Diagnostic and prognostic EEG analysis of critically ill patients: A deep learning study

Jonas, Stefan; Müller, Michael; Rossetti, Andrea O.; Rüegg, Stephan; Alvarez, Vincent; Schindler, Kaspar; Zubler, Frédéric

doi:10.1016/j.nicl.2022.103167

Cited by 4 publications

(5 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…When the networks were directly compared to other computational methods, their performance was equivalent, and the DL algorithms were more resistant to noise ( 18 ). When applied to a group of patients with various etiologies of coma ( 29 ), the network's performance dropped (AUC 70%), whereas visual scoring of specific features together with a random forest classifier achieved an AUC of 80% on the same data set ( 17 ). Note however that the combined visual feature/random forest approach leveraged the knowledge of EEG experts with specific qualification in ICU EEG, and was allowed to use information concerning EEG reactivity, namely the modification of the EEG background after auditory of somatosensory stimulus, whereas the DL algorithm was trained exclusively on resting state EEG (in absence of stimulation).…”

Section: Discussionmentioning

confidence: 99%

“…Algorithms should also be extensively tested in sub-groups of coma patients, to ensure that they perform as expected not only at group level, but also within individual sub-populations. This implies, among others, the tedious work of inspecting false positive and false negative rates, to understand the reason for the misclassification ( 29 ).…”

Section: Discussionmentioning

confidence: 99%

“…It is reassuring that most studies on DL-based prognostication include techniques to improve interpretability, or other control analyses. For instance ( 28 , 29 ), used a so-called gradient method, which highlighted which part of the EEG signal was discriminative for each class. Interestingly, some patterns which were relevant for the network were similar to the ones used by human experts such as a suppressed background or spiky transients for unfavorable outcome; and continuous alpha or theta activity with posterior-anterior gradient, or generalized rhythmic delta activity for favorable outcome.…”

Section: Discussionmentioning

confidence: 99%

“…Additionally, algorithms in clinical settings should offer a measure of confidence in their single-patient prognostication, for instance by considering probabilities instead binary decisions. In fact, two studies have shown that the predicted “probability” output by the last networks layer correlates with actual probability ( 30 ) and with accuracy ( 29 ) of the prediction.…”

Section: Discussionmentioning

confidence: 99%

“…The t-VGG-GAP architecture was then used for outcome prediction in multicentric cohort of patients with different etiologies of coma (of which 30% after CA) ( 29 ). The authors used five-fold cross validation (286 patients for training and 72 for testing).…”

Section: Review Of the Literaturementioning

confidence: 99%

See 4 more Smart Citations

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Zubler¹,

Tzovara²

2023

Front. Neurol.

View full text Add to dashboard Cite

Outcome prognostication in comatose patients after cardiac arrest (CA) remains to date a challenge. The major determinant of clinical outcome is the post-hypoxic/ischemic encephalopathy. Electroencephalography (EEG) is routinely used to assess neural functions in comatose patients. Currently, EEG-based outcome prognosis relies on visual evaluation by medical experts, which is time consuming, prone to subjectivity, and oblivious to complex patterns. The field of deep learning has given rise to powerful algorithms for detecting patterns in large amounts of data. Analyzing EEG signals of coma patients with deep neural networks with the goal of assisting in outcome prognosis is therefore a natural application of these algorithms. Here, we provide the first narrative literature review on the use of deep learning for prognostication after CA. Existing studies show overall high performance in predicting outcome, relying either on spontaneous or on auditory evoked EEG signals. Moreover, the literature is concerned with algorithmic interpretability, and has shown that largely, deep neural networks base their decisions on clinically or neurophysiologically meaningful features. We conclude this review by discussing considerations that the fields of artificial intelligence and neurology will need to jointly address in the future, in order for deep learning algorithms to break the publication barrier, and to be integrated in clinical practice.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Review Of the Literaturementioning

confidence: 99%

See 3 more Smart Citations

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Zubler¹,

Tzovara²

2023

Front. Neurol.

View full text Add to dashboard Cite

show abstract

Application of advanced biosensors in nervous system diseases

Meng,

Zhang,

Yang

et al. 2024

Interdisciplinary Medicine

View full text Add to dashboard Cite

Nervous system diseases are among the most common diseases globally, posing a severe threat to patients' quality of life and placing a considerable burden on families and society. With improvements in miniaturization, intelligence, and the safety of biosensors, the combination of machinery and organisms is becoming increasingly common. In neuroscience research, biosensors of different macroscopic dimensions have been uniquely utilized to harness their relevant properties. One‐dimensional (1D) biosensors can achieve in situ real‐time monitoring of neural markers at the subcellular, single‐cell, ex vivo, and in vivo levels, with reduced impacts on organisms. Two‐dimensional (2D) biosensors can monitor the chemical behavior of cells and the neural activity of living animals. They are helpful for objectively identifying the characteristics of cells in response to external stimuli and studying the neural circuits of living animals. Three‐dimensional (3D) biosensors have shown unique advantages in point‐of‐care testing, liquid biopsy, drug screening, and mechanistic research. In clinical practice, brain‐computer interfaces (BCIs) and wearable devices have become important tools for monitoring and treatment. To date, there has been widespread adoption of BCIs in clinical practice. BCIs not only exhibit good efficacy in severe neurological and mental diseases but also provide a method for early diagnosis and treatment of these diseases. Wearable sensor devices can accurately assess the symptoms of movement disorders and play an active role in rehabilitation and treatment. In this review, we summarize the application of advanced biosensors in neuroscience research and clinical practice. The challenges and prospects of biosensors as applied to nervous system diseases under interdisciplinary promotion are also discussed in depth.

show abstract

Machine Learning and Artificial Intelligence Applications to Epilepsy: a Review for the Practicing Epileptologist

Kerr,

McFarlane

2023

Curr Neurol Neurosci Rep

View full text Add to dashboard Cite

Diagnostic and prognostic EEG analysis of critically ill patients: A deep learning study

Cited by 4 publications

References 54 publications

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Application of advanced biosensors in nervous system diseases

Machine Learning and Artificial Intelligence Applications to Epilepsy: a Review for the Practicing Epileptologist

Contact Info

Product

Resources

About