Ultrafast review of ambulatory EEGs with deep learning.

OBJECTIVE: Interictal epileptiform discharges (IED) are hallmark biomarkers of epilepsy which are typically detected through visual analysis. Deep learning has shown potential in automating IED detection, which could reduce the burden of visual analysis in clinical practice. This is particularly relevant for ambulatory electroencephalograms (EEGs), as these entail longer review times. METHODS: We applied a previously trained neural network to an independent dataset of 100 ambulatory EEGs (average duration 20.6 h). From these, 42 EEGs contained IEDs, 25 were abnormal without IEDs and 33 were normal. The algorithm flagged 2 second epochs that it considered IEDs. The EEGs were provided to an expert, who used NeuroCenter EEG to review the recordings. The expert concluded if each recording contained IEDs, and was timed during the process. RESULTS: The conclusion of the reviewer was the same as the EEG report in 97% of the recordings. Three EEGs contained IEDs that were not detected based on the flagged epochs. Review time for the 100 EEGs was approximately 4 h, with half of the recordings taking <2 minutes to review. CONCLUSIONS: Our network can be used to reduce time spent on visual analysis in the clinic by 50-75 times with high reliability. SIGNIFICANCE: Given the large time reduction potential and high success rate, this algorithm can be used in the clinic to aid in visual analysis.

Machine learning for detection of interictal epileptiform discharges.

The electroencephalogram (EEG) is a fundamental tool in the diagnosis and classification of epilepsy. In particular, Interictal Epileptiform Discharges (IEDs) reflect an increased likelihood of seizures and are routinely assessed by visual analysis of the EEG. Visual assessment is, however, time consuming and prone to subjectivity, leading to a high misdiagnosis rate and motivating the development of automated approaches. Research towards automating IED detection started 45 years ago. Approaches range from mimetic methods to deep learning techniques. We review different approaches to IED detection, discussing their performance and limitations. Traditional machine learning and deep learning methods have yielded the best results so far and their application in the field is still growing. Standardization of datasets and outcome measures is necessary to compare models more objectively and decide which should be implemented in a clinical setting.

Efficient use of clinical EEG data for deep learning in epilepsy.

OBJECTIVE: Automating detection of Interictal Epileptiform Discharges (IEDs) in electroencephalogram (EEG) recordings can reduce the time spent on visual analysis for the diagnosis of epilepsy. Deep learning has shown potential for this purpose, but the scarceness of expert annotated data creates a bottleneck in the process. METHODS: We used EEGs from 50 patients with focal epilepsy, 49 patients with generalized epilepsy (IEDs were visually labeled by experts) and 67 controls. The data was filtered, downsampled and cut into two second epochs. We increased the number of input samples containing IEDs through temporal shifting and using different montages. A VGG C convolutional neural network was trained to detect IEDs. RESULTS: Using the dataset with more samples, we reduced the false positive rate from 2.11 to 0.73 detections per minute at the intersection of sensitivity and specificity. Sensitivity increased from 63% to 96% at 99% specificity. The model became less sensitive to the position of the IED in the epoch and montage. CONCLUSIONS: Temporal shifting and use of different EEG montages improves performance of deep neural networks in IED detection. SIGNIFICANCE: Dataset augmentation can reduce the need for expert annotation, facilitating the training of neural networks, potentially leading to a fundamental shift in EEG analysis.

Outcome Prediction in Postanoxic Coma With Deep Learning.

OBJECTIVES: Visual assessment of the electroencephalogram by experienced clinical neurophysiologists allows reliable outcome prediction of approximately half of all comatose patients after cardiac arrest. Deep neural networks hold promise to achieve similar or even better performance, being more objective and consistent. DESIGN: Prospective cohort study. SETTING: Medical ICU of five teaching hospitals in the Netherlands. PATIENTS: Eight-hundred ninety-five consecutive comatose patients after cardiac arrest. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Continuous electroencephalogram was recorded during the first 3 days after cardiac arrest. Functional outcome at 6 months was classified as good (Cerebral Performance Category 1-2) or poor (Cerebral Performance Category 3-5). We trained a convolutional neural network, with a VGG architecture (introduced by the Oxford Visual Geometry Group), to predict neurologic outcome at 12 and 24 hours after cardiac arrest using electroencephalogram epochs and outcome labels as inputs. Output of the network was the probability of good outcome. Data from two hospitals were used for training and internal validation (n = 661). Eighty percent of these data was used for training and cross-validation, the remaining 20% for independent internal validation. Data from the other three hospitals were used for external validation (n = 234). Prediction of poor outcome was most accurate at 12 hours, with a sensitivity in the external validation set of 58% (95% CI, 51-65%) at false positive rate of 0% (CI, 0-7%). Good outcome could be predicted at 12 hours with a sensitivity of 48% (CI, 45-51%) at a false positive rate of 5% (CI, 0-15%) in the external validation set. CONCLUSIONS: Deep learning of electroencephalogram signals outperforms any previously reported outcome predictor of coma after cardiac arrest, including visual electroencephalogram assessment by trained electroencephalogram experts. Our approach offers the potential for objective and real time, bedside insight in the neurologic prognosis of comatose patients after cardiac arrest.