Artificial Intelligence Versus Manual Measurement of Spinopelvic Sagittal Parameters Across Preoperative, Postoperative, and Instrumented Spines: A Systematic Review of Diagnostic Accuracy, Failure Rates, and Inferred Surgical Planning Implications.
7 affiliations
- 1Neurology, Instituto de Estudios Superiores de Chiapas, Tuxtla Gutiérrez, MEX.
- 2Medicine Department, Fundación Universitaria de Ciencias de la Salud, Bogotá, COL.
- 3Neuroradiology, Massachusetts General Hospital, Boston, USA.
- 4Healthcare Management and Administration, Private Practice, Florida, USA.
- 5Tropical Medicine, Ministerio de Salud Pública, Guayaquil, ECU.
- 6Medicine, Ministerio de Salud Pública, Montecristi, ECU.
- 7General Practice, Universidad Autónoma de Guadalajara, Guadalajara, MEX.
Abstract
Adult spinal deformity (ASD) demands precise radiographic quantification of spinopelvic parameters to guide corrective surgery. Artificial intelligence (AI) and machine learning (ML) tools offer automated alternatives to manual measurement. This systematic review evaluates the diagnostic accuracy, failure rates, and inferred surgical planning implications of AI-based spinopelvic measurement versus manual assessment, with particular focus on postoperative and instrumented spines. This Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020-compliant systematic review searched MEDLINE (PubMed), ScienceDirect, Cochrane Library, and Google Scholar (first 200 results) through June 2026. Studies comparing AI/ML tools, including convolutional neural networks (CNNs), U-Net, Mask R-CNN, and Keypoint R-CNN, against manual clinician measurement on radiographs or EOS biplanar systems were included. Quality was assessed using QUADAS-2. Narrative synthesis was performed given heterogeneity across architectures, populations, and outcome metrics. Sixteen studies (sample sizes ranging from 21 to 1,810 patients per study) compared machine vision against manual measurement across preoperative, postoperative, and instrumented cohorts. Compared to manual measurement, automated machine vision demonstrated agreement statistically comparable to manual measurement for sagittal vertical axis (SVA) (intraclass correlation coefficient (ICC) 0.99-1.00) and pelvic tilt (PT) (ICC 0.96-0.99), matching experienced clinician inter-rater reliability. For lumbar lordosis, automated systems achieved acceptable accuracy in preoperative populations (ICC 0.91-0.97) but exhibited directional biases absent in manual measurement (up to 6.31° overestimation). Thoracic kyphosis and sacral slope agreement was highly variable and dataset-dependent (ICC ranging from 0.58 to 0.97 across studies), generally among the least consistent parameters rather than uniformly weak. Unlike manual measurement, automated accuracy declined significantly in postoperative and instrumented spines. Machine vision failure rates ranged from 0% to 76.2%, versus near-zero manual failure rates. Processing speed was up to 238-fold faster than manual workflows. Automated-versus-manual discrepancies for LL and PI occasionally exceeded the 10° PI-LL surgical correction threshold used as a proxy for surgical-planning impact, since no included study measured a surgical outcome directly. Machine vision matches manual speed and accuracy preoperatively but is not yet reliable enough postoperatively to replace manual clinical assessment without standardised validation.
Related
- Preoperative CT-derived Hounsfield unit values and mechanical complications after thoracolumbar or lumbar instrumented fusion: a systematic review and meta-analysis.
- Preoperative Zinc Deficiency as a Novel Prognostic Predictor for Postoperative Recurrence in Hepatocellular Carcinoma
- Preoperative Social Connection and Postoperative Outcomes in Adults Undergoing Surgery: A Systematic Review and Meta-analysis
- Preoperative risk stratification for postoperative inflammatory–nutritional deterioration after colorectal cancer surgery: an interpretable machine learning study
- Preoperative and early postoperative predictors of nausea and vomiting after laparoscopic abdominal surgery in an enhanced recovery programme
BibTeX
@article{mendez2026artificial,
title = {{Artificial Intelligence Versus Manual Measurement of Spinopelvic Sagittal Parameters Across Preoperative, Postoperative, and Instrumented Spines: A Systematic Review of Diagnostic Accuracy, Failure Rates, and Inferred Surgical Planning Implications.}},
author = {Evelyn Abril Jiménez Méndez and Elkin Armando Gallo Tafur and Rafael E Martinez Imbett and Edison Alejandro Hablich Navarrete and Luis Honorio Velez Alarcon and Mishell Alejandra Palacios Cedeño and Antonio Cueva Pérez},
journal = {Cureus},
year = {2026},
month = sep,
doi = {10.7759/cureus.116863},
url = {https://europepmc.org/article/MED/42927406},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial Intelligence Versus Manual Measurement of Spinopelvic Sagittal Parameters Across Preoperative, Postoperative, and Instrumented Spines: A Systematic Review of Diagnostic Accuracy, Failure Rates, and Inferred Surgical Planning Implications.
AU - Evelyn Abril Jiménez Méndez
AU - Elkin Armando Gallo Tafur
AU - Rafael E Martinez Imbett
AU - Edison Alejandro Hablich Navarrete
AU - Luis Honorio Velez Alarcon
AU - Mishell Alejandra Palacios Cedeño
AU - Antonio Cueva Pérez
PY - 2026
DA - 2026/09/24
JO - Cureus
DO - 10.7759/cureus.116863
UR - https://europepmc.org/article/MED/42927406
AB - Adult spinal deformity (ASD) demands precise radiographic quantification of spinopelvic parameters to guide corrective surgery. Artificial intelligence (AI) and machine learning (ML) tools offer automated alternatives to manual measurement. This systematic review evaluates the diagnostic accuracy, failure rates, and inferred surgical planning implications of AI-based spinopelvic measurement versus manual assessment, with particular focus on postoperative and instrumented spines. This Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020-compliant systematic review searched MEDLINE (PubMed), ScienceDirect, Cochrane Library, and Google Scholar (first 200 results) through June 2026. Studies comparing AI/ML tools, including convolutional neural networks (CNNs), U-Net, Mask R-CNN, and Keypoint R-CNN, against manual clinician measurement on radiographs or EOS biplanar systems were included. Quality was assessed using QUADAS-2. Narrative synthesis was performed given heterogeneity across architectures, populations, and outcome metrics. Sixteen studies (sample sizes ranging from 21 to 1,810 patients per study) compared machine vision against manual measurement across preoperative, postoperative, and instrumented cohorts. Compared to manual measurement, automated machine vision demonstrated agreement statistically comparable to manual measurement for sagittal vertical axis (SVA) (intraclass correlation coefficient (ICC) 0.99-1.00) and pelvic tilt (PT) (ICC 0.96-0.99), matching experienced clinician inter-rater reliability. For lumbar lordosis, automated systems achieved acceptable accuracy in preoperative populations (ICC 0.91-0.97) but exhibited directional biases absent in manual measurement (up to 6.31° overestimation). Thoracic kyphosis and sacral slope agreement was highly variable and dataset-dependent (ICC ranging from 0.58 to 0.97 across studies), generally among the least consistent parameters rather than uniformly weak. Unlike manual measurement, automated accuracy declined significantly in postoperative and instrumented spines. Machine vision failure rates ranged from 0% to 76.2%, versus near-zero manual failure rates. Processing speed was up to 238-fold faster than manual workflows. Automated-versus-manual discrepancies for LL and PI occasionally exceeded the 10° PI-LL surgical correction threshold used as a proxy for surgical-planning impact, since no included study measured a surgical outcome directly. Machine vision matches manual speed and accuracy preoperatively but is not yet reliable enough postoperatively to replace manual clinical assessment without standardised validation.
SN - 2168-8184
ER -