University of Sharjah researchers test spinal alignment analysis


A research team led by the University of Sharjah published a peer-reviewed study in Scientific Reports examining the ability of statistical and machine learning methods to analyze patterns of thoracic, lumbar, and pelvic spinal alignment in 468 participants, revealing strong statistical correlations versus limited clinical predictive power for subclassification.
The University of Sharjah study presents a model of scientific rigor in biomedical research, demonstrating that the success of machine learning in extracting strong mathematical correlations from spinal data does not automatically validate its suitability for guiding clinical treatment.
On November 15, 2026, a research team led by the University of Sharjah in the United Arab Emirates published a peer-reviewed machine learning-based scientific analysis in Scientific Reports, examining kinematic and structural patterns in thoracic, lumbar, and pelvic spinal alignment with participation from researchers at scientific institutions in Egypt, Turkey, Canada, and the United States.
The researchers utilized baseline data covering 468 participants drawn from four research groups, alongside a separate paired dataset of 268 participants from three groups that underwent therapeutic interventions; the fourth group, which provided baseline measurements only, was excluded from the post-intervention change analysis to avoid producing misleading results.
The scientific team relied on “principal component analysis” techniques to describe overarching trends of variance across multiple measurements, “canonical correlation analysis” to evaluate how two sets of measurements relate to one another, and “exploratory clustering” methods to suggest groups with similar patterns.
The paper emphasized a strict methodological principle: groups suggested by statistical clustering algorithms do not automatically represent distinct medical conditions or separate real-world disease categories.
In the primary analysis, the first two principal components accounted for approximately 83.3% of the variance, while in the paired analysis they accounted for nearly 87.9% of the variance in the measured change, recording strong canonical correlations between selected measurement sets. However, these statistical findings demonstrate relationships within the available data and do not prove the capability of an artificial intelligence system to diagnose a medical condition or predict treatment success.
When testing exploratory clustering, one method suggested the presence of two partially overlapping patterns, but the alternative “Gaussian mixture” analysis did not support the conclusion that two truly distinct latent categories exist. Furthermore, a three-class classification using change scores recorded modest performance with a balanced accuracy of about 0.418 and an overall area under the receiver operating characteristic curve of approximately 0.610.
This work represents a retrospective secondary analysis of previously anonymized studies rather than a new prospective clinical trial, and it does not prove the existence of a shared biological mechanism behind the observed changes or the equal efficacy of different treatments.
The study’s prominent scientific value lies in its transparency and analytical discipline in not merely presenting strong correlation figures while concealing limited classification performance. Moving toward future clinical guidance and patient rehabilitation requires proactive studies that verify classifications, account for varying follow-up periods, and measure the actual impact on physical therapy and rehabilitation decisions.
In this sound methodological sense, the study by the University of Sharjah and its international partners serves as an important scientific reminder to researchers and medical rehabilitation app developers that the abundance of kinematic data and the ability of dimensionality reduction algorithms to extract high mathematical correlations are not enough on their own to innovate new diagnostic patient classifications. True merit in evidence-based medicine remains contingent upon testing these computational patterns against independent clinical and therapeutic outcomes in rigorous future studies that account for individual differences and temporal follow-up periods while demonstrating tangible improvements in evaluating back pain cases and guiding physical therapy programs.
Moreover, the research team’s transparent disclosure of the limited accuracy of the three-class classification reflects a high academic commitment that protects physicians and clinical practitioners from prematurely relying on statistical divisions whose therapeutic feasibility for patients has not yet been proven. This enhances the standing of cross-border scientific collaboration between Arab and international universities in establishing rigorous methodological standards for applied artificial intelligence research in motor rehabilitation and orthopedic sciences.
Question: How many participants were included in the University of Sharjah team’s analysis?
Answer: The primary analysis was based on data from 468 participants across four research groups, while the post-therapeutic intervention change analysis included 268 participants from three groups.
Question: Why did the researchers warn against overemphasizing the clinical interpretation of the study’s results?
Answer: Because the Gaussian mixture analysis did not clearly confirm the existence of separate disease categories, and the three-class classification recorded a modest balanced accuracy of about 0.418 despite strong initial statistical correlations.
Question: Which countries do the researchers participating in the study hail from, alongside the UAE?
Answer: The study was prepared by researchers from the University of Sharjah in collaboration with research institutions in Egypt, Turkey, Canada, and the United States.