[Agreement between Cervi-AI and expert measurements of cervical MRI parameters and association of AI-derived imaging metrics with surgical approach selection in multilevel degenerative cervical myelopathy].
Authors
Affiliations (2)
Affiliations (2)
- Department of Orthopedics, Spine Center, Shanghai Changzheng Hospital, Naval Medical University, Shanghai 200003, China.
- Department of Orthopedics, No.905 Hospital of the Navy of the Chinese People's Liberation Army, Naval Medical University, Shanghai 200052, China.
Abstract
<b>Objective:</b> To evaluate the agreement between cervical magnetic resonance imaging (MRI) parameters measured by the Cervi-AI model and expert assessments, and to investigate the association between these parameters and the selection of anterior or posterior surgical approaches in patients with multilevel degenerative cervical myelopathy (DCM). <b>Methods:</b> This retrospective study included patients with DCM who underwent cervical spine surgery at the Second Affiliated Hospital of Naval Medical University between June 2023 and October 2024. Patients were categorized into anterior and posterior surgical approach groups according to the actual surgical procedure performed. Intraclass correlation coefficient (ICC) and Bland-Altman analysis were used to compare the consistency of cervical spinal stenosis (CSS) grade, involved segments, and maximum spinal cord compression (MSCC) index assessed by Cervi-AI with the expert-measured reference standard. Multivariable logistic regression analysis was performed to identify factors associated with surgical approach selection. Receiver operating characteristic (ROC) curves were generated, and the area under the curve (AUC) was calculated to evaluate the predictive performance of clinical and imaging parameters for surgical approach selection. <b>Results:</b> A total of 146 patients were included (77 males and 69 females), with a mean age of (52.0±12.0) years. Eighty-six patients underwent anterior surgery, and 60 underwent posterior surgery. No significant differences were observed between the two groups in terms of symptom type or symptom duration (all P values>0.05). Cervi-AI measurements showed excellent agreement with expert assessments for the number of involved CSS levels, maximum stenosis grade, and MSCC index, with ICCs of 0.970, 0.958, and 0.984, respectively. Multivariable logistic regression analysis revealed that longer symptom duration (per 12-month increase: <i>OR</i>=0.752, 95%<i>CI</i>: 0.576-0.957, <i>P</i>=0.020), a greater number of involved CSS levels (per additional level: <i>OR=</i>0.058, 95%<i>CI</i>: 0.014-0.162, <i>P</i><0.001), and the presence of ossification of the posterior longitudinal ligament (OPLL) (<i>OR</i>=0.103, 95%<i>CI</i>: 0.027-0.328, <i>P</i><0.001) were independently associated with selecting an posterior surgical approach. ROC curve analysis demonstrated that the number of involved CSS levels had the highest predictive performance for surgical approach selection (AUC=0.903, 95%<i>CI</i>: 0.852-0.955, <i>P</i><0.001). <b>Conclusions:</b> The Cervi-AI model can provide highly consistent and standardized measurements of key cervical MRI parameters compared with expert assessments. Symptom duration, the extent of cervical spinal stenosis, and the presence of OPLL are independently associated with the selection of anterior versus posterior surgical approaches in patients with multilevel DCM.