Chi-Separation MRI Plus Machine Learning Detects Brain Iron Linked to Cognition in Beta-Thalassemia

In 50 β-thalassemia major patients vs 50 matched controls, sub-voxel chi-separation MRI revealed excess iron in hippocampus, insula & anterior cingulate (FDR q<0.05); an SVM classifier achieved mean AUC 0.919±0.054. Higher iron correlated with lower MoCA scores — retrospective,…

• Cross-sectional study; 50 β-thalassemia major patients and 50 age/sex-matched healthy controls underwent 3T multi-echo gradient-echo MRI with chi-separation to isolate paramagnetic (iron-related) susceptibility in anatomically defined ROIs.
• Among three classifiers, SVM (RBF kernel) outperformed random forest and XGBoost (mean AUC 0.919±0.054); SHAP analysis flagged hippocampal and insular susceptibility as the top discriminating features, both inversely associated with Montreal Cognitive Assessment scores.
• Key limitation: small single-cohort sample (n=100) with internal Monte Carlo cross-validation only — no external test set; generalizability to milder thalassemia phenotypes or other iron-overload disorders is unestablished.

RadPigeon summaries are original and for information only. They are not clinical advice.