Premium
P‐040: Database supported diagnostics of Alzheimer's disease
Author(s) -
Johannesson Gisli H.,
Johnsen Kristinn,
Gudmundsson Steinn,
Emilsdottir Asdis L.,
Blin Nicolas P.,
Helgadottir Halla,
Snaedal Jon,
Gudmundsson Thorkell E.
Publication year - 2007
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 - 10.1016/j.jalz.2007.04.256
Subject(s) - electroencephalography , dementia , receiver operating characteristic , feature (linguistics) , disease , artificial intelligence , alzheimer's disease , pattern recognition (psychology) , audiology , medicine , computer science , psychology , machine learning , pathology , neuroscience , linguistics , philosophy
Background: The possibility of using electroencephalograms (EEG) as a surrogate marker for Alzheimer’s disease (AD) has been indicated by various investigations. Although the EEG of AD patients on average differs from healthy individuals for a given feature of the EEG, using a single feature is not clinically useful since the overlap between the distributions of the feature between the groups is too great resulting in poor accuracy. Objective: In this project a database of EEGs and multiple EEG features are used for diagnostic purposes for AD. The main goal is to create a diagnostic tool for AD which is both sensitive and specific. A secondary objective is to investigate whether AD patients can be separated from patients suffering from Vascular Dementia (VD) using the same methodology. Methods: A database of 1000 individual EEG measurements is being constructed. The participant has eyes closed and is at rest and the EEG is measured for a few minutes. Currently 500 EEGs have been collected. The results of EEG measurements from AD patients, VD patients, and age matched controls are presented here. Statistical pattern recognition (SPR) is applied to the dataset in order to select the combination of EEG features which best separates the groups of AD patients and healthy individuals and the groups of AD patients and VD patients. Results: The AD group and the control group can be separated with 90% accuracy. And the area under the receiver operator curve (ROC) is 0.96 demonstrating the applicability of the method. Conclusions: The results indicate that this type of methodology is both accurate and specific and can be used for a detection of AD. It may therefore be a useful addition to traditional AD diagnostic procedure, in particular since the method is completely objective.