Lifelog-based classification of mild cognitive impairment using artificial neural networks

Abstract

The swift diagnosis and treatment of mild cognitive impairment (MCI), as a prestage of dementia, are important to reduce the enormous costs of dementia treatment. The aim of this paper is to investigate the potential features in human behavior to facilitate the early diagnosis of MCI. In order to extract specific features from lifelogs, we collected data of activity and sleep using Fitbit's wrist band worn day and night from 12 subjects, for 12 week each. These data were analyzed using the SPSS (Statistical Package for Social Science) for verification and 12 total numbers of the significant features are extracted, further these features used for classification based on artificial neural networks (ANNs). ANNs with 8 input neurons (extracted features), 4 hidden neurons, and output neurons (diagnosis) were used to classify the patients. The results indicate that lifelog-based classifier have a good capacity (AUC=0.81 ±0.08) to discriminate MCI patients from healthy controls.