Predictive value of multi-modal conventional MRI radiomics for early postoperative glioma recurrence(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2025年第2期
- Page:
- 208-212
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Predictive value of multi-modal conventional MRI radiomics for early postoperative glioma recurrence
- Author(s):
- ZHANG Yuhui1; YANG Yingsi1; FAN Weixiong1; JIANG Guihua2; XIONG Xiaoli1; YANG Rihui1
- 1. Department of Magnetic Resonance Imaging, Meizhou Peoples Hospital, Meizhou 514031, China 2. Department of Imaging, Guangdong Second Provincial General Hospital, Guangzhou 510403, China
- Keywords:
- Keywords: glioma tumor recurrence magnetic resonance imaging radiomics multi-modal
- PACS:
- R816.1
- DOI:
- DOI:10.3969/j.issn.1005-202X.2025.02.010
- Abstract:
-
Abstract: Objective To explore the preoperative non-invasive prediction of early postoperative glioma recurrence using multi-modal conventional MRI radiomics. Methods A retrospective analysis of the clinical and MRI data of 83 glioma patients who met the inclusion criteria was conducted. The Kruskal-Wallis test was used to compare clinical factors between recurrent and non-recurrent groups. The automated segmentation of the entire tumor lesion for glioma patients was accomplished with VB-Net algorithm, a deep learning approach developed by United Imaging Healthcare and the extraction of radiomics features from preoperative T1CE and T2WI images was carried out on URP platform. The optimal feature combination was determined using the maximum relevance and minimum redundancy and least absolute shrinkage and selection operator methods. Logistic regression and five-fold cross-validation were employed to analyze radiomics features and construct 4 prediction models, namely T2WI model, T1CE model, T2WI+T1CE model, and imaging-clinical fusion model. The diagnostic performances of these models were evaluated and compared using the area under the receiver operating characteristic curve (AUC) and the Delong test. In addition, the model sensitivity and specificity were calculated. Results Postoperatively, there were 40 recurrent cases and 43 non-recurrent cases. The clinical factors such as glioma grade showed statistical significance between two groups (P<0.05), while gender and age did not show significant statistical differences (P>0.05). For the single-sequence radiomics models, T1CE model (AUC: 0.804) outperformed T2WI model (AUC: 0.702). The multi-modal combined model exhibited a higher AUC than the single-sequence prediction models, with an AUC of 0.849, a sensitivity of 72.5%, and a specificity of 79.1%. The imaging-clinical fusion model whose predictive efficiency was close to that of multi-modal combined model (P=0.303) also performed well in forecasting postoperative glioma recurrence, with an AUC of 0.839, a sensitivity of 72.5%, and a specificity of 79.1%. Conclusion The multi-modal conventional MRI radiomics model can better predict early postoperative glioma recurrence. The imaging-clinical fusion model that includes glioma grade does not have the diagnostic performance superior to that of radiomics model.
Last Update: 2025-01-22