Abstract Background Poor posture and related musculoskeletal conditions represent a growing global health concern. Conventional postural assessment methods are often subjective, intermittent, and insufficient for accurate, continuous monitoring. Advances in artificial intelligence (AI), particularly in computer vision and human pose estimation (HPE), have introduced new possibilities for objective and real-time postural analysis. Main body This critical review synthesizes and evaluates current developments in AI technologies for postural management. The review draws on recent literature from computer science, bioengineering, and clinical research, focusing on studies from the past decade that explore the use of AI and HPE in the detection, monitoring, and correction of human posture. AI-based HPE models demonstrate high precision in identifying anatomical landmarks and quantifying postural parameters, offering a robust alternative to traditional assessment methods. Applications are expanding beyond laboratory environments to practical contexts such as ergonomic risk evaluation and sports performance analysis. In addition, AI-driven systems that deliver real-time feedback and support tele-rehabilitation are enhancing user engagement and enabling personalized interventions. Despite these advancements, the field faces several challenges. Evidence from large-scale clinical trials remains limited, and the generalizability of existing models across diverse populations and real-world conditions is uncertain. Concerns related to usability, data privacy, and integration within healthcare systems also pose significant barriers to clinical translation. Conclusion AI holds considerable potential to transform postural management through continuous, objective, and accessible assessment and intervention. To fully realize this potential, future work must extend beyond technical innovation to include rigorous clinical validation, user-centered design, and the establishment of ethical and regulatory frameworks that ensure safe, effective, and equitable implementation.
Köroglu et al. (Wed,) studied this question.