EEG Classification by Minimalistic Convolutional Neural Network Utilizing Context Information

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21460%2F21%3A00353219" target="_blank" >RIV/68407700:21460/21:00353219 - isvavai.cz</a>

Výsledek na webu

<a href="https://doi.org/10.1109/EHB52898.2021.9657572" target="_blank" >https://doi.org/10.1109/EHB52898.2021.9657572</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/EHB52898.2021.9657572" target="_blank" >10.1109/EHB52898.2021.9657572</a>

Jazyk výsledku

angličtina

Název v původním jazyce

EEG Classification by Minimalistic Convolutional Neural Network Utilizing Context Information

Popis výsledku v původním jazyce

Labelling of electroencephalography (EEG) or polysomnography (PSG) recordings can be time consuming for a physician (expert), especially if we consider long term monitoring (e.g. sleep stages). Convolutional neural networks (CNN), which use images created from raw or preprocessed data trials, appear to be a promising method for semi-automated label acquisition. To the best of our knowledge most of the existing studies use only one channel or a very small number of channels in combination with a complicated structure of the CNN. In our study we propose adding context information based on use of multiple channels in an input image. Based on our current understanding and state of the art we expect additional growth of final accuracy. Studies based on this approach show additional input from context information that enhances final validation accuracy. During our research we focused on small structure CNN with only three convolutional layers and two fully connected layers. Additionally, all methods which use CNN structure are heavily influenced by several parameters like number of classes or size of their convolution masks. Secondary goal of this study is to acquaint the reader with these parameters and with possible solutions for their problems. Based on more than six thousand different CNN settings and input settings we managed to acquire results up to 94.6% mean training accuracy and 88.2% mean validation accuracy in sleep stage classification on real PSG including sleep EEG recordings even with easy-to-create CNN structure.

Název v anglickém jazyce

EEG Classification by Minimalistic Convolutional Neural Network Utilizing Context Information

Popis výsledku anglicky

Labelling of electroencephalography (EEG) or polysomnography (PSG) recordings can be time consuming for a physician (expert), especially if we consider long term monitoring (e.g. sleep stages). Convolutional neural networks (CNN), which use images created from raw or preprocessed data trials, appear to be a promising method for semi-automated label acquisition. To the best of our knowledge most of the existing studies use only one channel or a very small number of channels in combination with a complicated structure of the CNN. In our study we propose adding context information based on use of multiple channels in an input image. Based on our current understanding and state of the art we expect additional growth of final accuracy. Studies based on this approach show additional input from context information that enhances final validation accuracy. During our research we focused on small structure CNN with only three convolutional layers and two fully connected layers. Additionally, all methods which use CNN structure are heavily influenced by several parameters like number of classes or size of their convolution masks. Secondary goal of this study is to acquaint the reader with these parameters and with possible solutions for their problems. Based on more than six thousand different CNN settings and input settings we managed to acquire results up to 94.6% mean training accuracy and 88.2% mean validation accuracy in sleep stage classification on real PSG including sleep EEG recordings even with easy-to-create CNN structure.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20601 - Medical engineering

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2021

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

2021 International Conference on e-Health and Bioengineering (EHB)

ISBN

978-1-6654-4000-4

ISSN

2575-5137

e-ISSN

2575-5145

Počet stran výsledku

4

Strana od-do

1-4

Název nakladatele

IEEE Industrial Electronic Society

Místo vydání

Vienna

Místo konání akce

Iasi

Datum konání akce

18. 11. 2021

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

000802227900036