Related Experiment Video
Updated: Sep 20, 2026
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Spatial association between pontine infarction and the corticospinal tract predicts early neurological deterioration:
Yang Du1, Bi Ma2, Yinglin Liu3
1Department of Neurology, West China School of Medicine, Sichuan University, Sichuan University Affiliated Chengdu Second People's Hospital, Chengdu, Sichuan, China; School of Clinical Medicine, Chengdu Medical College, Chengdu, Sichuan, China.
Objectives:
Branch atheromatous disease (BAD) involving the paramedian pontine artery (PPA) carries a high risk of early neurological deterioration (END), largely due to corticospinal tract (CST) injury. This study aimed to quantify CST impairment via an atlas-based registration approach and develop machine learning (ML) models to predict END.
Materials And Methods:
A total of 221 patients diagnosed with acute PPA infarction were retrospectively enrolled from two campuses. Diffusion-weighted imaging (DWI) was registered to the JHU white-matter atlas to delineate CST involvement. Shape-based radiomics features were extracted from infarct and CST regions, including CST-to-lesion ratio features. Following feature selection, six radiomics features were retained. A clinical logistic regression model was established using significant baseline characteristics. The Synthetic Minority Over-sampling Technique (SMOTE) was applied to balance the training cohort. Radiomics and clinical-radiomics models were developed using support vector machine (SVM), random forest (RF), and XGBoost algorithms.
Results:
The study* included 152 patients in the training cohort and 69 in the validation cohort. END occurred in 28.3% of the training cohort and 24.6% of the validation cohort. In the validation cohort, the radiomics SVM, RF, and XGBoost yielded AUC values of 0.910 (95% CI: 0.831-0.968), 0.910 (95% CI: 0.833-0.968), and 0.855 (95% CI: 0.752-0.938), respectively, all significantly outperforming the clinical model. Incorporating clinical characteristics into the radiomics models did not result in additional predictive benefit.
Conclusion:
Atlas-based CST radiomics combined with ML show promise for predicting END in PPA-BAD. These preliminary findings warrant further validation in large-scale, multicenter cohorts. *This study was approved by the Ethics Committee of Chengdu Second People's Hospital (Approval No. 2022046).
