Construction and validation of a deep learning prediction model for cervical instability

• A deep learning model using cervical MRI images achieved high predictive performance for early cervical instability, with AUC of 0.97 in both training and test sets. • The model incorporated manual annotations of five key anatomical structures, enhancing interpretability and clinical relevance. • The study included 308 young and middle-aged participants, with a balanced distribution of cervical instability patients and healthy controls. • The deep learning approach outperforms traditional radiomics by eliminating manual feature engineering and enabling end-to-end learning.