Permutation Entropy-Based Interpretability of Convolutional Neural Network Models for Interictal EEG Discrimination of Subjects with Epileptic Seizures vs. Psychogenic Non-Epileptic Seizures

Improving Clinician Performance in Classifying EEG Patterns on the Ictal-Interictal Injury Continuum Using Interpretable Machine Learning

BACKGROUND

In intensive care units (ICUs), critically ill patients are monitored with electroencephalography (EEG) to prevent serious brain injury. EEG monitoring is constrained by clinician availability, and EEG interpretation can be subjective and prone to interobserver variability. Automated deep-learning systems for EEG could reduce human bias and accelerate the diagnostic process. However, existing uninterpretable (black-box) deep-learning models are untrustworthy, difficult to troubleshoot, and lack accountability in real-world applications, leading to a lack of both trust and adoption by clinicians.

METHODS

We developed an interpretable deep-learning system that accurately classifies six patterns of potentially harmful EEG activity - seizure, lateralized periodic discharges (LPDs), generalized periodic discharges (GPDs), lateralized rhythmic delta activity (LRDA), generalized rhythmic delta activity (GRDA), and other patterns - while providing faithful case-based explanations of its predictions. The model was trained on 50,697 total 50-second continuous EEG samples collected from 2711 patients in the ICU between July 2006 and March 2020 at Massachusetts General Hospital. EEG samples were labeled as one of the six EEG patterns by 124 domain experts and trained annotators. To evaluate the model, we asked eight medical professionals with relevant backgrounds to classify 100 EEG samples into the six pattern categories - once with and once without artificial intelligence (AI) assistance - and we assessed the assistive power of this interpretable system by comparing the diagnostic accuracy of the two methods. The model's discriminatory performance was evaluated with area under the receiver-operating characteristic curve (AUROC) and area under the precision-recall curve. The model's interpretability was measured with task-specific neighborhood agreement statistics that interrogated the similarities of samples and features. In a separate analysis, the latent space of the neural network was visualized by using dimension reduction techniques to examine whether the ictal-interictal injury continuum hypothesis, which asserts that seizures and seizure-like patterns of brain activity lie along a spectrum, is supported by data.

RESULTS

The performance of all users significantly improved when provided with AI assistance. Mean user diagnostic accuracy improved from 47 to 71% (P<0.04). The model achieved AUROCs of 0.87, 0.93, 0.96, 0.92, 0.93, and 0.80 for the classes seizure, LPD, GPD, LRDA, GRDA, and other patterns, respectively. This performance was significantly higher than that of a corresponding uninterpretable black-box model (with P<0.0001). Videos traversing the ictal-interictal injury manifold from dimension reduction (a two-dimensional representation of the original high-dimensional feature space) give insight into the layout of EEG patterns within the network's latent space and illuminate relationships between EEG patterns that were previously hypothesized but had not yet been shown explicitly. These results indicate that the ictal-interictal injury continuum hypothesis is supported by data.

CONCLUSIONS

Users showed significant pattern classification accuracy improvement with the assistance of this interpretable deep-learning model. The interpretable design facilitates effective human-AI collaboration; this system may improve diagnosis and patient care in clinical settings. The model may also provide a better understanding of how EEG patterns relate to each other along the ictal-interictal injury continuum. (Funded by the National Science Foundation, National Institutes of Health, and others.).

Epileptic Seizure Detection with an End-to-End Temporal Convolutional Network and Bidirectional Long Short-Term Memory Model

Automatic seizure detection plays a key role in assisting clinicians for rapid diagnosis and treatment of epilepsy. In view of the parallelism of temporal convolutional network (TCN) and the capability of bidirectional long short-term memory (BiLSTM) in mining the long-range dependency of multi-channel time-series, we propose an automatic seizure detection method with a novel end-to-end TCN-BiLSTM model in this work. First, raw EEG is filtered with a 0.5-45 Hz band-pass filter, and the filtered data are input into the proposed TCN-BiLSTM network for feature extraction and classification. Post-processing process including moving average filtering, thresholding and collar technique is then employed to further improve the detection performance. The method was evaluated on two EEG database. On the CHB-MIT scalp EEG database, our method achieved a segment-based sensitivity of 94.31%, specificity of 97.13%, and accuracy of 97.09%. Meanwhile, an event-based sensitivity of 96.48% and an average false detection rate (FDR) of 0.38/h were obtained. On the SH-SDU database we collected, the segment-based sensitivity of 94.99%, specificity of 93.25%, and accuracy of 93.27% were achieved. In addition, an event-based sensitivity of 99.35% and a false detection rate of 0.54/h were yielded. The total detection time consumed for 1[Formula: see text]h EEG data was 5.65[Formula: see text]s. These results demonstrate the superiority and promising potential of the proposed method in real-time monitoring of epileptic seizures.

Classification of Epileptic and Psychogenic Nonepileptic Seizures via Time-Frequency Features of EEG Data

The majority of psychogenic nonepileptic seizures (PNESs) are brought on by psychogenic causes, but because their symptoms resemble those of epilepsy, they are frequently misdiagnosed. Although EEG signals are normal in PNES cases, electroencephalography (EEG) recordings alone are not sufficient to identify the illness. Hence, accurate diagnosis and effective treatment depend on long-term video EEG data and a complete patient history. Video EEG setup, however, is more expensive than using standard EEG equipment. To distinguish PNES signals from conventional epileptic seizure (ES) signals, it is crucial to develop methods solely based on EEG recordings. The proposed study presents a technique utilizing short-term EEG data for the classification of inter-PNES, PNES, and ES segments using time-frequency methods such as the Continuous Wavelet transform (CWT), Short-Time Fourier transform (STFT), CWT-based synchrosqueezed transform (WSST), and STFT-based SST (FSST), which provide high-resolution time-frequency representations (TFRs). TFRs of EEG segments are utilized to generate 13 joint TF (J-TF)-based features, four gray-level co-occurrence matrix (GLCM)-based features, and 16 higher-order joint TF moment (HOJ-Mom)-based features. These features are then employed in the classification procedure. Both three-class (inter-PNES versus PNES versus ES: ACC: 80.9%, SEN: 81.8%, and PRE: 84.7%) and two-class (Inter-PNES versus PNES: ACC: 88.2%, SEN: 87.2%, and PRE: 86.1%; PNES versus ES: ACC: 98.5%, SEN: 99.3%, and PRE: 98.9%) classification algorithms performed well, according to the experimental results. The STFT and FSST strategies surpass the CWT and WSST strategies in terms of classification accuracy, sensitivity, and precision. Moreover, the J-TF-based feature sets often perform better than the other two.

Effective Early Detection of Epileptic Seizures through EEG Signals Using Classification Algorithms Based on t-Distributed Stochastic Neighbor Embedding and K-Means

Epilepsy is a neurological disorder in the activity of brain cells that leads to seizures. An electroencephalogram (EEG) can detect seizures as it contains physiological information of the neural activity of the brain. However, visual examination of EEG by experts is time consuming, and their diagnoses may even contradict each other. Thus, an automated computer-aided diagnosis for EEG diagnostics is necessary. Therefore, this paper proposes an effective approach for the early detection of epilepsy. The proposed approach involves the extraction of important features and classification. First, signal components are decomposed to extract the features via the discrete wavelet transform (DWT) method. Principal component analysis (PCA) and the t-distributed stochastic neighbor embedding (t-SNE) algorithm were applied to reduce the dimensions and focus on the most important features. Subsequently, K-means clustering + PCA and K-means clustering + t-SNE were used to divide the dataset into subgroups to reduce the dimensions and focus on the most important representative features of epilepsy. The features extracted from these steps were fed to extreme gradient boosting, K-nearest neighbors (K-NN), decision tree (DT), random forest (RF) and multilayer perceptron (MLP) classifiers. The experimental results demonstrated that the proposed approach provides superior results to those of existing studies. During the testing phase, the RF classifier with DWT and PCA achieved an accuracy of 97.96%, precision of 99.1%, recall of 94.41% and F1 score of 97.41%. Moreover, the RF classifier with DWT and t-SNE attained an accuracy of 98.09%, precision of 99.1%, recall of 93.9% and F1 score of 96.21%. In comparison, the MLP classifier with PCA + K-means reached an accuracy of 98.98%, precision of 99.16%, recall of 95.69% and F1 score of 97.4%.

Epileptic seizure prediction and classification based on statistical features using LSTM fully connected neural network

Epilepsy, the most common neurological disorder by which over 65 million people are affected across the world. Recent research has shown a very large interest to predict and diagnose epilepsy well before time. The continuous monitoring of EEG signals for seizure detection in electroencephalogram (EEG) is a very tedious and time taking process and therefore requires a qualified and trained clinical specialist. This paper presents a novel approach to detect and predict the epileptic signal in the recorded electroencephalogram (EEG). There is always a requirement for a nonlinear technique to examine the EEG signals due to the random nature of EEG signals. Therefore, we are providing an alternate method that extracts various entropy measures such Sample Entropy, Spectral Entropy, Permutation Entropy, and Shannon Entropy as statistical features from EEG signal. Based on these extracted features LSTM Fully connected Neural Network is used to classify the EEG signal as Focal and Non-focal. The proposed method gives a new insight into EEG signals by providing sensitivity as an added measure using deep learning along with accuracy and precision.

Effects of Pharmacological Treatments in Alzheimer's Disease: Permutation Entropy-Based EEG Complexity Study

Alzheimer's disease (AD) is a neurodegenerative brain disease affecting cognitive and physical functioning. The currently available pharmacological treatments for AD mainly contain cholinesterase inhibitors (AChE-I) and N-methyl-D-aspartic acid (NMDA) receptor antagonists (i.e., memantine). Because brain signals have complex nonlinear dynamics, there has been an increase in interest in researching complexity changes in the time series of brain signals in individuals with AD. In this study, we explore the electroencephalographic (EEG) complexity for making better observation of pharmacological therapy-based treatment effects on AD patients using the permutation entropy (PE) method. We examined EEG sub-band (delta, theta, alpha, beta, and gamma) complexity in de-novo, monotherapy (AChE-I), dual therapy (AChE-I and memantine) receiving AD participants compared with healthy elderly controls. We showed that each frequency band depicts its own complexity profile, which is regionally altered between groups. These alterations were also found to be associated with global cognitive scores. Overall, our findings indicate that entropy measures could be useful to show medication effects in AD.

Convolutional Neural Network Classification of Rest EEG Signals among People with Epilepsy, Psychogenic Non Epileptic Seizures and Control Subjects

Identifying subjects with epileptic seizures or psychogenic non-epileptic seizures from healthy subjects via interictal EEG analysis can be a very challenging issue. Indeed, at visual inspection, EEG can be normal in both cases. This paper proposes an automatic diagnosis approach based on deep learning to differentiate three classes: subjects with epileptic seizures (ES), subjects with non-epileptic psychogenic seizures (PNES) and control subjects (CS), analyzed by non-invasive low-density interictal scalp EEG recordings. The EEGs of 42 patients with new-onset ES, 42 patients with PNES video recorded and 19 patients with CS all with normal interictal EEG on visual inspection, were analyzed in the study; none of them was taking psychotropic drugs before registration. The processing pipeline applies empirical mode decomposition (EMD) to 5s EEG segments of 19 channels in order to extract enhanced features learned automatically from the customized convolutional neural network (CNN). The resulting CNN has been shown to perform well during classification, with an accuracy of 85.7%; these results encourage the use of deep processing systems to assist clinicians in difficult clinical settings.

Entropy Measures of Electroencephalograms towards the Diagnosis of Psychogenic Non-Epileptic Seizures

Psychogenic non-epileptic seizures (PNES) may resemble epileptic seizures but are not caused by epileptic activity. However, the analysis of electroencephalogram (EEG) signals with entropy algorithms could help identify patterns that differentiate PNES and epilepsy. Furthermore, the use of machine learning could reduce the current diagnosis costs by automating classification. The current study extracted the approximate sample, spectral, singular value decomposition, and Renyi entropies from interictal EEGs and electrocardiograms (ECG)s of 48 PNES and 29 epilepsy subjects in the broad, delta, theta, alpha, beta, and gamma frequency bands. Each feature-band pair was classified by a support vector machine (SVM), k-nearest neighbour (kNN), random forest (RF), and gradient boosting machine (GBM). In most cases, the broad band returned higher accuracy, gamma returned the lowest, and combining the six bands together improved classifier performance. The Renyi entropy was the best feature and returned high accuracy in every band. The highest balanced accuracy, 95.03%, was obtained by the kNN with Renyi entropy and combining all bands except broad. This analysis showed that entropy measures can differentiate between interictal PNES and epilepsy with high accuracy, and improved performances indicate that combining bands is an effective improvement for diagnosing PNES from EEGs and ECGs.

Time-Frequency Representations of Brain Oscillations: Which One Is Better?

Brain oscillations are thought to subserve important functions by organizing the dynamical landscape of neural circuits. The expression of such oscillations in neural signals is usually evaluated using time-frequency representations (TFR), which resolve oscillatory processes in both time and frequency. While a vast number of methods exist to compute TFRs, there is often no objective criterion to decide which one is better. In feature-rich data, such as that recorded from the brain, sources of noise and unrelated processes abound and contaminate results. The impact of these distractor sources is especially problematic, such that TFRs that are more robust to contaminants are expected to provide more useful representations. In addition, the minutiae of the techniques themselves impart better or worse time and frequency resolutions, which also influence the usefulness of the TFRs. Here, we introduce a methodology to evaluate the "quality" of TFRs of neural signals by quantifying how much information they retain about the experimental condition during visual stimulation and recognition tasks, in mice and humans, respectively. We used machine learning to discriminate between various experimental conditions based on TFRs computed with different methods. We found that various methods provide more or less informative TFRs depending on the characteristics of the data. In general, however, more advanced techniques, such as the superlet transform, seem to provide better results for complex time-frequency landscapes, such as those extracted from electroencephalography signals. Finally, we introduce a method based on feature perturbation that is able to quantify how much time-frequency components contribute to the correct discrimination among experimental conditions. The methodology introduced in the present study may be extended to other analyses of neural data, enabling the discovery of data features that are modulated by the experimental manipulation.