Category: MSA, PSP, CBS: Neuroimaging
Objective: This study aims to evaluate whether automated classification based on individual voxel-based morphometry (VBM) adjusting covariates (iVAC) can achieve high diagnostic accuracy for MSA across heterogeneous imaging environments using a multicenter database.
Background: We previously reported in a single-center study that iVAC may achieve high diagnostic accuracy for MSA; however, its multicenter generalizability remains unclear.
Method: Clinical and MRI data from 937 participants collected at 8 institutions (246 MSA, 252 Parkinson’s disease [PD], 90 progressive supranuclear palsy [PSP], 346 healthy controls [HC]) were analyzed. 3D T1-weighted images were preprocessed using a standard VBM pipeline. Regression models including age, sex, and intracranial volume (ICV) were constructed using HC images to establish a referential database. The resulting statistical parameters were applied to patient images to generate adjusted voxel-level Z-score maps, enabling quantitative assessment of regional atrophy. To evaluate generalizability, leave-one-site-out cross-validation (LOSO-CV) was performed, and discriminative performance was assessed.
Results: Receiver operating characteristic (ROC) analyses using Z-scores of the putamen, pons, and middle cerebellar peduncle demonstrated that a mean AUC of 0.942 (range across iterations: 0.922–0.979) in held-out test data and 0.947 (range: 0.928–0.957) in the training data across LOSO-CV iterations. In machine learning approaches, gradient boosting achieved a mean AUC of 0.947, with a balanced accuracy of 0.852. Logistic regression yielded a mean AUC of 0.933 with a balanced accuracy of 0.868 in test data.
Conclusion: The iVAC-based automated classification demonstrated robust, scanner-agnostic performance across institutions and imaging platforms based on the largest dataset to date, supporting its real-world applicability as a clinical decision-support tool for MSA.
To cite this abstract in AMA style:
K. Kawabata, H. Takeshige-Amano, T. Hatano, T. Baba, A. Takeda, S. Kuwabara, A. Ogura, M. Katsuno, R. Hanajima, M. Matsushima, I. Yabe, M. Hatakeyama, O. Onodera, H. Watanabe. Multicenter Evaluation of Automated Structural MRI-based Diagnosis for Multiple System Atrophy [abstract]. Mov Disord. 2026; 41 (suppl 1). https://www.mdsabstracts.org/abstract/multicenter-evaluation-of-automated-structural-mri-based-diagnosis-for-multiple-system-atrophy/. Accessed October 1, 2026.« Back to 2026 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/multicenter-evaluation-of-automated-structural-mri-based-diagnosis-for-multiple-system-atrophy/