	58 ± 11	51 ± 10	<0.001
Silverman-Andersen score	5.1 ± 1.4	3.7 ± 1.3	2.6 ± 1.2	<0.001
Lactate, mmol/L	2.8 ± 0.9	2.1 ± 0.8	1.6 ± 0.6	<0.001
Mean arterial pressure, mmHg	39.5 ± 6.8	41.8 ± 6.5	43.1 ± 6.2	0.018
Heart rate, beats/min	152 ± 18	145 ± 16	139 ± 15	0.004
Respiratory support escalation within 72 h, n (%)	—	15 (25.9)	21 (36.2)	—
Repeat surfactant therapy within 72 h, n (%)	—	12 (20.7)	18 (31.0)	—
Invasive mechanical ventilation within 72 h, n (%)	—	7 (12.1)	11 (19.0)	—

Table 6: Longitudinal changes in nLUS and key clinical/physiologic variables after surfactant therapy Serial imaging and physiologic measurements at baseline (pre-PS), 24 h post-PS, and 72 h post-PS. The table summarizes dynamic response patterns, including nLUS improvement, oxygenation changes, and short-term clinical events such as escalation of support and repeat PS.

Supplementary Table S1: Standardized neonatal lung ultrasound (nLUS) regional scoring rubric and interpretation criteria used for training, calibration, and study scoringPlease click here to download this file. Detailed scoring definitions, interpretation rules, evaluability criteria, and score-calculation procedures for the 12-region nLUS protocol. This table supports reproducible image scoring and consistent application of the LUS-based severity framework across baseline and follow-up scans.

Discussion

This study developed and evaluated a structured bedside ultrasound pathway for neonatal respiratory distress syndrome (NRDS), using lung ultrasound (LUS) as the primary respiratory severity axis and cranial ultrasound (CrUS) as a neurologic risk axis¹⁵. Across the early admission window and short-term follow-up, the framework supported three practical tasks in NICU care: early NRDS identification, severity stratification with escalation prediction, and dynamic response monitoring after surfactant (PS) therapy. The value of this approach lies less in adding another imaging test and more in organizing bedside assessment into a repeatable, time-linked workflow that aligns imaging findings with treatment decisions.

The results show that baseline nLUS functioned as a stable marker of pulmonary disease burden, not only a diagnostic discriminator. Higher nLUS values were consistently associated with worse oxygenation, greater respiratory support intensity, and higher surfactant use, indicating that nLUS captured the severity of aeration loss and diffuse pulmonary involvement in clinically meaningful terms¹⁶. This is also reflected in the threshold analyses, where nLUS cutoffs remained useful across multiple management-relevant outcomes, including surfactant treatment, escalation of support, and invasive mechanical ventilation¹⁷. Compared with isolated clinical thresholds, a standardized nLUS-based approach provides a more direct and reproducible severity signal at the bedside¹⁸.

A second finding is that CrUS added value mainly as a contextual risk modifier rather than a primary diagnostic tool for NRDS. CrUS alone showed limited discriminative performance for NRDS, which is expected, but its contribution became clearer when combined with LUS and clinical variables, especially in borderline respiratory-risk cases. In those settings, CrUS-derived risk information improved classification and decision consistency, suggesting that neurologic vulnerability helps identify neonates with less stable early trajectories¹⁹. This interpretation is consistent with the dual-axis design of the protocol, in which pulmonary severity and neurologic risk are assessed in parallel rather than treated as separate bedside problems.

The dynamic monitoring results after PS therapy further support the clinical usefulness of serial LUS. Most PS-treated neonates showed measurable nLUS improvement within 24 h, and the magnitude of nLUS decline correlated with oxygenation improvement (r = 0.52, p < 0.001), supporting nLUS change as an early treatment-response marker²⁰. The 72 h follow-up showed heterogeneous recovery trajectories, with slower or less stable improvement in higher-severity and higher-risk strata. This pattern supports repeated reassessment within a predefined time window rather than a single post-treatment scan, and it strengthens the practical role of LUS in short-term monitoring and escalation planning²¹. In the same direction, the surfactant decision analysis showed fewer false negatives with LUS-guided and LUS+CrUS-guided strategies than with routine clinical criteria, which is clinically relevant when early under-treatment is a concern²².

Several limitations should be considered. First, this was a single-center study, and the case mix, treatment thresholds, and imaging workflow reflect local NICU practice; the proposed cutoffs therefore require external validation before broader use. Second, although acquisition and scoring were standardized, ultrasound interpretation remains partly operator-dependent, and inter-center differences in scanning technique may affect reproducibility²³. Third, the CrUS-integrated risk framework improved bedside stratification, but this study was not designed to establish a causal mechanism linking pulmonary severity to neurologic injury risk. In addition, the analysis focused on early and short-term clinical outcomes, so the prognostic value of these ultrasound trajectories for longer-term respiratory or neurodevelopmental outcomes remains uncertain.

Future work should prioritize multicenter validation of the nLUS thresholds and the integrated Clinical + LUS + CrUS model under harmonized imaging and quality-control protocols. A useful next step would be to test whether early trajectory-based ultrasound markers (rather than single baseline scores alone) improve prediction of clinically important outcomes, including prolonged respiratory support, repeat surfactant exposure, and follow-up neurodevelopmental risk. If confirmed, this would strengthen the role of a dual-axis ultrasound workflow as a practical decision-support structure in NICU care, with LUS providing the primary severity signal, CrUS adding neurologic context, and serial reassessment improving the timing and consistency of early management.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CrUS-Structural-Doppler-Worksheet-v1.0	Study-generated	version date: 21 Apr 2026; XLSX format	Worksheet used to record anterior fontanelle views, structural CrUS findings, IVH-related abnormality, ventricular enlargement, periventricular echogenicity, RI, PI, and Doppler evaluability.
De-identified image archive naming convention	Study-generated	ImageArchive-NamingRule-v1.0; version date: 21 Apr 2026	Naming rule for saved images and clips: StudyID_Modality_TimePoint _RegionOrPlane_FileType_OperatorCode. Example: NRDS001_LUS_T0_RAU_Cine_OP01.
High-frequency linear ultrasound transducer	GE HealthCare	12L-RS linear array probe; catalog / part no. 5499501	Linear probe used for neonatal lung ultrasound. Use for pleural-line visualization, A-line/B-line assessment, subpleural consolidation detection, and regional nLUS scoring.
Lung ultrasound scoring worksheet	Study-generated	nLUS-12Region-Worksheet-v1.0; version date: 21 Apr 2026; XLSX format	Worksheet used to enter 12 regional LUS scores, total nLUS, missing regions, reason for incomplete scanning, and ΔnLUS at 24 h and 72 h.
Micro-convex ultrasound transducer	GE HealthCare	8C-RS micro-convex array probe; catalog / part no. 5499508	Micro-convex probe used for cranial ultrasound through the anterior fontanelle and for pulsed-wave Doppler acquisition when feasible.
Probe cleaning and disinfection SOP	Study-generated	SOP-LUS-CrUS-Disinfection-v1.0; version date: 21 Apr 2026	Defines cleaning sequence before and after bedside ultrasound, including gel removal, probe-handle wiping, machine surface wiping, and documentation of cleaning completion.
ROC and diagnostic statistics software	MedCalc Software Ltd.	MedCalc Statistical Software 20.2	Used for ROC analysis, AUC calculation, AUC comparison, Youden-index cutoff selection, sensitivity, specificity, PPV, NPV, LR+, and LR−.
R package for decision curve analysis	CRAN / open-source R package	rmda 1.6 or dcurves 0.5.0	Use one package consistently for decision curve analysis. Record the exact package and version in the final analysis log.
R package for ROC analysis or visualization	CRAN / open-source R package	pROC 1.18.5	Optional R package for secondary ROC visualization or sensitivity checks. Use the same ROC method as the primary analysis plan if results are reported.
R package for plotting	CRAN / open-source R package	ggplot2 3.4.4	Used for reproducible plots, including predictor ranking, calibration curves, and trajectory plots when figures are generated in R.
Routine ultrasound coupling gel	Parker Laboratories	Aquasonic 100 Ultrasound Transmission Gel; product no. 01-08; 0.25 L dispenser, 12 per box	Use for routine intact-skin bedside lung ultrasound when local infection-control policy permits non-sterile gel. Use a consistent gel type across examinations to reduce image-quality variability.
Statistical analysis software	IBM	SPSS Statistics 26.0	Used for descriptive statistics, group comparisons, logistic regression, and basic model summaries.
Statistical computing environment	R Foundation for Statistical Computing	R 4.3.2	Used for calibration plots, decision curve analysis, sensitivity analyses, and reproducible figure generation when applicable. Freeze package versions before final analysis.
Sterile ultrasound coupling gel	Parker Laboratories	Sterile Aquasonic 100 Ultrasound Transmission Gel; product no. 01-01; 20 g sterile overwrapped foil pouch, 48 per box	Use for neonatal cranial ultrasound or any examination requiring sterile single-use gel according to NICU infection-control policy. Use one pouch per patient whenever sterility is indicated.
Surface disinfectant wipes for ultrasound equipment	PDI Healthcare	Super Sani-Cloth Germicidal Disposable Wipes; product no. Q55172	Use for external surface disinfection of ultrasound probe handles, machine contact surfaces, and bedside equipment according to local NICU infection-control policy. Avoid using disinfectants on probe surfaces unless compatible with the manufacturer’s probe-care instructions.