EEG‐based outcome prediction after cardiac arrest with convolutional neural networks: Performance and visualization of discriminative features

Abstract

Prognostication for comatose patients after cardiac arrest is a difficult but essential task. Currently, visual interpretation of electroencephalogram (EEG) is one of the main modality used in outcome prediction. There is a growing interest in computer‐assisted EEG interpretation, either to overcome the possible subjectivity of visual interpretation, or to identify complex features of the EEG signal. We used a one‐dimensional convolutional neural network (CNN) to predict functional outcome based on 19‐channel‐EEG recorded from 267 adult comatose patients during targeted temperature management after CA. The area under the receiver operating characteristic curve (AUC) on the test set was 0.885. Interestingly, model architecture and fine‐tuning only played a marginal role in classification performance. We then used gradient‐weighted class activation mapping (Grad‐CAM) as visualization technique to identify which EEG features were used by the network to classify an EEG epoch as favorable or unfavorable outcome, and also to understand failures of the network. Grad‐CAM showed that the network relied on similar features than classical visual analysis for predicting unfavorable outcome (suppressed background, epileptiform transients). This study confirms that CNNs are promising models for EEG‐based prognostication in comatose patients, and that Grad‐CAM can provide explanation for the models' decision‐making, which is of utmost importance for future use of deep learning models in a clinical setting.

Standardized visual EEG features predict outcome in patients with acute consciousness impairment of various etiologies

“…This is especially true for electroencephalography (EEG). In patients with HIE, EEG has become the main prognostic tool, as several visual and quantitative (computer-derived) features have been shown to predict functional outcome [6][7][8][9][10][11][12][13]. In particular, a continuous and reactive EEG background suggests a favorable outcome, whereas a suppressed background or burst suppression with identical bursts is usually predictor of poor outcome [7,8].…”

Targeted temperature management and early neuro-prognostication after cardiac arrest

“…100 Computer-assisted interpretation of early EEG (9-30 hours after CA) during TTM using a convolutional neural network achieved an AUC of 0.885 in discriminating good and poor neurological outcome. 101 Nonetheless, confounding variables such as the effect of hypothermia on EEG or sedatives used during TTM must be considered, 102 although recent studies suggested that sedation required for TTM did not alter the predictive value of standardized EEG interpretation, 93 and EEG patterns were not significantly changed in propofol-sedated patients undergoing TTM (both 33 C and 36 C targets). 103 Computerassisted interpretation of EEG is a promising option and warrants further investigation.…”

Exploring sMRI Biomarkers for Diagnosis of Autism Spectrum Disorders Based on Multi Class Activation Mapping Models

“…The CAM-based models have been successfully applied to the classification of muscular dystrophies [35], tumor diagnosis [36], EEG signal interpretation [37] and other medical research fields [38]. The biggest advantage of these models is that it can accurately locate the abnormal positions of features (such as electrical signals, cerebral cortex, pathological images, etc.)…”