Premium
Machine learning regression analysis predicts brain amyloid‐β burden: A multimodal neuroimaging research study
Author(s) -
Lin Li,
Wang Xiaoni,
Wang Xiaoqi,
Li Xuanyu,
Sun Yu,
Du Wenying,
Chen Guanqun,
Han Ying
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.041799
Subject(s) - parahippocampal gyrus , fractional anisotropy , neuroimaging , voxel , diffusion mri , amyloid (mycology) , medicine , magnetic resonance imaging , neuroscience , psychology , pathology , radiology , temporal lobe , epilepsy
Abstract Background Recent criteria have unveiled the importance of biomarkers in the continuum of Alzheimer's disease, which greatly enhances the progression of clinical trials of prevention and treatment. However, the expanding cost and resources limit the first‐line brain amyloid measurement techniques. Objective We aimed to develop an inexpensive and sensitive method to predict brain amyloid‐β burden by integrating machine learning and multimodal MRI. Methods Eighty‐three elders were enrolled from the Xuanwu Hospital. Brain amyloid‐β burden, structural and functional changes were respectively assessed with (18)F‐AV45 amyloid PET, structural and diffusion MRI, functional MRI. Using regression machine implementations, we predicted brain amyloid‐β level by voxel‐based morphometry, fractional anisotropy and fractional amplitude of low frequency fluctuations (fALFF). Results We found that extracted features could predict brain amyloid‐β burden, using voxel‐based morphometry (r=0.55, p=0.001) or/and fractional anisotropy (r=0.36, p=0.002), but not fALFF(r=0.07, p=0.2388). These results were replicated using cross‐validation procedures and these models were statistical significance determined by permutation testing. The brain regions that showed most important to predict amyloid‐β burden were superior temporal gyrus, parahippocampal gyrus, inferior frontal gyrus and hippocampus. Conclusion The current study showed high correlation between the structural and diffusion MRI indexes and amyloid burden in the elders which opens a new possibility for early diagnosis of AD.