Prediction of primary blast lung injury outcomes in goats using CT and injury factors

Abstract

Purpose

To investigate the feasibility of using CT combined with injury factors to predict the lethality and severity of primary blast lung injury through establishing a goat model simulating real explosion effects.

Methods

High explosives were detonated at two altitudes in natural fields, with goats positioned at distances of 2 to 6 meters from the explosion center, facing either the right side or the front chest. CT features and quantitative data post-injury were recorded, along with blood gas analysis in surviving goats and lung coefficient measurements. Correlation analysis between CT severity score and lung coefficient was performed. ROC curve analysis was conducted to evaluate the multi-factors in distinguishing severity (severe/serious vs. moderate/slight) and lethality (death vs. survival). Nomograms based on injury factors and CT were developed to predict the risk of severe/serious injury and death.

Results

After the experiment, a total of 87 goats were classified as severe/serious lung injury through pathology, including 40 goats that died on the spot. High altitude, close distance, and right-side orientation exacerbated lung injury severity, with close distance being the primary cause of death. Lung coefficient, CT score, and volume exhibited excellent performance in distinguishing outcomes (AUC: 0.967, 0.925, and 0.924 for lethality; 0.944, 0.915, and 0.909 for severity, respectively), with no significant difference in diagnostic efficacy. Moderate correlations were observed between lung coefficient and CT score for severity (r=0.440; 0.480), with a strong correlation for survival but none for death. Volume emerged as the primary risk factor, and the nomograms incorporating multiple factors demonstrated personalized predictive abilities.

Conclusion

CT combined with altitude, distance, and orientation can effectively predict the lethality and severity of primary blast lung injury, providing significant value for accurate injury assessment and clinical management.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00068-025-03014-2.

Introduction

With the intensification of human industrialization and military conflicts, the incidence of primary blast lung injury (PBLI) has been on the rise, with the lung being the most susceptible organ to damage [1–3]. PBLI primarily manifests as pulmonary hemorrhage and pulmonary edema, and if left untreated, it can progress to acute respiratory distress syndrome (ARDS) and multiple organ dysfunction syndrome (MODS), posing a significant threat to life [4–8]. Air embolism into the systemic circulation can even lead to immediate death [9].

Research on PBLI has a long history, encompassing various aspects such as blast wave damage effects, chest dynamic mechanical response, and the development of injury criteria and scoring systems [10]. However, current assessment methods for lung injury largely rely on comprehensive consideration of vital signs or pathology, which can be subjective, vague, or invasive. Non-invasive criteria and parameters that directly reflect the degree of lung injury remain scarce.

Some current studies have investigated the factors influencing PBLI. Deng et al. simulated a hypoxic and low-pressure plateau environment to establish an animal model of explosion-induced lung injury [11]. Their results revealed that the decrease in air pressure was closely related to the exacerbation of lung injury severity and the increase in mortality. In recent years, advancements in computed tomography (CT) have significantly enhanced the visualization of lung injury features. However, its application for assessing biological damage effects in complex field settings remains underdeveloped. Current CT research on PBLI is limited by small sample sizes or simplified conditions, and its generalizability is not yet sufficient for application in the classification of lung injury severity and prognostic analysis. In summary, the key gap lies in the lack of standardized methods to correlate CT imaging biomarkers with biological injury outcomes under real-world explosive conditions. This limits its clinical application, especially in military or disaster scenarios.

Here, In this study, we aimed to address these limitations by establishing a large sample model using goats to simulate the real explosive biological damage effects and evaluate the lethality and severity of PBLI at different altitudes, distances, and orientations through CT scans. Additionally, this study developed an intuitive and effective nomogram by combining CT-derived biomarkers with explosion-specific variables to predict individualized risks of severely lung injury and death. Traditional assessment methods rely on subjective clinical signs or invasive procedures, whereas our nomogram provides a non-invasive, quantitative tool tailored to dynamic field conditions. The results of this study not only serve as a key reference for preclinical research but also lay the foundation for translating CT-based nomograms into clinical protocols. By integrating imaging data with environmental injury factors, our study offers a new framework to further optimize real-time injury assessment and resource allocation in complex settings.

Methods

Experimental animals

Goats were acquired from the Geermu Animal Breeding Center. All methods were conducted in accordance with the relevant guidelines and regulations, including the Regulations on the Administration of Laboratory Animals (2017 revision) by the State Council of the People’s Republic of China, and the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines for reporting animal research. The study protocol was approved by the Animal Ethics Committee of Army Medical University (Approval No. AMUWEC20202140), ensuring that all animal experiments were performed in compliance with ethical standards. All experimental goats were housed individually in 3 m × 2 m pens with 6 m²space each. The facility was maintained at 20 ± 3 °C, 40–60% humidity, with natural ventilation and a 12 h light/dark cycle. Pens had concrete floors with straw bedding, automatic waterers, and were fed alfalfa hay twice daily.

Experimental setup

First, an appropriate temporary medical center was established, equipped with professional medical personnel and the vehicle-mounted CT scanner was calibrated. Several kilometers away from the temporary medical center, an open area with sunny weather and suitable humidity and temperature was selected as the explosion test site(wind speed at Level 3, relative humidity ranging from 9% to 33% and ambient temperature between 10 °C and 15 °C). Initially, 253 goats were included to conduct the explosion experiments in the natural field. Each goat was assigned to a specific combination of distance (2, 2.5, 3, 4, 5, and 6 m), altitude (plain, altitude = 200 m and plateau, altitude = 3300 m), and orientation towards explosion center(right or front chest) using a random number table. The explosion experiments were conducted in groups and phases, with the number of goats in each group controlled at 16 to 20. Before the explosion, the goats were secured to a specially designed rigid frame with adjustable straps and padding, positioning their right side of the chest or the front of the chest towards the explosion center, and this orientation was confirmed visually. During the explosion experiments, the restraint device was anchored to a stable platform or surface to ensure that it remained stationary during the explosion. After the explosion, the goats in each group were transported to the temporary medical center by vehicle for subsequent data collection.

The second generation of new thermobaric ammunition with an 8 kg TNT equivalent static explosion was used for the tests, with a core height of 0.8 m. Figure 1 shows the experimental flow diagram.

Fig. 1.

The workflow of the study

Data collection and pathological grading

At the temporary medical center, professional medical personnel first confirmed the survival status of each group of goats. Dead goats were directly subjected to CT scanning, while surviving goats first had their blood gas analysis data collected, followed by CT scanning. These processes could be carried out simultaneously. The surviving goats were eventually euthanized by intravenous injection of sodium pentobarbital(Dose of injection:150 mg/kg, Concentration:3%). Finally, all goats were dissected by pathology experts to observe lung injuries. An incision was made at the sternum, and the lung lobes were carefully identified and separated from the surrounding tissues such as the pleura, ribs, and intercostal muscles. The connection between the lungs and the trachea was located and severed, and the lungs were completely removed, weighed, and photographed. The examination of each group of goats was completed within 24 h. Based on the gross anatomical findings, the Pathologic Severity Scale of Lung Blast Injury (PSSLBI) was used to quantify the four levels of PBLI severity—sight, moderate, serious and severe—according to the degree and extent of pulmonary hemorrhage, edema and laceration, with scores of 1, 2, 3, or 4, respectively [12]. The lung coefficient (lung weight [g]/body weight [kg]) was also calculated.

CT examination and feature identification

Vehicle-mounted 16 row multi slice spiral CT (MinFound, China) was used to obtain images of the lungs of goats. Scanning parameters were as follows: tube voltage 120 kV, tube current 200mAs, layer thickness 1.0 mm, layer spacing 1.0 mm, matrix 512 × 512.

Two experienced chest radiologists, blinded to the pathological outcomes, independently assessed the CT scan of the lungs post-explosion. In cases of disagreement, a third senior radiologist was consulted for assistance in the evaluation. PBLI imaging features include: ①contusion; ②lacerations; ③hematoma; ④atelectasis; ⑤ pneumothorax; ⑥hemothorax; ⑦distribution of lesions. The density classification, number of involved lung lobes and single or double lung injury were recorded.

Quantization parameters and semi-quantitative score of imagings

The open-source software 3D Slicer (version 5.6.1, https://www.slicer.org/) was used to outline the ROI of lung injury. By meticulously outlining the injury boundaries on a slice-by-slice basis, the volume calculation was ensured to reflect only the areas damaged by the blast wave, excluding healthy lung tissue. Normal tissues, such as ribs, vessels, and areas of misidentification, were manually excluded through visual inspection of the segmented regions and reference to CT value thresholds, which are consistent with the typical density differences between lung tissue and bone/vasculature. Ultimately, the “grow from seed” function was utilized to automatically generate a three-dimensional image of the entire voxel, thereby obtaining the volume of interest (VOI), from which the mean density value of the lung injury area was calculated. After the delineation was completed, the original images and VOIs were saved as the required format files and marked one by one with goat numbers.

A visual semi-quantitative score system was implemented to assess the degree of involvement of each lung lobe based on the diameter of the largest lesion present in each lobe. Each lobe was scored in a four-point scale: score 1, diameter < 1 cm; score 2, 1 cm < diameter < 3 cm; score 3, diameter >3 cm and up to 50% of the lobe; score 4, >50% of the lobe [13]. All 7 lobe scores were summed to calculate the total lung scores.

Statistical analysis

An SPSS statistical software (version 26.0; IBM) was used to process the data. Continuous variables were described as mean ± standard deviation or median with inter-quartile range. The χ² test, t-test or Mann-Whitney U test was used to compare the differences between the groups. Correlations between lung coefficient and CT score at severity grouping were analysed with spearman correlation. P < 0.05 was considered significantly different. The Receiver Operating Characteristic curves (ROC curves) were used to evaluate the diagnostic performance for death and severe/serious risks, and the best cut-off value, sensitivity and specificity were calculated. The DeLong test was used to compare the differences between the ROC curves of variables.

To predict death and severe/serious risk, logistic regression analysis was used for univariate and multivariate analysis to obtain odds ratios (OR) and develop nomograms using weighted estimators corresponding to each covariate. The covariates were derived from regression coefficients and variance estimates using R software (version 4.3.2; http://www.Rproject.org).

Results

Baseline characteristics and CT

After preliminary screening, 33 goats were excluded due to no injuries, poor imaging quality or other unexpected reasons, and the baseline data and CT images of 220 (weight: 35.524 ± 5.847 kg) goats were ultimately included in the evaluation.

Of the 220 goats, 98 were in the plateau explosion field, and 122 were in the plain explosion field. The goats facing the explosion center were fixed with their right side (n = 174) and front chest (n = 46), respectively. The explosion distance was evaluated in groups of 2–3 m (n = 98) and 4–6 m (n = 122). According to real-time survival status and subsequent classification of PSSLBI, 41 goats died after explosion and 179 goats survived; 87 goats were classified as the severe/serious and 133 as the moderate/slight. The 220 goats contributed whole CT data. 134 goats were measured for lung coefficient, while blood gas indicators were obtained from 139 surviving goats.

The above parameters were independently analyzed by grouping (survival and death, severe/serious and moderate/slight) and summarized in Table 1. The analysis showed that close distance explosions was the main cause of death, accompanied by increases in lung coefficient, CT score, and injury volume. The CT morphological indicators for distinguishing death and survival included damage location, distribution, pneumothorax, hemothorax and involved lung lobes. Altitude, distance, and orientation were all related to the severity of the injury, and there were also differences in lung coefficient and blood gas indicators. The CT for distinguishing severity were score, volume, mean density value, damage location, distribution, density classification, lacerations and involved lung lobes. Partial cases are shown in Fig. 2.

Table 1.

Variables of lung injury on CT and results of baseline

Variable	Death n = 41	Survival n = 179	P value	Severe/Serious n = 87	Moderate/Slight n = 133	P value
region
plain	18	104	0.099	41	81	0.044*
plateau	23	75		46	52
distance(m)
2 ~ 3	35	63	0.000*	62	36	0.000*
4 ~ 6	6	116		25	97
orientation
right side	35	139	0.273	81	93	0.000*
front chest	6	40		6	40
lung coefficient(n = 134)	23.62 ± 1.53	12.15(10.04 ~ 13.66)	0.000*	13.87(12.04 ~ 21.25)	8.76 ± 1.93	0.000*
PaCO2(n = 139)	/	28.81 ± 4.47	/	27.70(23.70 ~ 30.80)	29.44 ± 4.26	0.018*
PaO2(n = 139)	/	52.92 ± 16.23	/	42.83 ± 1.76	56.63 ± 16.18	0.000*
SaO2(n = 139)	/	89.00(79.00 ~ 94.00)	/	81.00(71.00 ~ 93.00)	91.00(83.00 ~ 95.00)	0.000*
score	23.00(18.00 ~ 26.00)	9.00(6.00 ~ 14.00)	0.000*	18.16 ± 5.75	7.00(5.00 ~ 11.00)	0.000*
volume	934.22 ± 63.65	203.79(68.04 ~ 363.84)	0.000*	629.58(369.19 ~ 938.16)	136.07(53.19 ~ 264.36)	0.000*
mean density values(Hu)	−374.79(−404.58~−323.75)	−388.20(−461.54~−314.28)	0.226	−339.04 ± 82.82	−418.07 ± 113.75	0.000*
damage location
both lung	41	153	0.009*	87	107	0.000*
single lung	0	26		0	26
distribution
diffused	34	29	0.000*	46	17	0.000*
scattered	7	150		41	116
density classification
opacification	32	127	0.360	49	110	0.000*
density	9	52		38	23
contusions
presence	41	179	/	87	133	/
absent	0	0		0	0
lacerations
presence	4	5	0.111	7	2	0.017*
absent	37	174		80	131
hematoma
presence	0	2	0.497	1	1	0.761
absent	41	177		86	132
pneumothorax
presence	10	13	0.001*	10	13	0.684
absent	31	166		77	120
atelectasis
presence	5	21	0.934	8	18	0.330
absent	36	158		79	115
hemothorax
presence	3	0	0.000*	3	0	0.118
absent	38	179		84	133
involved lung lobes	7(7 ~ 7)	4(3 ~ 5)	0.000*	7(5 ~ 7)	4(2 ~ 5)	0.000*

*P < 0.05

Fig. 2.

Images of lung injuries and corresponding gross anatomy. (a)Number: 870, death, plateau, 3 m, right-orientation (severe, PSSLBI score is 4). diffuse butterfly distribution contusion of both lungs; score: 22; volume: 904.053cm³; mean density value: −402.821hu; density classification: opacification; involved lung lobes: 7; (b)Number: 524, survival, plain, 3 m, right-orientation (serious, PSSLBI score is 3). scattered patchy distribution of both lungs; score: 15; volume: 306.634cm³; mean density value: −166.836hu; density classification: density; involved lung lobes: 7; lung coefficient: 15.51; PaCO₂: 31mmHg; PaO₂: 40mmHg; SaO₂: 80%; (c) Number: 986, survival, plateau, 4 m, right-orientation (moderate, PSSLBI score is 2). scattered patchy distribution of right lungs; score: 11; volume: 211.121cm³; mean density value: −318.744hu; density classification: density; involved lung lobes: 4; lung coefficient: 10.62; PaCO₂: 25mmHg; PaO₂: 39mmHg; SaO₂: 72%; (d) Number: 962, survival, plateau, 4 m, front-orientation (slight, PSSLBI score is 1). scattered patchy distribution of right lungs; score: 7; volume: 12.298cm³; mean density value: −474.590hu; density classification: opacification; pneumothorax: presence; involved lung lobes: 3; lung coefficient: 10; PaCO₂: 22.6mmHg; PaO₂: 41mmHg; SaO₂: 79%;

ROC curve for lethality and severity

ROCs were plotted to evaluate the clinical value of distinguishing lethality and severity (Fig. 3). For the prediction of death, lung coefficient, CT score and volume performance were superior to other indicators (AUC = 0.967, CI: 0.928 ~ 1.000; 0.925, CI: 0.869 ~ 0.981 and 0.924, CI: 0.870 ~ 0.977, respectively) and contributed to equivalent discriminatory power. For severe/serious, AUCs of lung coefficient, CT score, and volume were 0.944, CI: 0.887 ~ 0.978; 0.915, CI:0.893 ~ 0.986and 0.909, CI: 0.883 ~ 0.978, respectively, which were higher than other indicators and also had equivalent discriminatory power. Detailed information can be found in Table 2. The Delong tests showed that in terms of lung coefficient vs. CT score, lung coefficient vs. volume, and CT score vs. volume, the comparisons between the survival/death group and the severity group did not have statistical significance (P = 0.402, 0.509, 0.286 and 0.835, 0.950, 0.286, respectively). Detailed information was provided in Supplementary Information (Figure S1/2 and Table S1/2).

Fig. 3.

Receiver operator characteristic(ROC) curves for predicting different outcomes

Table 2.

Receiver operating characteristic (ROC) curves of effective variables under two outcomes

Variable	Death vs. Survival					Variable	Severe/serious vs. Moderate/slight
	AUC	95%CI	Cut-off	SEN	SPE		AUC	95%CI	Cut-off	SEN	SPE
lc	0.967	0.99−1.00	17.01	96.67%	92.86%	lc	0.944	0.89–0.98	11.08	91.40%	85.00%
score	0.925	0.87–0.98	18	91.62%	87.80%	score	0.915	0.89–0.99	13	86.21%	84.21%
volume	0.924	0.87–0.98	543.54	89.20%	92.90%	volume	0.909	0.88–0.98	293.48	88.51%	82.71%
ill	0.893	0.84–0.95	5.5	83.30%	92.90%	ill	0.832	0.76–0.90	4.5	86.21%	60.90%
						mdv	0.854	0.79–0.92	-313.18	74.10%	77.50%
						PaCO2	0.605	0.52–0.72	30.50	74.47%	46.74%
						PaO2	0.732	0.66–0.82	49.50	70.21%	65.22%
						SaO2	0.739	0.66–0.83	86.50	68.09%	71.74%

OR odds ratio, CI confidence interval, lc lung coefficient, ill involved lung lobes, mdv mean density values

Correlation analysis between lung coefficient and CT score

The spearman correlation results are shown in Fig. 4. The lung coefficient was strongly corelated to CT score (r = 0.708, p = 0.000) in terms of survival, while there was no correlation regarding death (r = 0.171, p = 0.559). The lung coefficient and CT score were moderately correlated for both severe/serious and moderate/slight (r = 0.440, p = 0.002; r = 0.480, p = 0.000).

Fig. 4.

Correlations between lung coefficient and CT score at each condition

Logistic regression analysis and nomogram building

Univariable and multivariable analysis for logistics regression analysis suggested that two variables (volume and involved lung lobes) and five variables (region, orientation, distance, volume and score) were favorable predictive factors for death and severe/serious risk, respectively (Tables 3 and 4). Two nomograms to predict two outcomes were respectively formulated using the above predictive factors and the volume showed the largest contributions to prognosis (Fig. 5). The Hosmer Lemeshow test showed that the models fitted well (P > 0.05), as shown in Supplementary Information (Figure S3/4).

Table 3.

Univariate and multivariate logistic regression analysis for death

Variable	Univariate				Multivariate
	P	OR	B	95% CI	P	OR	B	95% CI
distance	0.000*	0.26	−1.34	0.13 ~ 0.54	0.593	0.74	−0.31	0.24 ~ 2.27
score	0.000*	1.41	0.34	1.24 ~ 1.60	0.439	0.90	−0.11	0.68 ~ 1.18
volume	0.000*	1.01	0.01	1.01 ~ 1.01	0.006*	0.01	1.01	1.01 ~ 1.01
damage location	0.990	0.00	−17.11	/
distribution	0.000*	0.07	−2.68	0.02 ~ 0.19	0.559	0.63	−0.46	0.13 ~ 2.97
pneumothorax	0.095	2.68	0.99	0.84 ~ 8.54
hemothorax	0.987	/	17.29	/
involved lung lobes	0.000*	4.68	1.54	2.48 ~ 8.83	0.036*	2.46	0.90	1.06 ~ 5.68

OR odds ratio, CI confidence interval, *P < 0.05;

Table 4.

Univariate and multivariate logistic regression analysis for severe/serious

Variable	Univariate				Multivariate
	P	OR	B	95% CI	P	OR	B	95% CI
region	0.045*	1.75	0.56	1.01 ~ 3.02	0.017*	2.96	1.09	1.22 ~ 7.23
orientation	0.000*	5.81	1.76	2.34 ~ 14.40	0.001*	9.53	2.25	2.41 ~ 37.66
distance	0.000*	0.29	−1.25	0.19 ~ 0.43	0.003*	0.40	−0.93	0.21 ~ 0.73
score	0.000*	1.39	0.33	1.28 ~ 1.52	0.000*	1.29	0.25	1.18 ~ 1.41
volume	0.000*	1.01	0.01	1.01 ~ 1.01	0.010*	1.01	0.01	1.01 ~ 1.01
mean density values	0.000*	1.01	0.01	1.01 ~ 1.02	0.057	2.69	0.99	0.97 ~ 7.46
damage location	0.982	0.00	−17.36	/
distribution	0.000*	0.13	−2.04	0.07 ~ 0.25	0.052	3.57	1.27	0.99 ~ 12.89
density classification	0.000*	0.27	−1.31	0.15 ~ 0.50	0.063	3.30	1.19	0.94 ~ 11.36
lacerations	0.032*	5.73	1.75	1.16 ~ 28.27	0.080	6.70	1.90	0.79 ~ 56.52
involved lung lobes	0.000*	4.68	1.54	2.48 ~ 8.83	0.655	0.90	−0.10	0.57 ~ 1.42

OR odds ratio, CI confidence interval, *P < 0.05;

Fig. 5.

Nomograms are used to predict the lethality and severity of PBLI

Discussion

The research on PBLI has garnered significant attention in recent years due to its high incidence and severe consequences in industrial and military settings. Battlefield data showed that up to 80%of battlefield injuries caused by explosions develop into lung injuries [14]. Harvey et al. found that explosions at close distance and enclosed spaces could increase the severity of PBLI [15, 16]. Yang et al. simulated low-pressure environments and found that reduced air pressure was associated with lung injury and mortality [17–19].

Various complex fields had different biological destructive effects and further increased the difficulty of clinical diagnosis and treatment, which was an urgent need for PBLI research. The clinical diagnosis of blast injury is based on the actual condition of the injury, clinical symptoms and radiology. Researches found that CT was the most recommended evaluation approach [20–22]. In this study, we established a large sample model with goats under a single destructive element, and set up three injury factors: altitude, distance and orientation to simulate the real explosive biological destructive effect. Based on the classification of lethality and severity, CT was used to evaluate the injury characteristics of different outcomes. After initial examination, 41 goats died on the spot, with a mortality rate of 18.64% (41/220), which was similar to Wolf’s discovery (range:17%−47%) [23]. The proportion of severe/serious case was 39.55% (87/220), with a death conversion rate of 45.98% (40/87). The results showed that close distance was the only injury factor associated with mortality, whereas CT metrics—including volume and the number of involved lung lobes—exhibited predictive value for outcomes. In terms of distinguishing the severity of injuries, high altitude, close distance and right-sided orientation toward the explosion all exacerbated the severity, while CT indicators such as score and volume emerged as key factors for differentiating the severity of lung injuries.

From a biological perspective, this study revealed the complex interactions between various injury factors and their impact on lung tissue damage. First, we observed that the degree of lung injury was more severe in plateau areas. Previous studies have confirmed that the hypobaric hypoxic environment of plateau regions can exacerbate the pathological process of blast-induced pulmonary injury through a dual mechanism. This is manifested as a significant decrease in the body’s compensatory capacity, leading to a marked reduction in the organ’s tolerance threshold to impact loads. Additionally, endothelial damage in the microcirculation and increased vascular permeability render capillaries more susceptible to rupture under the action of blast waves, thereby aggravating the bleeding of hollow organs [24, 25]. Moreover, the pressure difference effect between the plateau environment and the blast wave is more pronounced than in plain areas, which further amplifies biomechanical injury effect [26]. This synergistic action of pathophysiological and mechanical factors results in significantly higher severity of pulmonary injury in plateau areas compared to plain areas, necessitating the need of more effective prevention and treatment measures.

Our finding that right-sided orientation toward the explosion center increased the risk of severe/serious injury is novel. It is speculated that the mediastinum, which contains structures such as the heart, is positioned closer to the right thoracic wall. After the blast wave reaches the mediastinal area, the attenuation effect of dense tissues on the blast wave is limited, resulting in the repeated energy transmission within the right lung and thus exacerbating lung injury. Additionally, from the perspective of respiratory tract anatomy, the right main bronchus is shorter, thicker, and has a steeper branching angle compared to the left main bronchus. These anatomical characteristics may facilitate direct impact of the blast wave on deep right lung tissues, causing more extensive damage. While anatomical considerations suggest that right chest orientation toward the explosion center could enhance blast wave-induced lung damage, further in-depth mechanical studies are required to validate this. These insights help us understand the biological mechanisms underlying PBLI and can inform the development of targeted therapeutic and protective strategies.

PBLI is defined as “radiological and clinical evidence of acute lung injury occurring within 12 h of exposure and not due to secondary or tertiary injury” [27]. The definition of primary blast injury is as following: blast overpressure contacts the victim leading to direct tissue damage, and thoracic imaging findings (CT) are peribronchovascular ground-glass and consolidative opacities [28]. This attenuation around bronchovascular tissue is typical pulmonary contusion, which is common in the moderate/slight group in our study. While in the severe/serious group, it is more manifested as diffuse lung infiltration in single or both lungs, which can be symmetrical butterfly distribution, similar to the description reported by Hare [29]. The current clinical triage urgently requires accurate prediction of PBLI prognosis. Our study found that lung coefficient, CT score and injury volume all shared the optimal predictive performance without difference, which can be explained by the fact that the ability of the three indicators to distinguish risks is equivalent. Considering from a clinical operational perspective, CT scoring is faster and more convenient to use. This scoring method was previously used to differentiate the severity of COVID-19. For the first time, we used it to evaluate the prognosis of PBLI. CT scores were significantly elevated in death and severe/serious group. When the scoring thresholds were 18 and 13, death and severe/serious risk could be identified with sensitivity of 91.62% and 86.21%, specificity of 87.80% and 84.21%, respectively. The lung coefficient can measure the degree of pulmonary edema and reflect the severity of lung injury [12]. In different degrees of severity, it showed a moderate correlation with CT scores, which proves the reliability of CT scores in distinguishing the risk of severity. However, the response to death is not sufficient to be adopted. It is hypothesized that the small sample size of the death group (n = 41) increased the risk of false negatives, potentially masking actual weak-to-moderate correlations between indicators. This could lead to insufficient statistical power, thereby failing to accurately detect the true association between the lung coefficient and the CT score. Moreover, in the death group, the lung injuries often progressed to irreversible stages (such as extensive pulmonary consolidation and hemorrhagic necrosis). At this point, the biological correlation between the lung coefficient and CT features may be diluted by the complexity of the pathological process. Such heterogeneity further weakens the correlation within the group.

Upon reviewing the basic characteristics of the data, we found that there were 10 cases of pneumothorax in the death group, which may also have a weak correlation with pulmonary hemorrhage and edema. Despite the non-significant correlation observed in the death group, the CT score can rapidly and non-invasively quantify the severity of injury in the early stage, which is unparalleled by the lung coefficient. Multivariate regression analysis showed that the CT score had an AUC of 0.925 in predicting the risk of death, with no difference in classification ability compared to the lung coefficient. This indicates its key role in early triage and its irreplaceable advantage in real-time clinical decision-making.

Evaluating the severity of PBLI before treatment is helpful for clinical decision-making and predicting the efficacy of treatment. At present, it can be confirmed that image methods are recommended measures for the evaluation of PBLI, but this task is challenging as early diagnostic signs are scarce and lack specificity [15, 30]. The current imaging mostly describes the characteristics of the injury, so radiologists still lack confidence in distinguishing the severity of PBLI and predicting prognosis [21, 22, 31]. In the modern medical decision-making process, nomograms are widely used in tumor staging, and there is evidence to suggest that their predictive efficacy is more accurate than the multiple tumor staging system of the American Joint Committee on Cancer (AJCC) [32]. By combining multiple parameters to calculate the probability of certain events occurring, the nomogram shows a wide application prospect. In our results, volume was the chief contributing variable and was able to generate personalized prognostic reports when combined with other variables. Surprisingly, all morphological indicators were not included, this also illustrates from another perspective that imaging features alone cannot predict death and severe/serious risk. The predictive nomograms developed in this study can be used to stratify cases according to the risk of severe/serious and death, enabling clinicians to allocate resources more effectively and initiate timely interventions. Additionally, the use of CT scoring as a non-invasive assessment tool can facilitate early diagnosis and monitoring of PBLI, thereby optimizing clinical decision-making. The nomogram developed in this study integrates CT imaging indicators, scores and explosion environmental variables to achieve precise and rapid battlefield triage. Its core advantages are as follows: (1) It requires no invasive procedures, enabling adaptation to harsh frontline conditions; (2) It quantifies risk probabilities, reducing subjective bias; (3) It incorporates key variables such as explosion distance and direction, better meeting the demands of actual combat senarios. In the future, it can be further promoted to field hospitals and disaster relief sites in combination with portable CT devices.

There are some limitations to this study. First, given that the diaphragm of goats is flatter and has a smaller range of motion compared to that of humans, it may underestimate the amplifying effect of intrathoracic negative pressure on pulmonary injury. Subsequent primate or human trials are required to confirm the generalizability. Second, this study did not employ contrast-enhanced CT, which may have led to the omission of early injury biomarkers such as pulmonary vascular embolism or micro-hemorrhage. Future research should combine contrast - enhanced CT or angiography to verify the accuracy of the model. Finally, the standard segmentation method has not been determined.

In conclusion, the controllable animal model established in this study successfully simulated the real explosion effect of PBLI. CT was able to visualize the characteristic elements of lung injury. The research results show that integrating imaging data and injury factors is necessary for assessing the prognosis of PBLI, which is of great significance for optimizing clinical decision-making.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

SNW conceived, designed and supervised the study. BY performed data analysis and drafted the manuscript. HWW, LF and YSY conducted a major clinical experiment. HWW and BY contributed to imaging and clinical data collection. KLX supervised the literature review and data quality control. KLX and SNW revised the manuscript. All the authors have read and approved the final manuscript.

Funding

This work was supported by Project of Science and Technology Plan of Chongqing Clinical research Centre of Imaging and Nuclear Medicine (CSTC2015YFPT-gcjsyjzx0175); Research Project on Education Reform of Army Medical University (2022A14); Chongqing Medical Scientific Research Project (No. 2023MSXM009).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the animal ethics committee of Army Medical University(AMUWEC20202140).

Competing interests

The authors declare no competing interests.