Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clinical translation of machine learning algorithms for seizure detection in scalp electroencephalography: systematic review

Nina Moutonnet, Steven White, Benjamin P Campbell, Saeid Sanei, Toshihisa Tanaka, Hong Ji, Danilo Mandic, Gregory Scott

TL;DR

This paper addresses the gap between ML-based seizure-detection research and clinical deployment using scalp EEG by performing a structured, domain-informed review. It synthesizes evidence from public datasets, systematic search, and a broad range of ML approaches (feature-based and DL), emphasizing generalisability, data heterogeneity, and clinically relevant evaluation. The authors provide concrete guidelines for improving clinical translatability, including multi-domain representations, diverse training data, patient-independent validation, and transparent post-processing, while highlighting the need for robust ground-truth labeling and accounting for non-ictal activity. The work aims to accelerate real-world adoption of automated seizure detection by clarifying methodological best practices and identifying critical research gaps.

Abstract

Machine learning algorithms for seizure detection have shown considerable diagnostic potential, with recent reported accuracies reaching 100%. Yet, only few published algorithms have fully addressed the requirements for successful clinical translation. This is, for example, because the properties of training data may limit the generalisability of algorithms, algorithm performance may vary depending on which electroencephalogram (EEG) acquisition hardware was used, or run-time processing costs may be prohibitive to real-time clinical use cases. To address these issues in a critical manner, we systematically review machine learning algorithms for seizure detection with a focus on clinical translatability, assessed by criteria including generalisability, run-time costs, explainability, and clinically-relevant performance metrics. For non-specialists, the domain-specific knowledge necessary to contextualise model development and evaluation is provided. It is our hope that such critical evaluation of machine learning algorithms with respect to their potential real-world effectiveness can help accelerate clinical translation and identify gaps in the current seizure detection literature.

Clinical translation of machine learning algorithms for seizure detection in scalp electroencephalography: systematic review

TL;DR

This paper addresses the gap between ML-based seizure-detection research and clinical deployment using scalp EEG by performing a structured, domain-informed review. It synthesizes evidence from public datasets, systematic search, and a broad range of ML approaches (feature-based and DL), emphasizing generalisability, data heterogeneity, and clinically relevant evaluation. The authors provide concrete guidelines for improving clinical translatability, including multi-domain representations, diverse training data, patient-independent validation, and transparent post-processing, while highlighting the need for robust ground-truth labeling and accounting for non-ictal activity. The work aims to accelerate real-world adoption of automated seizure detection by clarifying methodological best practices and identifying critical research gaps.

Abstract

Machine learning algorithms for seizure detection have shown considerable diagnostic potential, with recent reported accuracies reaching 100%. Yet, only few published algorithms have fully addressed the requirements for successful clinical translation. This is, for example, because the properties of training data may limit the generalisability of algorithms, algorithm performance may vary depending on which electroencephalogram (EEG) acquisition hardware was used, or run-time processing costs may be prohibitive to real-time clinical use cases. To address these issues in a critical manner, we systematically review machine learning algorithms for seizure detection with a focus on clinical translatability, assessed by criteria including generalisability, run-time costs, explainability, and clinically-relevant performance metrics. For non-specialists, the domain-specific knowledge necessary to contextualise model development and evaluation is provided. It is our hope that such critical evaluation of machine learning algorithms with respect to their potential real-world effectiveness can help accelerate clinical translation and identify gaps in the current seizure detection literature.
Paper Structure (39 sections, 1 equation, 7 figures, 13 tables)

This paper contains 39 sections, 1 equation, 7 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Structure of this review
  3. Domain-specific knowledge
  4. EEG recording techniques
  5. Scalp EEG
  6. Conventional EEG frequency bands
  7. Seizure manifestations in scalp EEG
  8. Seizure types, duration, and evolution
  9. Seizure-related phases
  10. Epileptiform activity
  11. Seizure subtypes
  12. Non convulsive status epilepticus
  13. Scalp EEG corrupting artefacts
  14. Datasets
  15. PRISMA: A systematic search strategy
  16. ...and 24 more sections

Figures (7)

  • Figure 1: Potential use cases, challenges and recording modalities for automated scalp EEG seizure detection (Created with BioRender, created_with_biorender). Applications of seizure detection algorithms range widely, from (1) highlighting to clinicians sections of interest in long recordings to simplify offline annotation; (2) real-time seizure detection using continuous EEG in ambulatory patients, telemetry units, or ICU; to (3) automatic recording of seizure diary entries.
  • Figure 2: The 10-20 electrode placement system with front-back (nasion to inion) 10% and 20% electrode distances. The spatial resolution of scalp EEG setups can range from as few as 1 channel (low resolution) to 256 channels (high resolution). Each EEG electrode is labelled by one or two alphabetic characters followed by a digit. This combination describes the location of each electrode. The alphabetic characters are Fp for frontal-polar, F for frontal, P for parietal, T for temporal, O for occipital, and C for the central region of the brain. Odd-numbered electrodes are located on the left side of the brain, even-numbered electrodes on the right side, and Z electrodes lie along the midline of the scalp. shriram2013eeg (Created with BioRender, created_with_biorender_1020)
  • Figure 3: Bipolar and monopolar EEG montage. (Left) Double banana bipolar montage, where each electrode is referenced to the electrode behind it. Observe an outside temporal chain (e.g. Fp2-F8-T4-T6-O2) and an inside parasagittal chain (e.g. Fp2-F4-C4-P4-O2) on each side of the scalp, and a unique central chain in the middle (Fz-Cz-Pz). (Right) Monopolar montage, where all the electrodes are referenced to a single point. (Created with BioRender, created_with_biorender_1020)
  • Figure 4: Main seizure types and some EEG characteristics. (Top) Normal brain activity, focal seizure and focal onset seizure with secondary generalisation alongside their EEG correlate. (Bottom) Nomenclature of seizure phases including demonstrative EEG segments of inter-ictal, pre-ictal, ictal and post-ictal activity. (Created with BioRender, created_with_biorender_seizure)
  • Figure 5: PRISMA flowchart. Reasons for exclusion of reports are numbered following the list of exclusion criteria provided above. 'Other' includes retracted articles, studies that are not seizure detection and studies that did not meet reasonable quality standards. (Created with BioRender, created_with_biorender_prisma)
  • ...and 2 more figures