Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Epilepsy Seizures Classification to Detection: A Deep Learning-based Approach for Raw EEG Signals

Davy Darankoum, Manon Villalba, Clelia Allioux, Baptiste Caraballo, Carine Dumont, Eloise Gronlier, Corinne Roucard, Yann Roche, Chloe Habermacher, Sergei Grudinin, Julien Volle

TL;DR

The paper addresses automated seizure detection from raw EEG by developing a preclinical-to-clinical pipeline that avoids leakage-prone preprocessing. It introduces a preprocessing approach that does not require prior seizure labeling, a postprocessing step to reconstruct continuous EEG and align seizures with ground truth, and two evaluation strategies for classification and detection. Among architectures, a CNN+Transformer emerges as the top performer, with strong cross-species generalization from MTLE mice to human Bonn data, while emphasizing realistic, event-based evaluation. The work demonstrates robust generalization and discusses practical implications for preclinical research and ASM development, supported by rigorous data-splitting to prevent leakage and by a realistic detection pipeline that preserves temporal precision.

Abstract

Epilepsy represents the most prevalent neurological disease in the world. One-third of people suffering from mesial temporal lobe epilepsy (MTLE) exhibit drug resistance, urging the need to develop new treatments. A key part in anti-seizure medication (ASM) development is the capability of detecting and quantifying epileptic seizures occurring in electroencephalogram (EEG) signals, which is crucial for treatment efficacy evaluation. In this study, we introduced a seizure detection pipeline based on deep learning models applied to raw EEG signals. This pipeline integrates: a new pre-processing technique which segments continuous raw EEG signals without prior distinction between seizure and seizure-free activities; a post-processing algorithm developed to reassemble EEG segments and allow the identification of seizures start/end; and finally, a new evaluation procedure based on a strict seizure events comparison between predicted and real labels. Models training have been performed using a data splitting strategy which addresses the potential for data leakage. We demonstrated the fundamental differences between a seizure classification and a seizure detection task and showed the differences in performance between the two tasks. Finally, we demonstrated the generalization capabilities across species of our best architecture, combining a Convolutional Neural Network and a Transformer encoder. The model was trained on animal EEGs and tested on human EEGs with a F1-score of 93% on a balanced Bonn dataset.

From Epilepsy Seizures Classification to Detection: A Deep Learning-based Approach for Raw EEG Signals

TL;DR

The paper addresses automated seizure detection from raw EEG by developing a preclinical-to-clinical pipeline that avoids leakage-prone preprocessing. It introduces a preprocessing approach that does not require prior seizure labeling, a postprocessing step to reconstruct continuous EEG and align seizures with ground truth, and two evaluation strategies for classification and detection. Among architectures, a CNN+Transformer emerges as the top performer, with strong cross-species generalization from MTLE mice to human Bonn data, while emphasizing realistic, event-based evaluation. The work demonstrates robust generalization and discusses practical implications for preclinical research and ASM development, supported by rigorous data-splitting to prevent leakage and by a realistic detection pipeline that preserves temporal precision.

Abstract

Epilepsy represents the most prevalent neurological disease in the world. One-third of people suffering from mesial temporal lobe epilepsy (MTLE) exhibit drug resistance, urging the need to develop new treatments. A key part in anti-seizure medication (ASM) development is the capability of detecting and quantifying epileptic seizures occurring in electroencephalogram (EEG) signals, which is crucial for treatment efficacy evaluation. In this study, we introduced a seizure detection pipeline based on deep learning models applied to raw EEG signals. This pipeline integrates: a new pre-processing technique which segments continuous raw EEG signals without prior distinction between seizure and seizure-free activities; a post-processing algorithm developed to reassemble EEG segments and allow the identification of seizures start/end; and finally, a new evaluation procedure based on a strict seizure events comparison between predicted and real labels. Models training have been performed using a data splitting strategy which addresses the potential for data leakage. We demonstrated the fundamental differences between a seizure classification and a seizure detection task and showed the differences in performance between the two tasks. Finally, we demonstrated the generalization capabilities across species of our best architecture, combining a Convolutional Neural Network and a Transformer encoder. The model was trained on animal EEGs and tested on human EEGs with a F1-score of 93% on a balanced Bonn dataset.
Paper Structure (37 sections, 4 equations, 5 figures, 3 tables)

This paper contains 37 sections, 4 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Animals - MTLE mice model
  4. Datasets
  5. Dataset 1:
  6. Dataset 2:
  7. Pre-processing and post-processing pipelines
  8. Dataset 1
  9. Pre-processing with prior identification of seizure/seizure-free activity (Pre-processing I)
  10. Pre-processing without prior identification of seizure/seizure-free activity (Pre-processing II)
  11. Post-processing for signal reconstitution (Post-processing II)
  12. Evaluation strategy for seizure classification (Evaluation I)
  13. Evaluation strategy for seizure detection (Evaluation II)
  14. Dataset 2
  15. Training, Validation, and Test sets
  16. ...and 22 more sections

Figures (5)

  • Figure 1: Examples of EEG signals from Dataset 1. Snapshots of 3 portions of EEG signals measured in 3 mice and labeled by the same expert. Labels in red indicate detected seizures, and blue labels represent seizure-free activities. The total duration of the snapshots is 88 seconds.
  • Figure 2: Task-based (Classification and Detection) pipelines summary.
  • Figure 3: Illustration of the classification task pipeline at inference time. Segments built out of pre-processing I do not overlap across two activities. Segments colored in blue or red reflect an example of classification by the trained model. Blue color corresponds to classification into seizure-free activity. The red color indicates classification into seizure activity. TP: Segment labeled as seizure and detected as seizure by the model.TN: Segment labeled as seizure-free and detected as seizure-free. FP: Segment labeled as seizure-free and detected as seizure by the model. FN: Segment labeled as seizure and detected as seizure-free.
  • Figure 4: Illustration of the detection task pipeline at inference time. Segments built out of pre-processing II do overlap across activities exhibiting a real-world scenario. Segments predicted through the model application are re-assembled into a continuous signal using post-processing D. Finally, following evaluation II strategy, events formed in the reconstituted signal are compared with events labeled by the expert. TP: A seizure start & end labeled by the expert matches a seizure start & end detected by the model. FP: A seizure start & end detected by the model did not find a match with any seizure start & end labeled by the expert. FN: A seizure start & end labeled by the expert did not find a match with any seizure start & end detected by the model.
  • Figure 5: Schematic representation of the dual path CNN/transformer architecture. $K$ denotes the kernel size.