Non-invasive preoperative MRI-based deep transfer learning radiomics model for predicting meningioma brain invasion.
4 affiliations
- 1Department of Radiology, Huanhu Hospital Affiliated to Tianjin Medical University, Tianjin, China.
- 2Tianjin Key Laboratory of Cerebral Blood Flow Reconstruction and Head and Neck Tumor New Technology Translation, Tianjin, China.
- 3Department of Radiology, Nayong Xinli Hospital, Bijie, Guizhou, China.
- 4Institute of Neurosurgery, Huanhu Hospital Affiliated to Tianjin Medical University, Tianjin, China.
Abstract
Background
Current preoperative assessment of meningioma brain invasion (BI) demands imaging analytical frameworks that optimize computational efficiency while capturing multidimensional tumor-pathology correlations. Building upon the inherent advantages of 2.5D MRI in synthesizing multiplanar biological data with manageable processing loads, this study innovatively integrates deep transfer learning (DTL), a technique bridging pretrained network knowledge with target domain adaptation, to systematically identify tumor-invasive biological signatures and develop an optimized predictive system using preoperative contrast-enhanced T1-weighted imaging (CE-T1WI).Methods
This retrospective study enrolled 674 patients with pathologically confirmed meningiomas, including 53 with BI and 621 without BI. The cohort was stratified into a training set and an independent test set. Tumor regions of interest (ROIs) were delineated on CE-T1WI, and radiomics and DTL features were extracted. Radiomics and DTL features were fused via early fusion, followed by feature selection. Within the training cohort, five-fold cross-validation was performed, with SMOTE applied exclusively to the training portion of each fold; validation folds and the independent test set retained their original class distributions.Results
The effectiveness of the models was compared using the area under the ROC curve (AUC). In the training set, ComM achieved an AUC of 0.977, while in the test set, its AUC reached 0.935. ComM demonstrated the best performance (training set: AUCComM > AUCDTLRM > AUCRadM > AUCClinM > AUCDTLM; test set: AUCComM > AUCDTLRM > AUCRadM > AUCDTLM > AUCClinM), showing excellent preoperative predictive capability for meningioma BI. The Hosmer-Lemeshow test indicated good model fit, and the calibration curve revealed that ComM was closest to the ideal curve.Conclusion
ClinM, RadM, DTLM, DTLRM, and ComM based on 2.5D CE-T1WI all exhibited good performance in predicting meningioma BI. The clinical-radiomics-DTL fusion model (ComM) demonstrated the highest efficacy, suggesting that DTL provides additional features correlated with meningioma BI that differ from radiomics features. This model shows potential clinical utility for preoperative prediction; however, further external validation in larger multicenter cohorts is required before clinical implementation.
Related
- MRI-based preoperative WHO grade prediction in meningioma: development and comparison of deep learning, radiomics, clinical, and combined models in multicenter data.
- Multiparametric MRI radiomics and clinical features for preoperative prediction of high-grade meningioma: a two-centre retrospective study
- Development and Internal Validation of the BRICS Score: A Preoperative Clinical and Hematological Model for Predicting High-Grade and Muscle-Invasive Bladder Cancer
- Development of a multi-parameter MRI-based habitat radiomics model for differential diagnosis of vestibular schwannoma and meningioma in the cerebellopontine angle: a multi-center study.
- CT-based multi-station lymph-node radiomics for preoperative prediction of lymph node metastasis in colorectal cancer: a proof-of-concept comparison with primary-tumor radiomics
BibTeX
@article{yuan2026noninvasive,
title = {{Non-invasive preoperative MRI-based deep transfer learning radiomics model for predicting meningioma brain invasion.}},
author = {Dong Yuan and Jiajia Zhang and Chao Zhang and Xiyue Jing and Quanzhi Feng and Tong Han},
journal = {Frontiers in oncology},
year = {2026},
month = oct,
doi = {10.3389/fonc.2026.1935311},
url = {https://europepmc.org/article/MED/42908707},
note = {CC BY},
}
RIS
TY - JOUR
TI - Non-invasive preoperative MRI-based deep transfer learning radiomics model for predicting meningioma brain invasion.
AU - Dong Yuan
AU - Jiajia Zhang
AU - Chao Zhang
AU - Xiyue Jing
AU - Quanzhi Feng
AU - Tong Han
PY - 2026
DA - 2026/10/07
JO - Frontiers in oncology
DO - 10.3389/fonc.2026.1935311
UR - https://europepmc.org/article/MED/42908707
AB - Background Current preoperative assessment of meningioma brain invasion (BI) demands imaging analytical frameworks that optimize computational efficiency while capturing multidimensional tumor-pathology correlations. Building upon the inherent advantages of 2.5D MRI in synthesizing multiplanar biological data with manageable processing loads, this study innovatively integrates deep transfer learning (DTL), a technique bridging pretrained network knowledge with target domain adaptation, to systematically identify tumor-invasive biological signatures and develop an optimized predictive system using preoperative contrast-enhanced T1-weighted imaging (CE-T1WI).Methods This retrospective study enrolled 674 patients with pathologically confirmed meningiomas, including 53 with BI and 621 without BI. The cohort was stratified into a training set and an independent test set. Tumor regions of interest (ROIs) were delineated on CE-T1WI, and radiomics and DTL features were extracted. Radiomics and DTL features were fused via early fusion, followed by feature selection. Within the training cohort, five-fold cross-validation was performed, with SMOTE applied exclusively to the training portion of each fold; validation folds and the independent test set retained their original class distributions.Results The effectiveness of the models was compared using the area under the ROC curve (AUC). In the training set, ComM achieved an AUC of 0.977, while in the test set, its AUC reached 0.935. ComM demonstrated the best performance (training set: AUCComM > AUCDTLRM > AUCRadM > AUCClinM > AUCDTLM; test set: AUCComM > AUCDTLRM > AUCRadM > AUCDTLM > AUCClinM), showing excellent preoperative predictive capability for meningioma BI. The Hosmer-Lemeshow test indicated good model fit, and the calibration curve revealed that ComM was closest to the ideal curve.Conclusion ClinM, RadM, DTLM, DTLRM, and ComM based on 2.5D CE-T1WI all exhibited good performance in predicting meningioma BI. The clinical-radiomics-DTL fusion model (ComM) demonstrated the highest efficacy, suggesting that DTL provides additional features correlated with meningioma BI that differ from radiomics features. This model shows potential clinical utility for preoperative prediction; however, further external validation in larger multicenter cohorts is required before clinical implementation.
SN - 2234-943X
ER -