2022
DOI: 10.1109/tnsre.2022.3204540
|View full text |Cite
|
Sign up to set email alerts
|

TIE-EEGNet: Temporal Information Enhanced EEGNet for Seizure Subtype Classification

Abstract: labelers. To relieve the burden on clinicians and improve the treatment efficiency, automated epilepsy diagnosis algorithms are desired.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
23
0

Year Published

2023
2023
2025
2025

Publication Types

Select...
5
2

Relationship

1
6

Authors

Journals

citations
Cited by 30 publications
(23 citation statements)
references
References 52 publications
(60 reference statements)
0
23
0
Order By: Relevance
“…The performance of the proposed EEG-LSTMNet is compared to that of the state-of-theart model described in [25], both of which classify eight seizure types. The proposed model in [25] comprises three blocks, block one with temporal and depthwise convolutional layers, block two with a separable, depthwise convolutional layer and a subsequent point-wise convolutional layer, and block three with a single fully connected layer that is connected to a SoftMax activation function. The proposed EEG-LSTMNet outperforms this model in adopting two average pooling layers to compute the average value for the patches of a feature map to create a down-sampled (pooled) feature map.…”
Section: Comparison With State-of-the-art Methodsmentioning
confidence: 99%
See 2 more Smart Citations
“…The performance of the proposed EEG-LSTMNet is compared to that of the state-of-theart model described in [25], both of which classify eight seizure types. The proposed model in [25] comprises three blocks, block one with temporal and depthwise convolutional layers, block two with a separable, depthwise convolutional layer and a subsequent point-wise convolutional layer, and block three with a single fully connected layer that is connected to a SoftMax activation function. The proposed EEG-LSTMNet outperforms this model in adopting two average pooling layers to compute the average value for the patches of a feature map to create a down-sampled (pooled) feature map.…”
Section: Comparison With State-of-the-art Methodsmentioning
confidence: 99%
“…Unlike the single fully connected classification layer in [25], the proposed EEG-LSTMNet adopts two fully connected layers; the first layer is a time-distributed dense layer, while the second one is connected to a SoftMax function and includes eight neurons. Both models adopt different databases; the suggested EEG-LSTMNet model adopts the TUH database, whereas [25] adopts both the TUSZ and CHSZ databases. The results of both models prove that our model outperforms with a 98.4% F1-score compared with the state-of-the-art model in [25], which achieved 67.2% Accuracy.…”
Section: Comparison With State-of-the-art Methodsmentioning
confidence: 99%
See 1 more Smart Citation
“…To alleviate the above problem, TL has been proposed to leverage data and knowledge from the source domains and handle the domain discrepancy. Unsupervised domain adaptation assumes the source domains are labeled, whereas the target domain is not, which is more suitable for new patient diagnostics [17]. Unsupervised domain adaptation approaches utilize the unlabeled target domain data during learning to better calibrate the algorithm under distribution shift.…”
Section: Tlmentioning
confidence: 99%
“…The public Helsinki University Hospital neonatal intensive care unit (NICU) dataset [36] and the CHSZ dataset [17] were used. Table 1 summarizes their main characteristics.…”
Section: Datasetsmentioning
confidence: 99%