Hybrid neonatal EEG seizure detection algorithms achieve the benchmark of visual interpretation of the human expert

Stevenson, Nathan J.; Tapani, Karoliina; Vanhatalo, Sampsa

doi:10.1109/embc.2019.8857367

Cited by 8 publications

(8 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bag of features approaches to neonatal seizure detection are being superseded by deep neural networks [17]. We and others have shown that deep convolutional neural networks provide similar performance to the proposed 'bag of features' approach when applied to our training set [28,17] with deep convolutional networks providing more consistent performance over several data sets [17]. These methods will, nevertheless, face similar challenges such as lack of diversity within training data sets, subjectivity in the gold standard of human expert annotation (the training target) and performance assessment.…”

Section: Discussionmentioning

confidence: 83%

“…This was due to the relatively small size of our database and the expected difference in EEG quality between high density, short duration training data and low density, long duration validation data. Therefore we used a modified version of our previous system that was first proposed in [15,28]. Post-processing of the initial SDA output included eliminating outliers (we also eliminated EEG epochs detected as 'bad electrode' by the EEG monitor), applying a temporal moving average to each channel, taking the maximum value across channels and thresholding to form a binary (seizure/no seizure) decision.…”

Section: Sdamentioning

confidence: 99%

See 1 more Smart Citation

Validating an SVM-based neonatal seizure detection algorithm for generalizability, non-inferiority and clinical efficacy

Tapani¹,

Nevalainen²,

Vanhatalo³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Neonatal seizure detection algorithms (SDA) are approaching the benchmark of human expert annotation.Measures of algorithm generalizability and non-inferiority as well as measures of clinical efficacy are needed to assess the full scope of neonatal SDA performance. We validated our neonatal SDA on an independent data set of 28 neonates. Generalizability was tested by comparing the performance of the original training set (cross-validation) to its performance on the validation set. Non-inferiority was tested by assessing interobserver agreement between combinations of SDA and two human expert annotations. Clinical efficacy was tested by comparing how the SDA and human experts quantified seizure burden and identified clinically significant periods of seizure activity in the EEG. Algorithm performance was consistent between training and validation sets with no significant worsening in AUC (p > 0.05, n = 28). SDA output was inferior to the annotation of the human expert, however, re-training with an increased diversity of data resulted in non-inferior performance (∆κ=0.077, 95% CI: -0.002-0.232, n = 18). The SDA assessment of seizure burden had an accuracy ranging from 89-93%, and 87% for identifying periods of clinical interest. The proposed SDA is approaching human equivalence and provides a clinically relevant interpretation of the EEG.

show abstract

Section: Discussionmentioning

confidence: 83%

Section: Sdamentioning

confidence: 99%

Validating an SVM-based neonatal seizure detection algorithm for generalizability, non-inferiority and clinical efficacy

Tapani¹,

Nevalainen²,

Vanhatalo³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Applying the artefact detection algorithm to the entire EEG recordings resulted in an overall artefact burden of 30% over all of the channels (sum of five artefact types, Table 3). A seizure detection algorithm was initially used as an additional layer of automated annotation to exclude seizure events [32]. Figure 7 shows the amount of data labelled as Clean by the artefact detection algorithm for each channel in the Double Banana bipolar montage, with Table S.…”

Section: S2) the Majority Of Permutations Of Network Architecture And Training Options Gave Similar Resultsmentioning

confidence: 99%

“…This algorithm can be used i) to aid bedside nursing staff in monitoring EEG quality in real time, in order to take corrective actions (e.g. detection of EL artefact can suggest poor contact of specific electrodes), and ii) to support EEG review by clinician's and complement future diagnostic tools, such as seizure detectors, lesion detectors, or EEG background classifiers [28,[32][33][34][35].…”

Section: Discussionmentioning

confidence: 99%

“…To simulate a potential application of the artefact detection algorithm (trained on the Balanced Dataset), we applied it in conjunction with a seizure detection algorithm [32] to the complete EEG recordings from all 79 patients to estimate the level of contamination within our database. The level of contamination was compared between channels in the bipolar montage.…”

Section: Algorithm Performance Assessmentmentioning

confidence: 99%

See 1 more Smart Citation

Automated detection of artefacts in neonatal EEG with residual neural networks

Webb

Kauppila

Roberts

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Background and Objective: To develop a computational algorithm that detects and identifies different artefact types in neonatal electroencephalography (EEG) signals. Methods: As part of a larger algorithm, we trained a Residual Deep Neural Network on expert human annotations of EEG recordings from 79 term infants recorded in a neonatal intensive care unit (112 h of 18-channel recording). The network was trained using 10 fold cross validation in Matlab. Artefact types included: device interference, EMG, movement, electrode pop, and non-cortical biological rhythms. Performance was assessed by prediction statistics and further validated on a separate independent dataset of 13 term infants (143 h of 3-channel recording). EEG pre-processing steps, and other post-processing steps such as averaging probability over a temporal window, were also included in the algorithm. Results: The Residual Deep Neural Network showed high accuracy (95%) when distinguishing periods of clean, artefact-free EEG from any kind of artefact, with a median accuracy for individual patient of 91% (IQR: 81%-96%). The accuracy in identifying the five different types of artefacts ranged from 57%-92%, with electrode pop being the hardest to detect and EMG being the easiest. This reflected the proportion of artefact available in the training dataset. Misclassification as clean was low for each artefact type, ranging from 1%-11%. The detection accuracy was lower on the validation set (87%). We used the algorithm to show that EEG channels located near the vertex were the least susceptible to artefact. Conclusion: Artefacts can be accurately and reliably identified in the neonatal EEG using a deep learning algorithm. Artefact detection algorithms can provide continuous bedside quality assessment and support EEG review by clinicians or analysis algorithms.

show abstract