OBJECTIVE

Bone mineral density (BMD) and muscle health influence an occupant’s tolerance to motor vehicle crash (MVC), and these musculoskeletal factors can be used to design more effective countermeasures. CT imaging routinely acquired during trauma evaluation can be leveraged to obtain volumetric (v)BMD and muscle metrics, yet extracting measurements manually from CT is time-consuming and inconsistent, typically limiting it to single anatomical regions.

In this retrospective observational study, we applied an existing AI-based CT segmentation tool to characterize the musculoskeletal status of MVC occupants.

METHODS

An external cohort of 55 adults with CT scans including a bone calibration phantom was used to validate the automated segmentation of trunk muscle cross-sectional area (CSA) and an automated tissue-calibrated method for vBMD. Agreement with manual CSA and phantom-calibrated vBMD was assessed via correlation and Bland-Altman analyses.

The validated automated pipeline was then applied to 851 CT scans from Crash Injury Research and Engineering Network (CIREN) cases collected between 2005-2023. The Data Analysis and Facilitation Suite (v3.11.3, Voronoi Health Analytics) software was used to measure lumbar spine (L1-L4), pelvis, femur head, femur neck and femur trochanter + shaft vBMD, as well as trunk muscle CSA and density.

Sarcopenia was defined as a trunk muscle CSA divided by height squared <38.5 cm 2/m 2 for females and <52.4 cm 2/m 2 for males. Osteopenia was defined as lumbar spine vBMD <145 mg/cm 3.

These characteristics were examined in association to regional fracture count with abbreviated injury scale (AIS) severity and injury severity score (ISS) using negative binominal regression, including crash variables (delta-V, belt status, driver status, airbag deployment, principal direction of force, model year, curb weight) and occupant characteristics (age, sex, height, weight).

RESULTS

In the validation cohort (64% female; ages 66 ± 4), automated and manual trunk muscle CSA measurements agreed (r = 0.996; p < 0.0001), with small average differences (-4.9 cm 2). Automated tissue-calibrated lumbar vBMD closely matched phantom-calibrated values (r = 0.938, mean difference -0.21 mg/cm³), and femur regions showed similar agreement (correlation = 0.865-0.927; all p < 0.0001).In CIREN occupants (56% female; ages 47 ± 20), 31% had sarcopenia, 28% had osteopenia, and 14% osteosarcopenia according to the CT-based definitions.

vBMD declined with age across all regions (all p < 0.0001). Lower vBMD was associated with higher ISS in the pelvis (p = 0.003), femur head (p < 0.0001), femur neck (p < 0.0001), and trochanter + shaft (p = 0.00087).

Additionally, a 10 mg/cm³ higher femur head vBMD was associated with 1.6% fewer AIS 2 fractures (p-value = 0.005).

CONCLUSIONS

Application of an AI-based segmentation platform to MVC CT scans demonstrated that compromised musculoskeletal tissue quality is common and associated with fractures and injury severity beyond traditional occupant and crash characteristics. This suggests that musculoskeletal profiling can inform MVC injury prevention, occupant protection design, and post-crash care strategies.