Use of Artificial Intelligence for Posture Correction: A Systematic Review Using PRISMA Guidelines and PICO Framework

Abstract

Objective: To methodically locate and compile peer-reviewed RCTs, quasi-experimental studies, cohort studies, and
crossover trials (2014–2026) that assess AI-based posture detection, monitoring, or correction systems in human subjects.
Methodology: The PRISMA 2020 statement and a PICO(S) framework were used to organize this review, which was limited
to AI/ML-driven systems (such as computer-vision pose estimation, deep neural networks, or ML classifiers) assessed in
comparative human-trial designs. Using combinations of "artificial intelligence," "machine learning," "deep learning,"
"posture," "posture correction," "posture monitoring," "randomized controlled trial," "quasi-experimental," "cohort," and
"crossover," a multi-database exploratory search was carried out across PubMed/MEDLINE, IEEE Xplore, Scopus, ACM
Digital Library, and ProQuest Central, supplemented by JMIR/PMC full-text repositories and Google Scholar-indexed
sources. PRISMA 2020 guidelines were followed for screening, data extraction, and risk-of-bias assessment (Cochrane
RoB 2 for RCTs; ROBINS-I for non-randomized designs).
Results: 55 distinct studies (out of 57 citation records) satisfied the eligibility requirements and were added to the narrative
synthesis after duplicate cross-database indexing was eliminated. Five technological families comprised the included
studies: (1) camera/video-based human pose estimation (HPE); (2) wearable pressure-sensor and inertial measurement
unit (IMU) systems; (3) hybrid AIoT and edge-AI platforms; (4) explainable and causal AI frameworks for spinal/
postural monitoring; and (5) AI-assisted clinical and educational interventions assessed in randomized or quasiexperimental
designs. Numerous intervention trials indicated statistically significant changes in postural measurements,
pain, disability, and functional outcomes, and reported categorization accuracies were consistently high (usually 90–
99%). However, the majority of technical validation studies were in the prototype or laboratory stage, sample numbers
were often small, external validation was few, and outcome measures were inconsistent, making meta-analysis impossible.
Conclusions: Strong technical accuracy and growing clinical potential are demonstrated by AI-based posture systems,
especially for biofeedback-based correction, ergonomic risk monitoring, and tele-rehabilitation. Future research should
focus on multi-center randomised trials, standardized outcome reporting, external and cross-population validation, and
transparent reporting of AI model development in accordance with PRISMA-AI guidelines because there is still relatively
little evidence at the clinical-effectiveness level.
Registration: Prospectively registered on open Science Framework (OSF) Registration DOI: 10.17605/OSF.IO/E9PBH
Keywords: artificial intelligence; machine learning; deep learning; posture detection; posture correction; pose estimation;
rehabilitation; PRISMA 2020; PICO framework; systematic review