학술논문
Comparison of Alzheimer's dementia and healthy classification algorithms based on MRI data.
Document Type
Abstract
Source
Subject
*ALZHEIMER'S disease
*NAIVE Bayes classification
*CLASSIFICATION algorithms
*COMPUTER-aided diagnosis
*SUPPORT vector machines
*NEUROLOGICAL disorders
*FORENSIC psychiatry
*
*
*
*
*
*
Language
ISSN
1307-8798
Abstract
Objective: Dementia is a neurological condition including cognitive impairment, psychiatric and behavioral symptoms. Alzheimer's disease (AD) is the most common cause of dementia. Current diagnosis of the disease bases on neuropsychological test (NPT) scores and detailed history taken from the patient and his relatives. However, an approach solely based on empirically specified cut-off points for NPT scores is not sufficiently objective. For this reason, more objective diagnostic methods with help of computer-aided diagnosis systems are being studied. We examined the performance of different machine classification methods on T1-weighted structural magnetic resonance imaging (MRI) data of demented AD patients (ADD) and healthy individuals. Methods: MRI and clinical data were taken from the OASIS-2 (https://www.oasis-brains.org) database. T1 images of 30 ADD patients and 34 healthy control subjects aged between 60-96 were included in the analysis. Total gray matter, white matter, cerebrospinal fluid (CSF) and intracranial volumes of individuals were computed by using CAT 12 (Computational Anatomy Toolbox, https://neuro-jena.github.io/cat/) software. These data and the gender, age, total years of education and mini-mental state examination (MMSE) scores of the individuals were supplied to the classification algorithms as feature set. Naive Bayes, support vector machine and multilayer perceptron classifiers included in WEKA (https://www.cs.waikato.ac.nz/~ml/weka/) software package were tested. Total data set is divided into 70% training and 30% test subsets. Results: The success rates of the algorithms in the test data sets were found to be 94.73% for Naive Bayes, 84.21% for the support vector machine and 84.21% for the multilayer perceptron, respectively. Conclusion: Although acceptable success was achieved with all 3 machine learning methods applied on the data set including the total gray and white matter, CSF and intracranial volumes, the highest success rate was achieved with Naive Bayes Classifier. [ABSTRACT FROM AUTHOR]