Machine learning maps multiplanar spinal surface alignment and patterns of change across heterogeneous cohorts

Surface-derived thoracolumbar-pelvic alignment measures span sagittal, frontal, and transverse planes, but their multivariate organization remains incompletely characterized. This retrospective pooled analysis comprised a baseline cross-sectional sample of 468 participants from four cohorts (CGH, n = 200; FMS, n = 120; LSR, n = 80; AIS, n = 68) and a separate paired sample of 268 participants from the three intervention cohorts (FMS, n = 120; LSR, n = 80; AIS, n = 68). The CGH cohort was a matched cross-sectional study with one baseline assessment and was excluded from all post-treatment, change-score, and multiplanar change-magnitude analyses. In the primary cohort-centered baseline analysis, the first two PCA components explained 83.3% of variance, and CCA yielded Rc = 0.926 (95% cohort-stratified bootstrap CI 0.915–0.939; within-cohort permutation p < 0.001). In the cohort-centered paired analysis, PCA of post-minus-baseline changes explained 81.2% and 6.7% of variance in PC1 and PC2, respectively (87.9% cumulative), and CCA yielded Rc = 0.917 (95% cohort-stratified bootstrap CI 0.898–0.935; within-cohort permutation p < 0.001; repeated cohort-stratified cross-validated Rc = 0.912 ± 0.022). Exploratory K-means clustering yielded two partially overlapping descriptive profiles; an alternative Gaussian mixture analysis did not support two discrete latent subgroups. The profiles were not validated against independent clinical outcomes. Three-class classification using change scores alone was modest (balanced accuracy 0.418 ± 0.072; macro-AUC 0.610 ± 0.035). The paired findings describe multiplanar covariance of observed within-participant change across different interventions and follow-up intervals. They do not establish a shared adaptive mechanism or a common treatment effect, and they do not apply to the baseline-only CGH cohort.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-15
DOI
https://doi.org/10.1038/s41598-026-80472-7
Primary Topic
Scoliosis diagnosis and treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine learning maps multiplanar spinal surface alignment and patterns of change across heterogeneous cohorts

Paul A. Oakley, Iman Akef Khowailed, Ibrahim M. Moustafa, Deed E. Harrison et al.
Scientific Reports
Scoliosis diagnosis and treatment
article

Machine learning maps multiplanar spinal surface alignment and patterns of change across heterogeneous cohorts

Paul A. Oakley, Iman Akef Khowailed, Ibrahim M. Moustafa, Deed E. Harrison, Emre Selman Caniaz, Fatimah W. Abdelrahman
article en

Abstract

Surface-derived thoracolumbar-pelvic alignment measures span sagittal, frontal, and transverse planes, but their multivariate organization remains incompletely characterized. This retrospective pooled analysis comprised a baseline cross-sectional sample of 468 participants from four cohorts (CGH, n = 200; FMS, n = 120; LSR, n = 80; AIS, n = 68) and a separate paired sample of 268 participants from the three intervention cohorts (FMS, n = 120; LSR, n = 80; AIS, n = 68). The CGH cohort was a matched cross-sectional study with one baseline assessment and was excluded from all post-treatment, change-score, and multiplanar change-magnitude analyses. In the primary cohort-centered baseline analysis, the first two PCA components explained 83.3% of variance, and CCA yielded Rc = 0.926 (95% cohort-stratified bootstrap CI 0.915–0.939; within-cohort permutation p < 0.001). In the cohort-centered paired analysis, PCA of post-minus-baseline changes explained 81.2% and 6.7% of variance in PC1 and PC2, respectively (87.9% cumulative), and CCA yielded Rc = 0.917 (95% cohort-stratified bootstrap CI 0.898–0.935; within-cohort permutation p < 0.001; repeated cohort-stratified cross-validated Rc = 0.912 ± 0.022). Exploratory K-means clustering yielded two partially overlapping descriptive profiles; an alternative Gaussian mixture analysis did not support two discrete latent subgroups. The profiles were not validated against independent clinical outcomes. Three-class classification using change scores alone was modest (balanced accuracy 0.418 ± 0.072; macro-AUC 0.610 ± 0.035). The paired findings describe multiplanar covariance of observed within-participant change across different interventions and follow-up intervals. They do not establish a shared adaptive mechanism or a common treatment effect, and they do not apply to the baseline-only CGH cohort.

Scientific Reports
Balıkesir University (TR), Cairo University (EG), York University (CA), University of Sharjah (AE)
Scoliosis diagnosis and treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon