Prognostic prediction of patients in vegetative state based on quantitative analysis of diffusion tensor imaging(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2025年第9期
- Page:
- 1147-1152
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Prognostic prediction of patients in vegetative state based on quantitative analysis of diffusion tensor imaging
- Author(s):
- YE Simin1; ZHONG Haili2; LIANG Qimei3; HUANG Xiyan2; WANG Sixun1; HUANG Jing1
- 1. School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China 2. Department of Rehabilitation Medicine, Zhujiang Hospital of Southern Medical University, Guangzhou 510282, China 3. Department of Rehabilitation Medicine, the Tenth Affiliated Hospital of Southern Medical University (Dongguan Peoples Hospital), Dongguan 523000, China
- Keywords:
- Keywords: vegetative state fractional anisotropy support vector machine diffusion tensor imaging
- PACS:
- R318；R445.2
- DOI:
- DOI:10.3969/j.issn.1005-202X.2025.09.004
- Abstract:
- Abstract: Objective To analyze the differences in structural integrity of cerebral white matter fiber bundles in vegetative state (VS) patients with different prognoses, and to construct an early-stage prognostic prediction model for 1-year post-stabilization prognosis. Methods A retrospective analysis was conducted on 52 VS patients admitted to the Department of Rehabilitation Medicine at Zhujiang Hospital of Southern Medical University. Patients were stratified into good prognosis (n=22) and poor prognosis (n=30) at 1-year follow-up based on Coma Recovery Scale-Revised (CRS-R) scores. The fractional anisotropy values of cerebral white matter fiber bundles were derived from diffusion tensor imaging, and for the first time, the scores of the visual subscales of CRS-R were combined with FA values as input features for the prognostic model. To optimize model construction, the least absolute shrinkage and selection operator regression was employed for feature screening, and synthetic minority over-sampling technique for data balancing. The prognostic prediction model was subsequently developed using a ?upport vector machine algorithm and validated via ?eave-one-out cross-validation. Model performance was evaluated using area under receiver operating characteristic curve, along with accuracy, sensitivity, specificity, and F1 score metrics. Results Following LASSO regression feature screening, the pontine crossing tract, medial lemniscus, tapetum, splenium of corpus callosum, and visual subscale scores were identified as key predictors. A multimodal SVM-based prediction model constructed with the above features could effectively predict the 1-year prognosis of VS patients, achieving a high predictive performance (AUC=0.894). Conclusion The SVM-based model integrating FA values of specific white matter fiber bundles and visual subscale scores demonstrates excellent predictive performance in predicting the 1-year prognosis of VS patients.
Last Update: 2025-09-30