Background Low back pain is the most common spinal disorder among soldiers, and load carriage training (LCT) is considered the main cause. We aimed to investigate changes in the spine system of soldiers after LCT at high altitudes and the change trend of the lumbar spine and surrounding soft tissues under different load conditions. Methods Magnetic resonance imaging scans of the lumbar spines of nine soldiers from plateau troops were collected and processed. We used ImageJ and Surgimap software to analyze changes in the lumbar paraspinal muscles, intervertebral discs (IVDs), intervertebral foramina, and curvature. Furthermore, the multiple linear regression equation for spine injury owing to LCT at high altitudes was established as the mathematical prediction model using SPSS Statistics version 23.0 software. Results In the paraspinal muscles, the cross-sectional area (CSA) increased significantly from 9126.4 ± 691.6 mm2 to 9862.7 ± 456.4 mm2, and the functional CSA (FCSA) increased significantly from 8089.6 ± 707.7 mm2 to 8747.9 ± 426.2 mm2 after LCT (P < 0.05); however, the FCSA/CSA was not significantly different. Regarding IVD, the total lumbar spine showed a decreasing trend after LCT with a significant difference (P < 0.05). Regarding the lumbar intervertebral foramen, the percentage of the effective intervertebral foraminal area of L3/4 significantly decreased from 91.6 ± 2.0 to 88.1% ± 2.9% (P < 0.05). For curvature, the lumbosacral angle after LCT (32.4° ± 6.8°) was significantly higher (P < 0.05) than that before LCT (26.6° ± 5.3°), while the lumbar lordosis angle increased significantly from 24.0° ± 7.1° to 30.6° ± 7.4° (P < 0.05). The linear regression equation of the change rate, △FCSA% = − 0.718 + 23.085 × load weight, was successfully established as a prediction model of spinal injury after LCT at high altitudes. Conclusion The spinal system encountered increased muscle volume, muscle congestion, tissue edema, IVD compression, decreased effective intervertebral foramen area, and increased lumbar curvature after LCT, which revealed important pathophysiological mechanisms of lumbar spinal disorders in soldiers following short-term and high-load weight training. The injury prediction model of the spinal system confirmed that a load weight < 60% of soldiers’ weight cannot cause acute pathological injury after short-term LCT, providing a reference supporting the formulation of the load weight standard for LCT.