Machine learning-based radiomics modeling of combined subcortical nuclei from quantitative susceptibility mapping for Parkinson's disease diagnosis.

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

BackgroundQuantitative susceptibility mapping (QSM) enables non-invasive assessment of brain iron deposition in Parkinson's disease (PD), yet existing radiomics studies have predominantly focused on single subcortical nuclei, without systematically evaluating whether combining multiple regions improves diagnostic accuracy.MethodsA total of 59 PD patients and 73 healthy controls underwent QSM. Radiomic features were extracted from the substantia nigra (SN), caudate nucleus (CN), globus pallidus (GP), red nucleus (RN), and putamen (Put). For all 31 possible combinations of the five nuclei, feature selection was performed on the training set using Wilcoxon rank-sum tests, minimal redundancy maximal relevance, and least absolute shrinkage and selection operator regression. Four classifiers-logistic regression (LR), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost)-were trained and evaluated under a stratified five-fold nested cross-validation framework. Diagnostic performance was assessed primarily using the area under the receiver operating characteristic curve (AUC).ResultsLR achieved its best mean test AUC of 0.879 ± 0.045 with CN + GP. SVM and RF yielded optimal mean test AUCs of 0.880 ± 0.047 (GP + RN + Put) and 0.891 ± 0.028 (CN + GP), respectively. XGBoost with GP + RN + Put produced the highest mean test AUC among all models (0.921 ± 0.040) and was designated the global best model; bootstrap validation confirmed its robustness (AUC = 0.940, 95% confidence interval: 0.891-0.977). A significant positive correlation was observed between the number of combined nuclei and XGBoost test AUC (Spearman's rho = 0.614, P ConclusionA multi-nuclei QSM-based radiomics model integrating GP, RN, and Put with XGBoost achieves excellent diagnostic performance for PD, supporting the value of multi-regional information fusion in precision imaging-based diagnosis.

Related topics