Toward a clinically integrated AI framework for personalized spine care

Background and Objective: Low back pain (LBP) is the leading global cause of disability, arising from complex structural, clinical, behavioral, and psychosocial factors. Although artificial intelligence (AI) has demonstrated considerable promise in spine imaging, automated diagnosis, and outcome prediction, most existing approaches remain task-specific, with limited integration of multimodal clinical information into routine spine-care workflows. Rather than proposing another disease-specific predictive model, this work presents a modular, clinically oriented AI framework that integrates imaging analysis, patient-reported outcomes, structured clinical information, and retrieval-aware language synthesis within a unified, clinician-supervised workflow to support future multimodal decision-making in spine care. Framework Design: The conceptual, implementation-independent architecture comprises three interoperable components: (1) an imaging-analysis module supporting automated interpretation and severity grading with human-in-the-loop validation; (2) a multimodal assessment component integrating patient-reported outcomes with demographic and clinical variables; and (3) a retrieval-aware large language model (LLM) synthesis layer generating clinician-reviewable radiological summaries and contextualized clinical considerations grounded in structured evidence. Core design principles include interoperability, explainability, governance, security, and human-centered workflow integration. Early Feasibility Demonstration: An exploratory proof-of-concept prototype incorporating AI-assisted lumbar MRI interpretation demonstrated how selected framework components may support explainable imaging analysis, multimodal contextualization, editable reporting, and clinician supervision. This demonstration illustrates architectural feasibility and workflow integration only, and should not be interpreted as evidence of diagnostic accuracy, clinical effectiveness, or generalizable clinical performance. Conclusions: The proposed framework provides an implementation-independent translational blueprint for future technical development, prospective multicentre validation, regulatory evaluation, and real-world deployment of clinically integrated, trustworthy multimodal AI systems for spine care.