Toward a clinically integrated AI framework for personalized spine care

Background and Objective: Low back pain remains the leading global cause of disability and is driven by a multifactorial interplay of structural, clinical, behavioral, and psychosocial factors. Although artificial intelligence has demonstrated strong potential in spine imaging and outcome prediction, existing solutions are typically single-purpose and do not reflect this complexity. This work proposes a modular artificial intelligence framework designed to unify imaging analysis, patient-reported outcomes, and clinical reasoning within a single, clinically deployable ecosystem. Methods: We developed a conceptual architecture comprising three integrated modules: (1) an automated imaging analysis and severity grading system with radiologist-in-the-loop validation through a dedicated user interface; (2) a multimodal assessment engine incorporating patient-reported outcomes such as pain intensity, disability score, quality-of-life scores, coping strategies, as well as demographic and clinical factors; and (3) a language-generation module based on large language models for synthesizing imaging findings and patient-reported data into editable radiology reports and personalized clinical considerations. The design emphasizes security, interpretability, human-centered interfaces, and interoperability with existing hospital systems.Results:To explore the feasibility of integrating patient-reported outcome measures (PROMs) with AI-assisted imaging assessment for personalized treatment planning, a small proof-of-concept analysis was conducted using real-world patient data. Three representative patients with distinct psychosocial profiles were examined to demonstrate how combining AI-generated imaging findings with structured PROMs can shift treatment recommendations beyond what imaging alone would suggest. These illustrative cases showcase the practical potential of augmenting automated radiology reports with patient-reported contextual data to better guide individualized clinical management. This analysis is provided as supplementary material to demonstrate feasibility rather than to draw quantitative conclusions.Conclusions:The proposed framework outlines a clear pathway toward a comprehensive, multimodal AI decision-support ecosystem for spine care. By unifying imaging analysis, patient-reported measures, and AI-assisted reporting, it highlights how modular architectures can better reflect the biopsychosocial nature of LBP and support future clinical translation. The proof-of-concept findings provide initial evidence of feasibility and motivate subsequent large-scale validation.