Premium
A cost‐efficient model for predicting cerebral Aβ burden using MRI and neuropsychological markers in the ADNI‐2 cohort
Author(s) -
Ko Hyunwoong,
Ihm Jungjoon
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.041715
Subject(s) - predictability , cohort , neuropsychology , neuroimaging , medicine , lasso (programming language) , demographics , machine learning , computer science , artificial intelligence , psychology , statistics , mathematics , cognition , demography , psychiatry , sociology , world wide web
Background Detecting cerebral Aβ is still expensive, invasive, and have limited accessibility. In this respect, the current study aims to identify and compare predictability of MRI markers with neuropsychological markers for accurate prediction of cerebral Aβ status in AD cohort through machine learning (ML) approaches. Methods Predictability of the candidate markers for cerebral Aβ status was examined by analyzing 724 participants from the ADNI‐2 cohort at baseline visit (170 control subjects, 95 with SMC, 324 with MCI and 135 with AD; mean age 73.2 years, range 55–90). Demographic variables (age, gender, education, and APOE status), structural MRI markers (cortical thickness and volume), and neuropsychological test scores were used as input in several ML algorithms. Cerebral Aβ burden was measured using florbetapir PET images. We first calculated the predictability of each ML model, and the adaptive LASSO algorithm with 10‐fold cross validation was implemented to identify the relative predictability of predictors in the selected model. Results ML models with MRI markers predicted cerebral Aβ status with the 85% predictability. Out of five combination of candidate markers, neuropsychological markers with demographics showed the most cost‐efficient result compared to MRI methods. The adaptive LASSO model with out‐of‐sample classification was able to distinguish abnormal levels of Aβ based on significantly predictable features. The AUC was 0.85 in the ADNI‐2 cohort, indicating the same performance with MRI‐based models. Conclusions Our results are twofold: the result has first identified the predictability in MRI markers using ML approaches, and secondly demonstrated the neuropsychological model with demographics could predict Aβ positivity, suggesting a more cost‐efficient method for detecting cerebral Aβ status compared to MRI markers. More specifically, with advantages in cost and its non‐invasive procedure, the method could be utilized as a brief screening tool in AD risk population for clinical trials or AD therapy. To this end, This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC(Information Technology Research Center) support program(IITP‐2020‐2017‐0‐01630) supervised by the IITP(Institute for Information & communications Technology Promotion).