AI in Clinical Neurophysiology: from IED detection to neuroprognostication.
Abstract
Invited talk about role of cEEG. monitoring in the ICU and how AI/machine learning can assist in interpretation.
Full text
AI in Clinical Neurophysiology from IED detection to neuroprognostication Michel J.A.M. van Putten MD MSc PhD University of Twente & Medisch Spectrum Twente ISCN/BSCN 2025 meeting, Dublin
…..
…..
Machine learning Deep Learning Deep Neural networks
Human defines features Human generates model
abnormalnormal PDR 8-13 Hz? YES NO Human defines features Human generates model
Symmetric PDR with a frequency of 9.8 Hz. Reactivity (EC/EO) is within normal limits. f (Hz)f (Hz)
Machine Learning
William Holbrook Beard
Neural Network
Neural Network
2018 Here we show, in a ground truth scenario, that a deep neural net can predict sex from scalp electroencephalograms with an accuracy of > 80% (p < 0.0001), revealing that brain rhythms are sex specific.
1 s
1 s
1 s
1 s
Eyes open, no response… 1 s
+ some oxygen
40
Bronder et al, Resusc. Plus, 2022
Continuous background pattern Burst suppression with identical bursts
GOOD POOR
2 s Van Putten & Tjepkema (in preparation), 2025
Van Putten & Tjepkema (in preparation), 2025 DeepCRI
poor outcome (spec > 99%) sensitivity 37-45% good outcome (spec > 95%) sensitivity 50% < 24 h Van Putten & Tjepkema (in preparation), 2025
poor outcome (spec > 99%) sensitivity 37-45% good outcome (spec > 95%) sensitivity 50% < 24 h Van Putten & Tjepkema (in preparation), 2025 DeepCRI
poor outcome (spec > 99%) sensitivity 37-45% good outcome (spec > 95%) sensitivity 50% poor outcome (spec > 99%) sensitivity 81% good outcome (spec > 95%) sensitivity 90% < 24 h Van Putten & Tjepkema (in preparation), 2025 DeepCRI
Van Putten & Tjepkema (in preparation), 2025
Detection of anomalies (IEDs)
Detection of IEDs Modified from Noebels et al, 2012
Catarina da Silva Lourenço. Clin Neurophys, 2021
IED detection Seizure detection Seizure prediction
EEG No IED IED
Kural et al, Epilepsia 2022 Accurate identification of EEG recordings with interictal epileptiform discharges using a hybrid approach: Artificial intelligence supervised by human experts Encevis SpikeNet Persyst Sensitivity Specificity
Tveit et al, JAMA Neurology, 2023 30.000 routine EEGs SCORE-AI Normal vs abnormal Abnormal: focal IEDs, generalised IEDs, nonepileptiform focal & non-epileptiform diffuse Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence Validated 100 EEGs (multicenter, 11 experts) 10000 (single center, 14 experts) 60 EEGs (external reference)
Tveit et al, JAMA Neurology, 2023 30.000 routine EEGs SCORE-AI Normal vs abnormal Abnormal: focal IEDs, generalised IEDs, nonepileptiform focal & non-epileptiform diffuse Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence Validated 100 EEGs (multicenter, 11 experts) 10000 (single center, 14 experts) 60 EEGs (external reference) Algorithm Sensitivity (%) Specificity (%) SCORE-AI 87 90 Encevis 97 17 Persyst 100 3 SpikeNet 67 63
3 montages Efficient use of clinical EEG data for deep learning in epilepsy Da Silva Lourenço et al, Clin Neurophys 2021
3 montages Temporal shifting Efficient use of clinical EEG data for deep learning in epilepsy Da Silva Lourenço et al, Clin Neurophys 2021
2,200 25,000 IEDS 3 montages Temporal shifting Efficient use of clinical EEG data for deep learning in epilepsy Da Silva Lourenço et al, Clin Neurophys 2021
Explainability Da Silva Lourenço et al., Clinical Neurophysiology, 2021 Occlusion p=0.78p=0.79p=0.45 Input Sample
Explainability Da Silva Lourenço et al., Clinical Neurophysiology, 2021
Da Silva Lourenço et al., Clinical Neurophysiology, 2021 Lourenço et al, Clin Neurophys 2021 Sensitivity increased from 63% to 96% at 99% specificity Efficient use of clinical EEG data for deep learning in epilepsy
Expert level of detection of interictal discharges with a deep neural network Tjepkema-Cloostermans et al, Epilepsia, 2025
DeepCRI assists in reliable outcome prediction after cardiac arrest Deep Learning for IED detection is on par with human experts Review times can become 10-50 times faster It will have major impact on our clinical work
Clinical Neurophysiology