TurboRhythmPat and Directed Rhythm Lobish: An Explainable Multi-Selector EEG Model for Recognising Attack and Post-Attack States in Multiple Sclerosis

Background/Objectives: Multiple sclerosis (MS) advances through attacks and partial recoveries, yet electroencephalography (EEG) has hardly been used to recognise an attack from the period that follows it, and most EEG models neither test unseen patients nor show which rhythm and electrode pair shaped a decision. Methods: We present an explainable model built around TurboRhythmPat, a feature extractor that codes each 15 s, 19-channel segment as 1805 addressable source–rhythm–target routes in five EEG rhythms. Four iterative feature selectors and their six pairwise intersections form ten selector branches, a self-organised k-nearest-neighbour classifier (tkNN) labels every branch, and iterative majority voting fuses the branch outputs. Directed Rhythm Lobish (DRLobish) decodes the routes behind the selected output into rhythm and electrode-pair statements. We analysed 57 recordings from 19 patients recorded during and after an attack and from 19 healthy controls. Results: Under fully nested leave-one-subject-out validation, in which every data-driven choice was made within the training participants, the attack and post-attack recordings of unseen patients were separated with 65.61% segment-level and 76.32% recording-level accuracy, and patient-level accuracy exceeded chance in 13 of 19 patients (Wilcoxon p = 0.0022). The between-person tasks did not exceed chance (37.92%, 52.51% and 33.17%), and the non-nested estimates (85.71% for attack versus post-attack) were inflated by model selection on the test participants. A permutation-based maxT test over all 1805 routes found no route after family-wise correction (smallest adjusted p = 0.076). Conclusions: Because the attack recording always came first, treatment timing was not documented, no dedicated artefact correction was applied, and the controls were older, the differences are not interpreted as relapse-related changes, and larger prospective cohorts are needed before clinical use.

Authors

Institutions

Publication Details

Journal
Diagnostics
Published
2026-11-19
DOI
https://doi.org/10.3390/diagnostics16223888
Primary Topic
EEG and Brain-Computer Interfaces
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

TurboRhythmPat and Directed Rhythm Lobish: An Explainable Multi-Selector EEG Model for Recognising Attack and Post-Attack States in Multiple Sclerosis

Şengül Doğan, Mehmet Bayğın, İrem Taşçı, Türker Tuncer et al.
Diagnostics
EEG and Brain-Computer Interfaces
article

TurboRhythmPat and Directed Rhythm Lobish: An Explainable Multi-Selector EEG Model for Recognising Attack and Post-Attack States in Multiple Sclerosis

Şengül Doğan, Mehmet Bayğın, İrem Taşçı, Türker Tuncer, Burak Taşçı, Caner Feyzi Demir, Bilal Ay
article en

Abstract

Background/Objectives: Multiple sclerosis (MS) advances through attacks and partial recoveries, yet electroencephalography (EEG) has hardly been used to recognise an attack from the period that follows it, and most EEG models neither test unseen patients nor show which rhythm and electrode pair shaped a decision. Methods: We present an explainable model built around TurboRhythmPat, a feature extractor that codes each 15 s, 19-channel segment as 1805 addressable source–rhythm–target routes in five EEG rhythms. Four iterative feature selectors and their six pairwise intersections form ten selector branches, a self-organised k-nearest-neighbour classifier (tkNN) labels every branch, and iterative majority voting fuses the branch outputs. Directed Rhythm Lobish (DRLobish) decodes the routes behind the selected output into rhythm and electrode-pair statements. We analysed 57 recordings from 19 patients recorded during and after an attack and from 19 healthy controls. Results: Under fully nested leave-one-subject-out validation, in which every data-driven choice was made within the training participants, the attack and post-attack recordings of unseen patients were separated with 65.61% segment-level and 76.32% recording-level accuracy, and patient-level accuracy exceeded chance in 13 of 19 patients (Wilcoxon p = 0.0022). The between-person tasks did not exceed chance (37.92%, 52.51% and 33.17%), and the non-nested estimates (85.71% for attack versus post-attack) were inflated by model selection on the test participants. A permutation-based maxT test over all 1805 routes found no route after family-wise correction (smallest adjusted p = 0.076). Conclusions: Because the attack recording always came first, treatment timing was not documented, no dedicated artefact correction was applied, and the controls were older, the differences are not interpreted as relapse-related changes, and larger prospective cohorts are needed before clinical use.

DiagnosticsVol. 16(22)
Fırat University (TR), Erzurum Technical University (TR)
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon