Scholarlink

Background <p>The study of spinopelvic alignment in asymptomatic individuals is essential for understanding physiological sagittal balance and establishing reference values for spinal deformity assessment. However, the availability of large datasets of healthy subjects is limited by ethical, logistical, and radiation-related constraints. Artificial intelligence (AI)-based synthetic data generation may represent a promising strategy to overcome these limitations.</p> Methods <p>Full-spine standing radiographs from 123 asymptomatic subjects were retrospectively analyzed. Demographic characteristics and multiple spinopelvic parameters, including pelvic incidence (PI), pelvic tilt (PT), sacral slope (SS), lumbar lordosis (LL), thoracic kyphosis (TK), and cervical alignment measures, were recorded. An AI-driven probabilistic Gaussian resampling approach with anatomical constraints was used to generate a synthetic dataset of 10,000 biologically plausible cases. Correlations identified within the synthetic dataset were subsequently validated against the original cohort using Pearson correlation analysis and bootstrap resampling (1,000 iterations).</p> Results <p>The synthetic dataset preserved the statistical distribution of the original population while substantially increasing analytical power. Significant correlations were identified between PI and PT (PT = 0.34 × PI − 7.2), PI and SS (SS = 0.66 × PI + 7.2), and PI and LL (|LL| = 0.55 × PI + 32.0). These relationships were consistent with previously published anatomical models and remained robust when tested in the original cohort and through bootstrap validation. No significant correlation was observed between PI and TK. A significant association was also identified between T1 slope and cervical lordosis.</p> Conclusions <p>AI-driven dataset amplification represents a feasible and reproducible approach for investigating spinopelvic relationships in limited clinical cohorts. The combination of synthetic data generation, validation on real-world observations, and bootstrap resampling enables the identification of biologically plausible correlations while minimizing the need for additional imaging studies. This methodology may serve as a valuable exploratory tool in spine research and other fields characterized by limited datasets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence–driven data expansion for the validation of spinopelvic parameter correlations in asymptomatic subjects

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Domenico Compagnone,
  • Davide Lamartina,
  • Marco Giacalone,
  • Riccardo Cecchinato,
  • Pedro Berjano,
  • Claudio Lamartina

摘要

Background

The study of spinopelvic alignment in asymptomatic individuals is essential for understanding physiological sagittal balance and establishing reference values for spinal deformity assessment. However, the availability of large datasets of healthy subjects is limited by ethical, logistical, and radiation-related constraints. Artificial intelligence (AI)-based synthetic data generation may represent a promising strategy to overcome these limitations.

Methods

Full-spine standing radiographs from 123 asymptomatic subjects were retrospectively analyzed. Demographic characteristics and multiple spinopelvic parameters, including pelvic incidence (PI), pelvic tilt (PT), sacral slope (SS), lumbar lordosis (LL), thoracic kyphosis (TK), and cervical alignment measures, were recorded. An AI-driven probabilistic Gaussian resampling approach with anatomical constraints was used to generate a synthetic dataset of 10,000 biologically plausible cases. Correlations identified within the synthetic dataset were subsequently validated against the original cohort using Pearson correlation analysis and bootstrap resampling (1,000 iterations).

Results

The synthetic dataset preserved the statistical distribution of the original population while substantially increasing analytical power. Significant correlations were identified between PI and PT (PT = 0.34 × PI − 7.2), PI and SS (SS = 0.66 × PI + 7.2), and PI and LL (|LL| = 0.55 × PI + 32.0). These relationships were consistent with previously published anatomical models and remained robust when tested in the original cohort and through bootstrap validation. No significant correlation was observed between PI and TK. A significant association was also identified between T1 slope and cervical lordosis.

Conclusions

AI-driven dataset amplification represents a feasible and reproducible approach for investigating spinopelvic relationships in limited clinical cohorts. The combination of synthetic data generation, validation on real-world observations, and bootstrap resampling enables the identification of biologically plausible correlations while minimizing the need for additional imaging studies. This methodology may serve as a valuable exploratory tool in spine research and other fields characterized by limited datasets.