Characterizing MRI biomarkers for early prediction of amnestic mild cognitive impairment among the community‐dwelling Chinese: Neuroimaging / Optimal neuroimaging measures for early detection. (7th December 2020)

• Record Type: Journal Article
• Authors: Yue, Ling; Pan, Yongsheng; Wang, Tao; Liu, Mingxia; Shen, Dinggang; Xiao, Shifu
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.041450
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Identifying subjects at the preclinical amnestic mild cognitive impairment (pre‐MCI), is fundamental for early intervention of Alzheimer's disease. Extensive studies have shown that structural magnetic resonance imaging (MRI) provides objective biomarkers for automated diagnosis of Alzheimer's disease and its prodromal stage (i.e., aMCI). This study aims to investigate a deep‐learning‐based framework for predicting the progression from normal control (NC) to MCI, by characterizing imaging biomarkers in structural MRI data. Method: Motivated by previous disease diagnosis systems with MRI, we create a unique deep‐learning framework for pre‐MCI conversion prediction, by evaluating the discriminative capability of single region‐of‐interest (ROI) or multiple ROIs in the brain. Our framework includes 2 steps: 1) a single‐ROI‐based network (SRNet) for locating the informative regions in the brain, and 2) a multi‐ROI‐based network (MRNet) for pre‐MCI conversion prediction (see Fig 1). We use 845 subjects with baseline MRI in ADNI‐1 to train SRNet and MRNet. To balance the sample size in each category, we use both AD and progressive MCI (within 3 years) patients as positive samples, while NC and stable MCI (within 3 years) samples as negative samples. We then apply the model trained on ADNI‐1 to our private dataset from the Chinese Longitudinal Aging Study (CLAS). The dataset of CLAS included 109 subjects with a baseline NC diagnosis and a 7‐year follow‐up diagnosis. Among these subjects, 40 NC are absolutely converted to aMCI within 7 years (pNC) and 69 NC remain cognitively normal (sNC). Result: We illustrate the predicted scores of all testing subjects derived by our method, where the subjects are roughly sorted by their stages. We further compared these two groups by logistic analysis while controlled for age, gender, education level, depression symptoms, anxiety symptom, physical disease and baseline cognitive score, pNC individuals remained significant higher predicted score than sNC (see table 1). Conclusion: This study develops an ROI‐based deep‐learning framework to predict the progression from cognitive normal to aMCI. Experimental results suggest that our method can discriminate the cognitive normal individuals who would be diagnosed with MCI 7 years later, which is helpful for early diagnosis.