PurposeTo investigate the predictive value of different vertebral specificity of BMD for cage subsidence among patients undergoing ACDF. MethodsBMD at different vertebrae was measured by CT image and MRI image. Cage subsidence was defined as >= 3 mm loss of fusion segmental height. Logistic regression analysis was used to identify BMD at specific vertebrae associated with cage subsidence. Receiver operating characteristic curve analysis was used to evaluate the value of BMD of different vertebrae in predicting cage subsidence. Results117 patients (182 intervertebral spaces) were included, with a mean age of 54.6 +/- 10.9years. The mean follow-up was 12.5 +/- 3.8months. Of the 182 intervertebral spaces, subsidence was found in 85 (46.7%) intervertebral spaces. The mean Hounsfield unit (HU) values measured by CT in the subsidence group were significantly lower than those in the non-subsidence group (upper vertebral body: 285.8 vs. 361.8, p < 0.001; lower vertebral body: 262.1 vs. 324.0, p < 0.001; average: 273.9 vs. 342.9, p < 0.001). Similarly, the mean bone quality scores measured by MRI in the subsidence group were higher than those in the non-subsidence group (upper vertebral body: 2.69 vs. 2.19, p < 0.001; lower vertebral body: 2.58 vs. 2.18, p < 0.001; average: 2.63 vs. 2.19, p < 0.001). Logistic regression analysis showed that the HU values were significantly negatively correlated with subsidence (p < 0.001), while the bone quality scores were significantly positively correlated with subsidence (p < 0.001). The results show that the Pre-T1 slope, Post-C2-C7 Cobb angle, and Post-T1 slope in the subsidence group were significantly larger than those in the non-subsidence group, with statistically significant differences (Pre-T1 slope: P = 0.015; Post-C2-C7: P < 0.001; Post-T1 slope: P = 0.006). The results from the multivariate logistic regression analysis indicated that bone density indicators remained statistically significant in predicting fusion device subsidence, while the Post-C2-C7 Cobb angle from the sagittal plane parameters also demonstrated statistical significance.The areas under the curve (AUCs) of BMD in the upper vertebrae were greater than those in the lower vertebrae, whether measured by CT (upper vertebral body: 0.754, lower vertebral body: 0.723; upper endplate: 0.766, lower endplate: 0.712) or MRI (upper vertebral body: 0.755, lower vertebral body: 0.717; upper endplate: 0.702, lower endplate: 0.646). These results indicate that the BMD of the upper vertebrae may be a better predictor of cage subsidence than that of the lower vertebrae. ConclusionThe bone density (HU value and VBQ score) at the surgical site is a strong predictor of cage subsidence in patients undergoing ACDF. For BMD assessment, we recommend that site-specific measurements, particularly those of the upper vertebrae, be given greater consideration rather than relying solely on generalized or averaged values.