Objective: This study reports recent updates in the development of a support vector machine learning model to discriminate between dementia variants (i.e., Alzheimer’s disease (AD) and dementia with Lewy bodies (DLB)) and healthy age-matched controls (HC) using diffusion MRI.
Background: Freewater imaging is a noninvasive diffusion imaging metric that is associated with neurodegeneration and neuroinflammation. Freewater imaging has been used to distinguish between Parkinsonian disorders and shows promise in the domain of dementias. Like Parkinsonian disorders, dementias are difficult to differentiate in early stages of disease. AD and DLB are two of the most common types of dementia, with DLB often misdiagnosed as AD.
Method: Diffusion MRI scans were collected from three different consortiums and preprocessed for eddy current corrections, freewater estimation, freewater elimination, and normalization. The primary outcomes of preprocessing are freewater and freewater-corrected fractional anisotropy values across the 233 regions of interest. The diffusion metrics are used as inputs to the support vector machine. The three comparisons evaluated in the studies are: AD vs DLB, AD vs HC, and DLB vs HC. We developed a training and validation cohort, and testing cohorts for each comparison using an 80:20 split. The training and validation cohorts are used to train the support vector machine with a 5-fold cross validation. Following training, we tested each comparison using area under the curve (AUC) of the receiver operator characteristic curve as a metric of diagnostic selectivity.
Results: The final sample consisted of 610 participants across three disease states from three consortia. The training and validation AUCs were 0.945, 0.978, and 0.962 and test AUCs were 0.959, 0.958, and 0.930 for AD vs DLB, DLB vs HC, and AD vs HC, respectively. Further, we looked at one-versus-all classifications. The training and validation AUCs were 0.977, 0.993, and 0.947 and test AUCs were 0.973, 0.919, and 0.947 for AD vs DLB/HC, DLB vs AD/HC, and HC vs AD/DLB, respectively.
Conclusion: This study indicates that freewater imaging, a contrast free approach, can aid in the differentiation of AD and DLB using an automated image analysis pipeline.
To cite this abstract in AMA style:
R. Chen, W. Wang, A. Barmpoutis, D. Vaillancourt. Diffusion MRI and Machine Learning Distinguish Alzheimer’s Disease and Dementia with Lewy Bodies [abstract]. Mov Disord. 2024; 39 (suppl 1). https://www.mdsabstracts.org/abstract/diffusion-mri-and-machine-learning-distinguish-alzheimers-disease-and-dementia-with-lewy-bodies/. Accessed October 10, 2024.« Back to 2024 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/diffusion-mri-and-machine-learning-distinguish-alzheimers-disease-and-dementia-with-lewy-bodies/