Vše

Co hledáte?

Vše
Projekty
Výsledky výzkumu
Subjekty

Rychlé hledání

  • Projekty podpořené TA ČR
  • Významné projekty
  • Projekty s nejvyšší státní podporou
  • Aktuálně běžící projekty

Chytré vyhledávání

  • Takto najdu konkrétní +slovo
  • Takto z výsledků -slovo zcela vynechám
  • “Takto můžu najít celou frázi”

Semi-supervised classification of iEEG using temporal autoencoder neural network

Identifikátory výsledku

  • Kód výsledku v IS VaVaI

    <a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21460%2F22%3A00361897" target="_blank" >RIV/68407700:21460/22:00361897 - isvavai.cz</a>

  • Nalezeny alternativní kódy

    RIV/68407700:21730/22:00361897

  • Výsledek na webu

    <a href="https://doi.org/10.1111/epi.17388" target="_blank" >https://doi.org/10.1111/epi.17388</a>

  • DOI - Digital Object Identifier

    <a href="http://dx.doi.org/10.1111/epi.17388" target="_blank" >10.1111/epi.17388</a>

Alternativní jazyky

  • Jazyk výsledku

    angličtina

  • Název v původním jazyce

    Semi-supervised classification of iEEG using temporal autoencoder neural network

  • Popis výsledku v původním jazyce

    Purpose: The visual review and interpretation of electrophysiology recordings is a time-consuming and challenging task subject to human operator variability and bias. This challenge has only grown with the emergence of large-scale electrophysiology recordings with high channel counts spanning long time scales. Methods: This research presents a semi-supervised deep learning method that utilizes temporal autoencoders to classify iEEG data into four distinct groups (i.e., physiological, pathological, movement & muscle artifacts, power line noise). The model was trained on publicly available datasets consisting of 3-second iEEG clips originating from 14 and 25 patients with drug-resistant epilepsy from two hospitals. Results: Each dataset consists of at least 150 thousand annotated clips. In general, the semisupervised technique utilizes a small amount of gold standard labeled data and large numbers of unlabelled data to capture underlying patterns in the data. The temporal autoencoders projects input iEEG data into low-dimensional embedding space, where data can be efficiently clustered or classified. The classification method utilizes kernel density estimation (KDE) and a naive Bayesian classifier, which process low dimensional embeddings from temporal autoencoder. We believe that this method might be efficiently used to preprocess long-term recordings, where visual inspection of the small amount of data might be easily extrapolated onto the whole recording, which significantly speeds up the iEEG revision. Once the entire recording is analyzed with the autoencoder, the signals segments of interest might be selected and presented to the physicians for further analysis and quantification. Conclusion: Our findings principally support the utility of the semi-supervised method for the training of classifiers on large EEG datasets, where only a minor portion of the data has standard gold labels. Numerical results show that the proposed semi-supervised method archives classification F1-score of 0.7 on hidden testing set while using only one thousand gold standard samples from each classification category.

  • Název v anglickém jazyce

    Semi-supervised classification of iEEG using temporal autoencoder neural network

  • Popis výsledku anglicky

    Purpose: The visual review and interpretation of electrophysiology recordings is a time-consuming and challenging task subject to human operator variability and bias. This challenge has only grown with the emergence of large-scale electrophysiology recordings with high channel counts spanning long time scales. Methods: This research presents a semi-supervised deep learning method that utilizes temporal autoencoders to classify iEEG data into four distinct groups (i.e., physiological, pathological, movement & muscle artifacts, power line noise). The model was trained on publicly available datasets consisting of 3-second iEEG clips originating from 14 and 25 patients with drug-resistant epilepsy from two hospitals. Results: Each dataset consists of at least 150 thousand annotated clips. In general, the semisupervised technique utilizes a small amount of gold standard labeled data and large numbers of unlabelled data to capture underlying patterns in the data. The temporal autoencoders projects input iEEG data into low-dimensional embedding space, where data can be efficiently clustered or classified. The classification method utilizes kernel density estimation (KDE) and a naive Bayesian classifier, which process low dimensional embeddings from temporal autoencoder. We believe that this method might be efficiently used to preprocess long-term recordings, where visual inspection of the small amount of data might be easily extrapolated onto the whole recording, which significantly speeds up the iEEG revision. Once the entire recording is analyzed with the autoencoder, the signals segments of interest might be selected and presented to the physicians for further analysis and quantification. Conclusion: Our findings principally support the utility of the semi-supervised method for the training of classifiers on large EEG datasets, where only a minor portion of the data has standard gold labels. Numerical results show that the proposed semi-supervised method archives classification F1-score of 0.7 on hidden testing set while using only one thousand gold standard samples from each classification category.

Klasifikace

  • Druh

    O - Ostatní výsledky

  • CEP obor

  • OECD FORD obor

    20601 - Medical engineering

Návaznosti výsledku

  • Projekt

  • Návaznosti

    S - Specificky vyzkum na vysokych skolach

Ostatní

  • Rok uplatnění

    2022

  • Kód důvěrnosti údajů

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