Unsupervised functional stage detection from EEG using topological data analysis

A new analysis method found clear transitions in many meditation and sleep EEG recordings without needing labeled training examples. This is an early signal-processing advance, not a diagnostic or treatment tool.

Abramov A, Mikhaylets E, Chernyshev V, Kaplan A, Ratnikov F.
Medical & Biological Engineering & Computing · Publication date: 2026-10-07
Added to site: 2026-10-09
DOI: 10.1007/s11517-026-03683-0

Published October 7, 2026. Added to Neurofeedback.com October 9, 2026.

Modalities

EEG, Quantitative signal analysis

Clinical focus

Unsupervised brain-state segmentation, Topological data analysis

Population

EEG recordings from Guhyasamaja Tantra meditation and whole-night polysomnographic sleep

Sample

Recording and participant counts were not reported in the accessible abstract

Intervention / method

Takens embeddings and Pearson dissimilarity matrices were transformed into tens of thousands of persistence-based topological features; a quick state-detecting algorithm selected features according to segmentation performance

Comparator

Conventional spectral features and temporally shuffled epochs served as methodological reference checks

Clinician summary

The topological feature pipeline produced well-separated stage boundaries across most examined recordings, remained encouraging on noisy data and weakened after epoch shuffling. The accessible abstract does not report participant counts, clinical accuracy metrics or external validation, so the work is best viewed as feature-engineering feasibility.

How to interpret the findings

The abstract provides no cohort denominators, externally held-out validation, clinical decision thresholds or direct evidence that the detected boundaries correspond to patient-important states. Evaluation was limited to meditation and sleep data, and the physiological interpretation and performance across broader EEG paradigms remain open questions.

Equipment and disclosures

EEG datasets and an open-source TopoEEG analysis pipeline were used; no clinical device or sponsor-equipment attribution is made.

Supported by Russian Science Foundation grant 24-68-00030. The authors declared no competing interests and state that source code is available publicly while raw data are available on request.

Related evidence

Compare findings from the same clinical area, including mixed results.