使用基于机器学习的模型对脊髓神经鞘瘤和脑膜瘤进行术前鉴别：一项系统评价和荟萃分析

Preoperative Differentiation of Spinal Schwannoma and Meningioma Using Machine Learning-Based Models: A Systematic Review and Meta-Analysis.

作者信息

Hajikarimloo Bardia, Mohammadzadeh Ibrahim, Hashemi Rana, Sheikhzadeh Mohsen, Najari Dorsa, Hezaveh Ehsan Bahrami, Ghorbanpouryami Fatemeh, Habibi Mohammad Amin

机构信息

Department of Neurosurgery, Shohada Tajrish Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran.

Skull Base Research Center, Loghman-Hakim Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran.

出版信息

World Neurosurg. 2025 Jul;199:124096. doi: 10.1016/j.wneu.2025.124096. Epub 2025 May 19.

DOI:10.1016/j.wneu.2025.124096

PMID:40398809

Abstract

BACKGROUND

Regarding the differences in surgical approaches for spinal schwannomas and meningiomas, preoperative differentiation of spinal schwannomas and meningiomas can be important in managing these lesions. This study evaluated the diagnostic performance of machine learning (ML)-based models in the differentiation of spinal schwannomas and meningiomas.

METHODS

On December 18, 2024, a comprehensive search was conducted. The data for the best-performing model were used to calculate pooled sensitivity, specificity, area under the curve (AUC), and diagnostic odds ratio.

RESULTS

Six studies with 644 patients were included, encompassing 364 schwannomas (59.9%) and 258 meningiomas (40.1%). Deep learning-based models (66.7%, 4/6) were the most frequent, followed by ML-based models (33.3%, 2/6). The best performance models' AUC and accuracy ranged from 0.876 to 0.998 and 0.8 to 0.982, respectively. Our findings showed a pooled sensitivity rate of 91% (95%CI: 81%-96%), a specificity rate of 92% (95%CI: 84%-96%), and a diagnostic odds ratio of 97.34 (95%CI: 23.5-403.6), concurrent with an AUC of 0.944.

CONCLUSIONS

ML-based models have a high diagnostic accuracy in preoperative differentiation of spinal schwannomas and meningiomas.

摘要

背景

鉴于脊柱神经鞘瘤和脊膜瘤手术方法的差异，术前区分脊柱神经鞘瘤和脊膜瘤对于处理这些病变可能很重要。本研究评估了基于机器学习（ML）的模型在区分脊柱神经鞘瘤和脊膜瘤方面的诊断性能。

方法

于2024年12月18日进行全面检索。使用表现最佳模型的数据来计算合并敏感度、特异度、曲线下面积（AUC）和诊断比值比。

结果

纳入了6项研究，共644例患者，其中包括364例神经鞘瘤（59.9%）和258例脊膜瘤（40.1%）。基于深度学习的模型最为常见（66.7%，4/6），其次是基于ML的模型（33.3%，2/6）。表现最佳模型的AUC和准确率分别在0.876至0.998以及0.8至0.982之间。我们的研究结果显示合并敏感度为91%（95%CI：81%-96%），特异度为92%（95%CI：84%-96%），诊断比值比为97.34（95%CI：23.5-403.6），AUC为0.944。