Cognitive Function Assessment and Prediction for Subjective Cognitive Decline and Mild Cognitive Impairment

• Published in: Brain imaging and behavior 2022-04, Vol.16 (2), p.645-658
• Main Authors: Li, Aojie, Yue, Ling, Xiao, Shifu, Liu, Manhua
• Format: Article
• Language: English
• Subjects: Alzheimer Disease - pathology; Alzheimer's disease; Amygdala; Atrophy; Biomarkers; Biomedical and Life Sciences; Biomedicine; Brain; Coding; Cognition; Cognitive ability; Cognitive Dysfunction - pathology; Dementia disorders; Fornix; Hippocampus; Humans; Impairment; Machine learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Medical imaging; Neural coding; Neurodegenerative diseases; Neuroimaging; Neuropsychology; Neuroradiology; Neurosciences; Original Research; Predictions; Psychiatry; Regression models; Temporal lobe
• ISSN: 1931-7557; 1931-7565
• DOI: 10.1007/s11682-021-00545-1

Alzheimer’s disease (AD) is a progressive and irreversible neurodegenerative dementia. Recent studies found that subjective cognitive decline (SCD) may be the early clinical precursor that precedes mild cognitive impairment (MCI) for AD. SCD subjects with normal cognition may already have some medial temporal lobe atrophy. Although brain changes by AD have been widely studied in the literature, it is still challenging to investigate the anatomical subtle changes in SCD. This paper proposes a machine learning framework by combination of sparse coding and random forest (RF) to identify the informative imaging biomarkers for assessment and prediction of cognitive functions and their changes in individuals with MCI, SCD and normal control (NC) using magnetic resonance imaging (MRI). First, we compute the volumes from both the regions of interest from whole brain and the subregions of hippocampus and amygdala as the features of structural MRIs. Then, sparse coding is applied to identify the relevant features. Finally, the proximity-based RF is used to combine three sets of volumetric features and establish a regression model for predicting clinical scores. Our method has double feature selections to better explore the relevant features for prediction and is evaluated with the T1-weighted structural MR images from 36 MCI, 112 SCD, 78 NC subjects. The results demonstrate the effectiveness of proposed method. In addition to hippocampus and amygdala, we also found that the fimbria, basal nucleus and cortical nucleus subregions are more important than other regions for prediction of Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) scores and their changes.