Distinguishing age-related cognitive decline from dementias: A study based on machine learning algorithms.

BACKGROUND AND AIM: This study aims to examine the distinguishability of age-related cognitive decline (ARCD) from dementias based on some neurocognitive tests using machine learning.

MATERIALS AND METHODS: 106 subjects were divided into four groups: ARCD (n=30), probable Alzheimer's disease (AD) (n=20), vascular dementia (VD) (n=21) and amnestic mild cognitive impairment (MCI) (n=35). The following tests were applied to all subjects: The Wechsler memory scale-revised, a clock-drawing, the dual similarities, interpretation of proverbs, word fluency, the Stroop, the Boston naming (BNT), the Benton face recognition, a copying-drawings and Öktem verbal memory processes (Ö-VMPT) tests. A multilayer perceptron, a support vector machine and a classification via regression with M5-model trees were employed for classification.

RESULTS: The pairwise classification results show that ARCD is completely separable from AD with a success rate of 100% and highly separable from MCI and VD with success rates of 95.4% and 86.30%, respectively. The neurocognitive tests with the higher merit values were Ö-VMPT recognition (ARCD vs. AD), Ö-VMPT total learning (ARCD vs. MCI) and semantic fluency, proverbs, Stroop interference and naming BNT (ARCD vs. VD).

CONCLUSION: The findings show that machine learning can be successfully utilized for distinguishing ARCD from dementias based on neurocognitive tests.