Machine Learning Integration of MRI Intratumoral and Peritumoral Radiomics Features for Predicting PNSTs
Jia Hao Liu1, Fang Ying Tang1, Ji Feng Wang1
1Wound Repair Department, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China (J.H.L, F.Y.T., J.F.W., Y.H.C., R.W.H., J.W.); Hand Surgery & Peripheral Nerve Surgery, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China (J.H.L, F.Y.T., J.F.W., Y.H.C., R.W.H., J.W.).
Academic Radiology
|February 24, 2026
Summary
Integrating intratumoral and peritumoral radiomics effectively predicts peripheral nerve sheath tumor complications. Fusion models show superior performance, offering a robust clinical pathway for risk assessment.
Area of Science:
- Radiology
- Oncology
- Medical Imaging
Background:
- Postoperative complications occur in 15-76% of patients with peripheral nerve sheath tumors (PNSTs).
- Objective preoperative risk stratification tools for PNSTs are currently lacking.
- Magnetic resonance imaging (MRI) radiomics research has primarily focused on intratumoral regions, leaving the predictive value of the peritumoral microenvironment unclear.
Purpose of the Study:
- To investigate the predictive value of integrating intratumoral and peritumoral radiomics for postoperative complications in PNST patients.
- To develop and evaluate radiomics models for preoperative risk assessment of PNST complications.
Main Methods:
- A retrospective study of 280 PNST patients with preoperative MRI.
- Segmentation of intratumoral and peritumoral regions (2 mm expansion) for radiomic feature extraction.
- Development and comparison of four models: intratumoral, peritumoral, fused region (Imagefusion), and concatenated features (intraPeri2mm), evaluated using AUC and DCA.
Main Results:
- Radiomics models significantly outperformed the clinical-only model (p<0.001).
- Fusion models (intraPeri2mm AUC 0.899, Imagefusion AUC 0.895) demonstrated superior performance compared to single-region models.
- The Imagefusion + clinical pathway showed robust clinical net benefit, especially with standardized data.
Conclusions:
- Integration of intratumoral and peritumoral radiomics provides effective preoperative prediction of PNST postoperative complications.
- The Imagefusion + clinical pathway is recommended for standardized clinical data settings.
- An intraPeri2mm-only strategy may be suitable for settings with limited clinical data.
Keywords:
Machine learningMagnetic resonance imagingPeripheral Nerve Sheath TumorsPostoperative complicationsRadiomics
