Traumatic rib fracture patterns associated with bone mineral density statuses derived from CT images

Background The impact of decreased bone mineral density (BMD) on traumatic rib fractures remains unknown. We combined computed tomography (CT) and artificial intelligence (AI) to measure BMD and explore its impact on traumatic rib fractures and their patterns. Methods The retrospective cohort comprised patients who visited our hospital from 2017–2018; the prospective cohort (control group) was consecutively recruited from the same hospital from February–June 2023. All patients had blunt chest trauma and underwent CT. Volumetric BMD of L1 vertebra was measured by using an AI software. Analyses were done by using BMD categorized as osteoporosis (<80 mg/cm3), osteopenia (80–120 mg/cm3), or normal (>120 mg/cm3). Pearson’s χ2, Fisher’s exact, or Kruskal–Wallis tests and Bonferroni correction were used for comparisons. Negative binomial, and logistic regression analyses were used to assess the associations and impacts of BMD status. Sensitivity analyses were also performed. Findings The retrospective cohort included 2,076 eligible patients, of whom 954 (46%) had normal BMD, 806 (38.8%) had osteopenia, and 316 (15.2%) had osteoporosis. After sex- and age-adjustment, osteoporosis was significantly associated with higher rib fracture rates, and a higher likelihood of fractures in ribs 4–7. Furthermore, both the osteopenia and osteoporosis groups demonstrated a significantly higher number of fractured ribs and fracture sites on ribs, with a higher likelihood of fractures in ribs 1–3, as well as flail chest. The prospective cohort included 205 eligible patients, of whom 92 (44.9%) had normal BMD, 74 (36.1%) had osteopenia, and 39 (19.0%) had osteoporosis. The findings observed within this cohort were in concurrence with those in the retrospective cohort. Interpretation Traumatic rib fractures are associated with decreased BMD. CT-AI can help to identify individuals who have decreased BMD and a greater rib fracture rate, along with their fracture patterns.


Introduction
Blunt chest trauma is a significant global health concern, comprising 15% of emergency department admissions and ranking as the third leading cause of trauma-related deaths, with substantial morbidity and mortality rates (1,2).It is commonly caused by incidents such as traffic accidents, violent acts, and falls, which are not uncommon in daily life and work, implying that any individual engaged in social activities is at risk of experiencing blunt chest trauma (3,4).Among these causes, traffic accidents are the most prevalent reason for blunt chest trauma (3).In China, the number of motor vehicles has grown by 250% in the past decade, reaching 319 million by the end of 2022, making it the world's largest automobile market.This explosive growth and the substantial overall volume have contributed to a consistently high number of blunt chest trauma patients (5).
Rib fractures are the most common associated injuries among these patients and serve as a crucial indicator of injury severity (6).The severity of trauma is influenced by different types of rib fractures.The number of rib fractures is a known risk factor for rib fracture-associated mortality, incidence, intensive care unit admission rate, hospitalization duration, and lung injury (7,8).In some countries, the number of rib fractures is also a forensic assessment indicator for accident penalties and sentencing (9).Specific rib fractures correlate with different organ injuries; ribs 1-3 are associated with a higher risk of major vascular and brachial plexus injuries, ribs 4-7 with an increased risk of lung contusion and cardiovascular injuries, and ribs 8-12 with a higher risk of abdominal solid organ injuries (10).Flail chest, the most severe pattern of rib fractures, necessitates an immediate assessment of the patient's condition and appropriate measures (11).Therefore, different types of rib fractures have a significant impact on the management and prognosis of patients with thoracic trauma.
Bone mineral density (BMD) is a crucial factor influencing the risk of fracture (12)(13)(14)(15).In addition to increasing the susceptibility to fragility fractures, decreased BMD also raise the likelihood of high-energy traumatic fractures (16,17).With the global aging trend intensifying, the population with decreased BMD is expected to increase, thereby drawing increasing attention to its adverse consequences.However, there is a relative dearth of research regarding how BMD impacts the risk and patterns of rib fractures in patients with blunt chest trauma.
Early identification of osteoporosis/osteopenia is crucial in preventing fractures and reducing associated medical costs (18).Recently, there has been growing interest in the concept of opportunistic BMD screening, which involves extracting BMD information from CT scans performed initially for non-BMD related clinical purposes (19,20).Additionally, the integration of artificial intelligence (AI) has facilitated the automation of this process, eliminating the laborious manual operations (21,22).This method, requiring no additional time costs or radiation exposure, provides additional information through a single scan, aiding in the identification of individuals at higher risk of future fractures and their potential patterns of rib fractures.This may contribute to raising awareness in patients, prompting thoughtful consideration of improving BMD and preventing rib fractures.Therefore, this study aims to evaluate the impact of BMD on the risk and patterns of rib fractures following blunt chest trauma through opportunistic BMD screening based on CT-AI.

Study population
This study was approved by the Ethics Committee of Huadong hospital (approval number: 20220069).The requirement to obtain written informed consent was waived.The study dataset of CT images was identified by searching the hospital's Picture Archiving and Communication System using the keyword "chest trauma evaluation".The inclusion criteria included: (1) individuals who had experienced chest collision events, such as car accidents, violent acts, and falls, excluding puncture events like gunshot wounds and stabbings; (2) patients who underwent CT scans for a definitive diagnosis; (3) a time interval not exceeding a year between the event and the CT examination; (4) age >18 years old; (5) no prior diagnosis of osteoporosis or receipt of osteoporosis treatment.Patients meeting any of the following criteria were excluded: (1) presence of skeletal diseases associated with an increased risk of fractures (such as osteogenesis imperfecta, or malignant tumors); (2) CT scan not including the L1 vertebra; (3) history of rib fractures; and (4) absence of reconstructed images with a thickness of ≤ 1.5 mm. Figure 1 shows a flowchart of the study design.
The eligible patients from January 1, 2017, to December 31, 2018, formed the retrospective cohort.To validate our findings and assess the predictive value of the BMD-based model, our researchers prospectively collected data continuously from eligible patients undergoing CT scans due to chest impact events at the radiology department (single center) from February 13 to June 2, 2023, forming a prospective cohort.Following their CT examinations, their thin-slice CT images were immediately processed by an AI algorithm to obtain BMD measurements and predict rib fracture patterns.Subsequently, the rib fracture patterns predicted by the BMD-based model will be compared with diagnostic reports provided by radiologist after a standard waiting period to validate the predictive value of the model.Detailed scan parameters are described in the Supplementary Methods of the online Supplementary Material.

AI-based BMD measurements
A two-stage, deep-learning model was adopted.First, a threedimensional UNet model was employed to complete vertebral segmentation, segmenting trabecular bone while excluding Abbreviations: AI, artificial intelligence; AUC, area under the receiver operating characteristic curve; BMD, bone mineral density; CI, confidence interval; CT, computed tomography; CT-AI, computed tomography and artificial intelligence combined; DXA, dual-energy X-ray absorptiometry; OR, odds ratio; vBMD, volumetric BMD.The three-dimensional UNet model used in the first stage consisted of an encoder and a decoder network, with featurefusion achieved through skip connections between the encoding and decoding modules.In the encoder network, two 3 × 3 × 3 convolutional layers and one 2 × 2 × 2 max pooling layer (stride = 2) were used for feature extraction, followed by downsampling.Four downsampling operations were performed to extract features of different dimensions.In the decoder network, a 2 × 2 × 2 upsampling convolutional layer and two 3 × 3 × 3 convolutional layers were used for feature extraction, followed by four upsampling operations.The final network layer adjusted the number of channels to the desired number of classes using a 1 × 1 convolutional layer.
In the second stage, the DenseNet model for BMD detection was divided into two parts: feature extraction and fully connected regression detection, with BMD measured using quantitative CT as the training target.The feature extraction part consisted of three Dienes blocks, each of which comprised an equal number of 1 × 1 × 1 and 3 × 3 × 3 convolutional layers.The fully connected regression detection head contained two layers of 2 × 2 × 2 global average pooling.The input size of the model was [16,32,32].
We restricted our analysis to the L1 vertebra, previously considered as the optimal target for opportunistic BMD measurements using CT images (25)(26)(27).BMD status was categorized as normal (> 120 mg/cm 3 ), osteopenic (80-120 mg/ cm 3 ), or osteoporotic (≤ 80 mg/cm 3 ) based on the recommendation of the International Society for Clinical Densitometry and the American College of Radiology (28, 29).This criterion was considered applicable in the Chinese population (30).The AI interface of BMD measurement is displayed in Figure 2.

Outcome measures
Our primary outcome was rib fracture patterns, confirmed by a radiologist (engaged in musculoskeletal imaging diagnosis for 15 Workflow of this study.CT, computed tomography; BMD, bone mineral density.years), through reference to both the CT diagnostic report and CT images, incorporating the following: 1.The presence of at least one rib fracture.2. Characteristics: The number of fractured ribs and rib fracture sites; the location of the fractured rib(s) categorized as ribs 1-3, 4-7, and 8-12; and the presence of flail chest.3. Classification based on fracture displacement: The degree of displacement of a single-rib fracture was classified as undisplaced, offset, or displaced (31).Patients were categorized into three groups: Group 1, undisplaced fractures only; Group 2, any offset but no displaced fractures; and Group 3, any case with displaced fractures (32).Patients who underwent internal fixation were excluded (as the fracture type present may change after internal fixation).

Statistical analysis
Data were analyzed using R software version 4.3.0(R Foundation for Statistical Computing, Vienna, Austria).Categorical data are given as frequencies and percentages.Medians and first to third quartiles (interquartile range) are used to summarize continuous variables.We compared variables between groups using Pearson's c 2 or Fisher's exact tests and Kruskal-Wallis tests for categorical and continuous variables, respectively.Furthermore, Bonferroni correction was used for post-hoc multiple comparisons.The association between BMD status and categorical outcomes was also assessed using binary logistic regression analysis.The impact of BMD status on the numbers of fractured ribs and rib fracture sites was assessed using negative binomial regression analysis.Unadjusted and adjusted (for sex and age [per 10 years]) logistic regression analyses were performed to calculate the odds ratios (ORs) and corresponding confidence intervals (CIs) to determine the association between BMD status and outcomes.Sensitivity analyses were performed using a three-knot piecewise polynomial model for age, with separate analyses for men and women, respectively.Statistical significance was set at P < 0.05.
We conducted further investigations into the ability of BMD alone and BMD combined with age and sex to predict rib fracture patterns using five-fold cross-validation.Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity.Owing to the potential overestimation of performance when quantifying the predictive value of the model using derived data, we utilized a prospective cohort to calculate the performance metrics, enabling us to evaluate the performance after optimistic correction.The detailed AI interface of BMD measurement.The left column of this interface comprises, from top to bottom, the patient's coronal, axial, and three-dimensional reconstructed CT images, respectively.The middle column displays the patient's sagittal CT images, while the right column presents the corresponding BMD results of the vertebrae.Vertebrae with BMD results <80 mg/m3 are annotated in red on the CT images, along with their respective BMD values.Vertebrae with BMD results in the range of 80-120 mg/m3 are marked in yellow on the CT images, together with their corresponding BMD values.BMD, bone mineral density; CT, computed tomography; AI, artificial intelligence.
Moreover, BMD status was associated with the numbers of fractured ribs and rib fracture sites.As shown in Figure 3, the median numbers (IQR) of fractured ribs in the groups with normal BMD, osteopenia, and osteoporosis were 4 (IQR: 3-6), 5 (IQR: 3-7), and 6 (IQR:4-8), respectively.After Bonferroni correction, both the osteopenia and osteoporosis groups exhibited significantly higher numbers of fractured ribs compared with that in the normal BMD group (both P<0.001).The unadjusted negative binomial regression analysis results were OR: 1.2, 95% CI: 1.1-1.3 and OR: 1.4, 95% CI: 1.3-1.5 for the osteopenia and osteoporosis groups, respectively.
Association of BMD status with flail chest BMD status was significantly associated with flail chest.Among a total of 954 patients with normal BMD, 806 patients with osteopenia, and 316 patients with osteoporosis, there were 99 (10.4%),154(19.1%),and 82 (25.9%) with flail chest, respectively.In the Bonferroni post-hoc analysis, both the osteopenia and osteoporosis groups had higher proportions of patients with flail chest than that in the normal BMD group (both P<0.001).The unadjusted and sex-and age-adjusted analyses results revealed an association between osteopenia and flail chest (OR: 2.0, 95% CI: 1.6-2.6 and OR: 1.6, 95% CI: 1.
Supplementary Figure 2 illustrates the fracture frequency per rib in the normal and decreased BMD (osteopenia and osteoporosis) groups.
The outcomes according to BMD group are shown in Figure 4 (Supplementary Table 1).Sensitivity analyses, including a three-knot piecewise polynomial model for age, separate analyses for men and women, and separate analyses for patients over 55 years old (Supplementary Tables 2-4), did not substantially change our results.

Discussion
We demonstrated that patients with chest trauma and decreased BMD had higher rib fracture rates and numbers of fracture sites, with a higher likelihood of fractures in ribs 1-3 and 4-7 as well as increased susceptibility to flail chest.Furthermore, we created a model encompassing BMD, sex, and age that moderately predicted rib fracture patterns.Unadjusted and adjusted odds ratios (95% confidence intervals) of events for the osteopenia and osteoporosis groups compared with that for the normal bone mineral density group.OR, odds ratio.Our results represent that rib fractures, even with trauma, are associated with decreased BMD.However, decreased BMD presents as a silent condition, with patients generally not undergoing BMD screening prior to fractures, leading to fracture treatment constituting a significant portion of expenses associated with osteoporosis (33).Opportunistic BMD assessment using CT-AI, conducted through CT imaging for other clinical purposes, may aid in identifying patients at an elevated risk of fractures and provide additional information regarding rib fracture patterns, intending to raise awareness in patients and enabling consideration of preventive and treatment.
Previous studies have extensively explored the association between BMD and fractures based on CT images (34)(35)(36)(37)(38). Chalhoub et al. (35) found that a greater risk of various fracture types was associated with low BMD.Other studies have investigated the predictive value of BMD for fractures using CT images (15,39,40); Gruenewald et al. (15) utilized the vBMD measurements of L1 to predict osteoporosis-related fractures and achieved satisfactory performance (AUC: 0.937, 95% CI: 0.867-0.977).Nevertheless, most previous studies have focused on traditional osteoporotic fractures (hip or spine), with limited exploration of the role of BMD in trauma.Prins et al. (41) investigated the association between BMD and rib fractures using DXA and showed that BMD impacted rib fractures.However, additional DXA scans increased the scan time and radiation exposure.Furthermore, this small-sample study (n = 119) did not establish an association between BMD and rib fracture patterns.
This study had some limitations First, it was conducted at a single center, the patient data were derived from individuals who underwent chest trauma evaluation at our institution.These patients had all experienced chest impact events and underwent CT scans for the purpose of a definitive diagnosis, often for forensic clinical assessment.This particular feature may have resulted in a limited number of patients without fractures, potentially rendering the sample insufficient to comprehensively represent the entire patient population.Second, our inability to obtain the actual accident circumstances made it impossible to differentiate between high-and low-energy fractures in trauma cases.Consequently, the correlation between BMD and low-energy fractures may have been underestimated due to the inclusion of high-energy fractures.However, non-osteoporotic fractures resulting from high-energy trauma are associated with low BMD    (17).Future studies should include detailed accident information to provide improved correlation results.Finally, the model performance in predicting patterns by combining BMD status, age, and sex was moderate, indicating that factors other than these variables were influential.Future large-scale prospective studies should explore other risk factors and incorporate vBMD derived from CT-AI to assess fracture risk and patterns accurately.
Our study provides evidence for an association between BMD status and traumatic rib fractures.We further highlighted the potential predictive value of BMD status regarding the risk of rib fractures.Opportunistic BMD screening based on CT-AI can be utilized for the identification of individuals with decreased BMD, along with their fracture patterns.

FIGURE 3
FIGURE 3Distribution of number of fractured ribs and locations of rib fractures among bone mineral density groups.BMD, bone mineral density.

FIGURE 4
FIGURE 4 FIGURE 5Area under the receiver operating characteristic curve results for the retrospective (A-C) and prospective (D-F) cohorts.The AUCs from left to right represent the predictions of flail chest, fractures in ribs 1-3, and fractures in ribs 4-7, respectively.AUC, area under the receiver operating characteristic curve; BMD, bone mass density.

TABLE 1
Demographics and rib fracture features in the three BMD groups in the retrospective cohort.

TABLE 2
Comparisons of bone mineral density status and demographic characteristics between retrospective and prospective cohorts.

TABLE 3
Demographics and rib fracture features of the three bone mineral density groups in the prospective cohort.

TABLE 4
Receiver operating characteristic analysis results in the retrospective and prospective cohorts a .
AUC, area under the receiver operating characteristic curve; SENS, sensitivity; SPEC, specificity; BMD, bone mineral density; R-C, retrospective cohort; P-C, prospective cohort; Ref, reference.a 95% confidence intervals were obtained by setting the seed number to 123 and performing 1,000 bootstrap iterations.