A Novel Radiomics-Based Bone Quality Assessment for Early Prediction of Proximal Junctional Kyphosis in Patients with...

A retrospective analysis.

To explore the potential of radiomics as a novel bone assessment tool for early prediction of proximal junctional kyphosis (PJK) in adult spinal deformity (ASD).

PJK is a prevalent complication following ASD surgery. Since impaired bone quality is a major risk factor, accurate preoperative bone assessment is crucial for early identification of high-risk patients. However, conventional metrics such as T-score, Hounsfield unit (HU) value, and vertebral bone quality (VBQ) score exhibit limitations in accuracy and reliability.

A total of 358 ASD patients were analyzed and randomly assigned to training and test sets (7:3). Radiomic features were extracted from lumbar CT and MRI scans to construct CT and MRI radiomics scores (CTRS/MRIRS). In parallel, T-score, HU value, and VBQ score were also evaluated. Univariable prediction models were first developed for each of the five bone metrics. Subsequently, multimodal models were constructed using the best-performing bone metric as the core variable, with additional selected clinical and radiographic parameters incorporated to further enhance predictive performance. Model performance was evaluated using AUROC, net reclassification improvement (NRI), and integrated discrimination improvement (IDI).

Among univariable models, CTRS (AUROC=0.780) and T-score (AUROC=0.793) exhibited superior predictive performance compared to MRIRS (AUROC=0.694), HU value (AUROC=0.713), and VBQ score (AUROC=0.658). Multimodal models significantly outperformed their univariable counterparts (CTRS/T-score multimodal AUROC=0.880/0.885), with improved reclassification ability (CTRS univariable vs. multimodal: IDI=-0.2; T-score univariable vs. multimodal: NRI=-0.332, IDI=-0.247; all P <0.001).

The CT-based radiomics score presents a promising alternative to conventional bone quality metrics for early prediction of PJK after ASD surgery. Integrating CTRS with clinical and radiographic factors further enhances predictive accuracy, providing a valuable framework for preoperative risk stratification.

III.