OSCR

SSDLabeler: realistic semi-synthetic data generation for multi-label artifact classification in EEG.

Code ↔ Paper

8 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 8 matches · 2 of them tie a paragraph to a whole file, not to given lines: weak matches, whose lines are not tinted
  1. [1] § Methods › Semi-synthetic data generation (SSDLabeler) ↔ SSDLabeler/SSDLabeler.m, lines 1–88 · score 0.86 · brain ICs, raw EEG recordings, clean EEG epochs, raw EEG epochs, artifact components, multi artifact
  2. [2] § Methods › Semi-synthetic data generation (SSDLabeler) ↔ SSDLabeler/SSDLabeler.m, lines 1–88 · score 0.71 · artifact presence, ICLabel, SSDLabeler, confident, Raw EEG, ICs
  3. [3] § Methods › SSD labeler for artifact annotation › SSDLabeler hyperparameter optimization ↔ SSDLabeler/main.m, the whole file · a weak match · score 0.63 · PSD verification, probability threshold, ICLabel, SSDLabeler, motor, hyperparameters
  4. [4] § Methods › Model › Model architecture and losses ↔ EEG_Artifact_Classification/artifact_classification/module/eeg_artifact_classification.py, lines 158–166 · score 0.58 · binary cross entropy, loss, predictions, classification
  5. [5] § Methods › SSD labeler for artifact annotation ↔ SSDLabeler/main.m, the whole file · a weak match · score 0.56 · PSD verification, PSD thresholding, events, labeler, RMS, channel
  6. [6] § Methods › Dataset and preprocessing ↔ EEG_Artifact_Classification/main_EEGArtifact.py, lines 42–110 · score 0.55 · cross validation, artifact classification, subset, training, model, raw
  7. [7] § Methods › ICA and artifact IC selection ↔ SSDLabeler/SSDLabeler.m, lines 123–131 · score 0.53 · mixing matrix, clean EEG, reconstructing, brain, heart, signals
  8. [8] § Methods › Model ↔ EEG_Artifact_Classification/main_EEGArtifact.py, lines 42–110 · score 0.51 · EEG encoder, artifact classification, module, model, trained, SSD

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

MATLAB · 473 lines · 19 KB · CC-BY-SA-4.0 · 3 matches

  1. function [SSD, contamination_label_all] = SSDLabeler(EEG_data, fs, target_fs, locs, EEG_channels, ...
  2. event_idx, trial_mixing, prob_threshold, mix_artifact, thresh_use, clean_use, ...
  3. thresh_e, thresh_h, thresh_l, thresh_c, thresh_o, thresh_m)
  5. % SSDLabeler
  6. % -------------------------------------------------------------------------
  7. % Generate semi-synthetic EEG datasets (SSD) using ICA + ICLabel-derived
  8. % artifact component signals, plus RMS/PSD-based epoch-level artifact masks.
  9. %
  10. % The function:
  11. % 1) Filters + epochs raw EEG around event indices
  12. % 2) Runs PCA+FastICA, then ICLabel to identify IC classes (A)
  13. % 3) Reconstructs:
  14. % - "clean" EEG from Brain ICs
  15. % - artifact-only signals (Eye/Muscle/Heart/Line/Channel/Other ICs)
  16. % 4) Computes epoch-level artifact presence masks:(B)
  17. % - Eye/Heart/Line/Channel/Other: Standardized RMS thresholds
  18. % - Muscle: PSD-template correlation threshold
  19. % 4.1) Mixes artifacts together (C)
  20. % 5) Generates mixed trials according to one of the supported modes:
  21. % A) SSDLabel (A+B+C): add artifact signals only when that epoch is
  22. % flagged by RMS/PSD masks (multi-artifact per epoch allowed)
  23. % B) Single-artifact (A+B): generate separate datasets per artifact
  24. % type (one artifact type at a time), only when that epoch is flagged
  25. % C) A+C: add all artifact signals regardless of masks (debug/ablation)
  26. %
  27. % Outputs:
  28. % SSD : [N x nChannels x nSamples] dataset
  29. % contamination_label_all : [N x 7] labels [clean, eye, muscle, heart, line, channel, other]
  30. %
  31. % -------------------------------------------------------------------------
  32. % Inputs
  33. % -------------------------------------------------------------------------
  34. % EEG_data : [nChannels x nTime] continuous EEG recording (raw or minimally processed).
  35. % NOTE: channels should correspond to 'locs' and 'EEG_channels'.
  36. %
  37. % fs : Original sampling rate (Hz) of EEG_data (e.g., 500).
  38. %
  39. % target_fs : Target sampling rate (Hz) after resampling (e.g., 125).
  40. % trial_length_sec * target_fs determines the samples per epoch.
  41. %
  42. % locs : EEGLAB-style channel locations struct array (chanlocs).
  43. % Used by ICLabel for topography features.
  44. %
  45. % EEG_channels : Vector of channel indices used for ICA/ICLabel (e.g., 1:19).
  46. % This should match the channels in EEG_data that are EEG (exclude EOG if desired).
  47. %
  48. % event_idx : Vector of sample indices (in fs) indicating epoch start points
  49. % (e.g., movement event onset indices). Each event creates a
  50. % trial_length_sec-long epoch if it fits in EEG_data.
  51. %
  52. % trial_mixing : Integer controlling trial mixing replication.
  53. % In the current method, ONLY trial_mixing == 0 is supported
  54. % (because only the 3 requested generation modes are enabled).
  55. %
  56. % prob_threshold : ICLabel confidence threshold in [0,1].
  57. % If max class probability for an IC < prob_threshold, that IC is treated as "unassigned"
  58. % (i.e., not included in brain or artifact groups via thresholding).
  59. %
  60. % mix_artifact : Boolean selecting generation mode:
  61. % - true : multi-artifact mixing modes (SSDLabel or A+C)
  62. % - false : single-artifact-per-dataset mode (A+B)
  63. %
  64. % thresh_use : Boolean selecting whether RMS/PSD epoch masks are respected:
  65. % - true : SSDLabel (A+B+C) or mode (A+B) mode uses masks to decide which artifacts
  66. % are injected into each epoch
  67. % - false : A+C mode injects ALL artifact signals for every epoch (debug/ablation)
  68. %
  69. % clean_use : Boolean controlling final dataset assembly:
  70. % - true : SSD = [clean_EEG_epochs; raw_epochs; generated_mixed_epochs]
  71. % labels = [clean_label; artifact_label; mix_label]
  72. % - false : SSD = [raw_epochs; generated_mixed_epochs]
  73. % labels = [artifact_label; mix_label]
  74. %
  75. % thresh_e : Percentile (0–100) for Eye RMS z-score thresholding.
  76. % Higher => fewer epochs labeled as Eye artifact (more conservative).
  77. %
  78. % thresh_h : Percentile (0–100) for Heart RMS z-score thresholding.
  79. %
  80. % thresh_l : Percentile (0–100) for Line-noise RMS z-score thresholding.
  81. %
  82. % thresh_c : Percentile (0–100) for Channel-noise RMS z-score thresholding.
  83. %
  84. % thresh_o : Percentile (0–100) for Other RMS z-score thresholding.
  85. %
  86. % thresh_m : Muscle PSD-template correlation threshold (typically in [-1, 1], often ~0.6–0.9).
  87. % Epochs with corr(Pxx_avg, template_PSD) < thresh_m are labeled as Muscle artifact.
  88. % -------------------------------------------------------------------------
  90. %% -------------------- Params --------------------
  91. trial_length_sec = 3;
  92. lp = 1;
  93. hp = 50;
  94. order = 6600;
  95. powerline_freq = 60;
  97. %% -------------------- 1) Filter + detrend --------------------
  98. EEG_data = filter_notch(EEG_data, fs, powerline_freq);
  99. EEG_data = filter_data(EEG_data, fs, lp, hp, order);
  100. EEG_data = detrend(EEG_data')';
  102. %% -------------------- 2) Epoch --------------------
  103. [epochs3d, ~] = epoch_by_events(EEG_data, event_idx, trial_length_sec*fs);
  104. % epochs3d: [ch x samp x trials]
  105. [num_channels, num_samples_fs, num_trials] = size(epochs3d);
  107. % ICA expects [ch x time]
  108. EEG_ica = reshape(epochs3d, num_channels, num_samples_fs*num_trials);
  110. %% -------------------- 3) ICA (PCA whitening + fastICA) --------------------
  111. [mixing_matrix, unmixing_matrix, IC] = run_fastica_with_pca(EEG_ica, fs);
  113. %% -------------------- 4) ICLabel --------------------
  114. class_prob = run_iclabel(EEG_ica, fs, locs, EEG_channels, mixing_matrix, unmixing_matrix);
  115. thresh_class_index = iclabel_threshold_classes(class_prob, prob_threshold);
  116. icIdx = ic_indices_from_thresh(thresh_class_index);
  118. brain_ic_index = icIdx.brain;
  119. if isempty(brain_ic_index)
  120. warning('No brain IC recovered! Please check.')
  121. end
  123. %% -------------------- 5) Build clean EEG + per-artifact reconstructed signals (at fs) --------------------
  124. EEG_clean_fs = mixing_matrix(:, brain_ic_index) * IC(brain_ic_index, :);
  126. Signal_fs.eye = mixing_matrix(:, icIdx.eye) * IC(icIdx.eye, :);
  127. Signal_fs.muscle = mixing_matrix(:, icIdx.muscle) * IC(icIdx.muscle, :);
  128. Signal_fs.heart = mixing_matrix(:, icIdx.heart) * IC(icIdx.heart, :);
  129. Signal_fs.line = mixing_matrix(:, icIdx.line) * IC(icIdx.line, :);
  130. Signal_fs.channel = mixing_matrix(:, icIdx.channel) * IC(icIdx.channel, :);
  131. Signal_fs.other = mixing_matrix(:, icIdx.other) * IC(icIdx.other, :);
  133. %% -------------------- 6) Resample + reshape to [trials x ch x samples] --------------------
  134. num_samples_target = trial_length_sec * target_fs;
  136. EEG_epochs = to_trials(EEG_clean_fs, num_channels, num_samples_target, num_trials, fs, target_fs);
  138. Signal_trials.eye = to_trials(Signal_fs.eye, num_channels, num_samples_target, num_trials, fs, target_fs);
  139. Signal_trials.muscle = to_trials(Signal_fs.muscle, num_channels, num_samples_target, num_trials, fs, target_fs);
  140. Signal_trials.heart = to_trials(Signal_fs.heart, num_channels, num_samples_target, num_trials, fs, target_fs);
  141. Signal_trials.line = to_trials(Signal_fs.line, num_channels, num_samples_target, num_trials, fs, target_fs);
  142. Signal_trials.channel = to_trials(Signal_fs.channel, num_channels, num_samples_target, num_trials, fs, target_fs);
  143. Signal_trials.other = to_trials(Signal_fs.other, num_channels, num_samples_target, num_trials, fs, target_fs);
  145. % Artifact-added (for RMS/PSD detection)
  146. EEG_plus.eye = EEG_epochs + Signal_trials.eye;
  147. EEG_plus.muscle = EEG_epochs + Signal_trials.muscle;
  148. EEG_plus.heart = EEG_epochs + Signal_trials.heart;
  149. EEG_plus.line = EEG_epochs + Signal_trials.line;
  150. EEG_plus.channel = EEG_epochs + Signal_trials.channel;
  151. EEG_plus.other = EEG_epochs + Signal_trials.other;
  153. %% -------------------- 7) RMS-based detection for eye/heart/line/channel/other --------------------
  154. rms_clean_mean = epoch_rms_mean(EEG_epochs);
  156. z_eye = (epoch_rms_mean(EEG_plus.eye) - rms_clean_mean) ./ std(rms_clean_mean);
  157. z_heart = (epoch_rms_mean(EEG_plus.heart) - rms_clean_mean) ./ std(rms_clean_mean);
  158. z_line = (epoch_rms_mean(EEG_plus.line) - rms_clean_mean) ./ std(rms_clean_mean);
  159. z_chan = (epoch_rms_mean(EEG_plus.channel) - rms_clean_mean) ./ std(rms_clean_mean);
  160. z_other = (epoch_rms_mean(EEG_plus.other) - rms_clean_mean) ./ std(rms_clean_mean);
  162. eye_artifact_idx = z_eye > prctile(z_eye, thresh_e);
  163. heart_artifact_idx = z_heart > prctile(z_heart, thresh_h);
  164. line_artifact_idx = z_line > prctile(z_line, thresh_l);
  165. channel_artifact_idx = z_chan > prctile(z_chan, thresh_c);
  166. other_artifact_idx = z_other > prctile(z_other, thresh_o);
  168. %% -------------------- 8) Muscle detection with PSD-template correlation --------------------
  169. muscle_artifact_idx = muscle_psd_template_mask(EEG_plus.muscle, target_fs, thresh_m);
  171. %% -------------------- 9) Build masks struct --------------------
  172. artifactMasks.eye = eye_artifact_idx;
  173. artifactMasks.muscle = muscle_artifact_idx;
  174. artifactMasks.heart = heart_artifact_idx;
  175. artifactMasks.line = line_artifact_idx;
  176. artifactMasks.channel = channel_artifact_idx;
  177. artifactMasks.other = other_artifact_idx;
  179. %% -------------------- 10) Build "epochs_all" (resampled original) --------------------
  180. epochs_all = to_trials(reshape(epochs3d, num_channels, []), num_channels, num_samples_target, num_trials, fs, target_fs);
  182. %% -------------------- 11) Generation stage (ONLY 3 MODES) --------------------
  183. if ~(trial_mixing == 0)
  184. error('Only trial_mixing == 0 is supported in this refactor (per your requested modes).');
  185. end
  187. if thresh_use == true && mix_artifact == true
  188. % SSDLabel: A+B+C
  189. [trial_mix_data, trial_mix_label] = gen_SSDLabel_ABC(EEG_epochs, Signal_trials, artifactMasks);
  191. elseif mix_artifact == false
  192. % A+B: single-artifact mixing
  193. [trial_mix_data, trial_mix_label] = gen_singleArtifact_AB(EEG_epochs, Signal_trials, artifactMasks);
  195. elseif thresh_use == false && mix_artifact == true
  196. % A+C: no RMS or PSD verification
  197. [trial_mix_data, trial_mix_label] = gen_allArtifacts_AC(EEG_epochs, Signal_trials, artifactMasks);
  199. else
  200. error('Unsupported configuration. Only the 3 requested generation modes are enabled.');
  201. end
  203. %% -------------------- 12) Labels for clean + artifact (non-mixed) --------------------
  204. clean_label = make_clean_labels(size(EEG_epochs,1));
  205. artifact_label = make_artifact_labels(size(EEG_epochs,1), artifactMasks);
  207. %% -------------------- 13) Concatenate outputs --------------------
  208. if clean_use
  209. contamination_label_all = [clean_label; artifact_label; trial_mix_label];
  210. SSD = [EEG_epochs; epochs_all; trial_mix_data];
  211. else
  212. contamination_label_all = [artifact_label; trial_mix_label];
  213. SSD = [epochs_all; trial_mix_data];
  214. end
  216. end
  218. %% ========================================================================
  219. % Helpers
  220. % ========================================================================
  222. function [epochs3d, valid_mask] = epoch_by_events(EEG, event_idx, epoch_samples)
  223. n_ch = size(EEG,1);
  224. epochs3d = [];
  225. valid_mask = false(size(event_idx));
  226. for i = 1:numel(event_idx)
  227. idx = event_idx(i);
  228. if idx + epoch_samples - 1 <= size(EEG,2)
  229. epochs3d = cat(3, epochs3d, EEG(:, idx:(idx+epoch_samples-1)));
  230. valid_mask(i) = true;
  231. end
  232. end
  233. if isempty(epochs3d)
  234. epochs3d = zeros(n_ch, epoch_samples, 0);
  235. end
  236. end
  238. function [A, W, IC] = run_fastica_with_pca(X, fs)
  239. [dewhitening, ~, latent] = pca(X','Centered','off');
  240. whitening = inv(dewhitening);
  241. PC = whitening * X;
  243. latent = 100 * latent / sum(latent);
  244. n_pcs = sum(latent > 0.01*mean(latent));
  246. [~, mixing, unmixing] = fastica(PC(1:n_pcs,:), 'approach','defl','g','tanh','maxNumIterations',fs);
  248. IC = unmixing * PC(1:n_pcs,:);
  250. A = dewhitening(:,1:n_pcs) * mixing;
  251. W = unmixing * whitening(1:n_pcs,:);
  252. end
  254. function class_prob = run_iclabel(EEG_ica, fs, locs, EEG_channels, A, W)
  255. data_len = size(EEG_ica,2);
  256. t = linspace(0, data_len/fs, data_len);
  258. EEG_struct = struct();
  259. EEG_struct.times = t;
  260. EEG_struct.data = EEG_ica;
  261. EEG_struct.chanlocs = locs;
  262. EEG_struct.srate = fs;
  263. EEG_struct.trials = 1;
  264. EEG_struct.pnts = data_len;
  265. EEG_struct.icawinv = A;
  266. EEG_struct.icaweights = W;
  267. EEG_struct.icaact = W * EEG_ica;
  268. EEG_struct.icachansind = EEG_channels;
  269. EEG_struct.ref = 'averef';
  271. EEG_struct = iclabel(EEG_struct);
  272. class_prob = EEG_struct.etc.ic_classification.ICLabel.classifications;
  273. end
  275. function thresh_class_index = iclabel_threshold_classes(class_prob, prob_threshold)
  276. [nIC, ~] = size(class_prob);
  277. thresh_class_index = zeros(1, nIC);
  278. for i = 1:nIC
  279. [mx, idx] = max(class_prob(i,:));
  280. if mx >= prob_threshold
  281. thresh_class_index(i) = idx;
  282. else
  283. thresh_class_index(i) = 0;
  284. end
  285. end
  286. end
  288. function icIdx = ic_indices_from_thresh(thresh_class_index)
  289. icIdx.brain = find(thresh_class_index == 1);
  290. icIdx.muscle = find(thresh_class_index == 2);
  291. icIdx.eye = find(thresh_class_index == 3);
  292. icIdx.heart = find(thresh_class_index == 4);
  293. icIdx.line = find(thresh_class_index == 5);
  294. icIdx.channel = find(thresh_class_index == 6);
  295. icIdx.other = find(thresh_class_index == 7);
  296. end
  298. function Xtr = to_trials(Xfs, n_ch, n_samp, n_trials, fs, target_fs)
  299. Xrs = resample(Xfs', target_fs, fs)'; % [ch x (samples_target*trials)]
  300. Xtr = reshape(Xrs, n_ch, n_samp, n_trials);
  301. Xtr = permute(Xtr, [3 1 2]); % [trials x ch x samp]
  302. end
  304. function m = epoch_rms_mean(X)
  305. m = squeeze(mean(sqrt(mean(X.^2, 3)), 2)); % [trials x 1]
  306. end
  308. function muscle_mask = muscle_psd_template_mask(EEG_plus_muscle, target_fs, corr_threshold)
  309. [epoch_num, channel_num, ~] = size(EEG_plus_muscle);
  310. frequencies = 1:50;
  311. template_PSD = exp(-1 * (1:length(frequencies)) / 20)';
  313. PSD_corr = zeros(epoch_num, 1);
  314. for i = 1:epoch_num
  315. epoch_data = squeeze(EEG_plus_muscle(i,:,:)); % [ch x samp]
  316. Pxx_all = zeros(length(frequencies), channel_num);
  317. for ch = 1:channel_num
  318. Pxx_all(:, ch) = pwelch(epoch_data(ch,:), [], [], frequencies, target_fs);
  319. end
  320. Pxx_avg = mean(Pxx_all, 2);
  321. PSD_corr(i) = corr(Pxx_avg, template_PSD);
  322. end
  324. muscle_mask = PSD_corr < corr_threshold;
  325. end
  327. function L = make_clean_labels(n)
  328. L = zeros(n,7);
  329. L(:,1) = 1;
  330. end
  332. function L = make_artifact_labels(n, masks)
  333. L = zeros(n,7);
  334. L(masks.eye, 2) = 1;
  335. L(masks.muscle, 3) = 1;
  336. L(masks.heart, 4) = 1;
  337. L(masks.line, 5) = 1;
  338. L(masks.channel, 6) = 1;
  339. L(masks.other, 7) = 1;
  341. no_art = sum(L(:,2:7),2) < 1;
  342. L(no_art,1) = 1;
  343. end
  345. %% ========================================================================
  346. % ONLY 3 GENERATION FUNCTIONS
  347. % ========================================================================
  349. function [trial_mix_data, trial_mix_label] = gen_SSDLabel_ABC(EEG_epochs, Signal_trials, masks)
  350. % thresh_use==true, trial_mixing==0, mix_artifact==true
  352. disp('SSDLabel')
  354. nTrials = size(EEG_epochs,1);
  355. trial_mix_data = [];
  356. trial_mix_label = [];
  358. for i = 1:nTrials
  359. mixed_trial = squeeze(EEG_epochs(i,:,:));
  361. if masks.eye(i)
  362. mixed_trial = mixed_trial + squeeze(Signal_trials.eye(i,:,:));
  363. end
  364. if masks.muscle(i)
  365. mixed_trial = mixed_trial + squeeze(Signal_trials.muscle(i,:,:));
  366. end
  367. if masks.heart(i)
  368. mixed_trial = mixed_trial + squeeze(Signal_trials.heart(i,:,:));
  369. end
  370. if masks.line(i)
  371. mixed_trial = mixed_trial + squeeze(Signal_trials.line(i,:,:));
  372. end
  373. if masks.channel(i)
  374. mixed_trial = mixed_trial + squeeze(Signal_trials.channel(i,:,:));
  375. end
  376. if masks.other(i)
  377. mixed_trial = mixed_trial + squeeze(Signal_trials.other(i,:,:));
  378. end
  380. temp_label = [0, masks.eye(i), masks.muscle(i), masks.heart(i), masks.line(i), masks.channel(i), masks.other(i)];
  381. if sum(temp_label) == 0
  382. temp_label = [1,0,0,0,0,0,0];
  383. end
  385. trial_mix_data = cat(1, trial_mix_data, reshape(mixed_trial, [1, size(mixed_trial,1), size(mixed_trial,2)]));
  386. trial_mix_label = [trial_mix_label; temp_label];
  387. end
  389. end
  391. function [trial_mix_data, trial_mix_label] = gen_singleArtifact_AB(EEG_epochs, Signal_trials, masks)
  392. % trial_mixing==0, mix_artifact==false
  394. nTrials = size(EEG_epochs,1);
  395. trial_mix_data = [];
  396. trial_mix_label = [];
  398. for art = 2:7
  400. if art == 2 && sum(masks.eye) < 1, continue; end
  401. if art == 3 && sum(masks.muscle) < 1, continue; end
  402. if art == 4 && sum(masks.heart) < 1, continue; end
  403. if art == 5 && sum(masks.line) < 1, continue; end
  404. if art == 6 && sum(masks.channel) < 1, continue; end
  405. if art == 7 && sum(masks.other) < 1, continue; end
  407. for i = 1:nTrials
  408. mixed_trial = squeeze(EEG_epochs(i,:,:));
  409. temp_label = [0,0,0,0,0,0,0];
  411. if art == 2 && masks.eye(i)
  412. mixed_trial = mixed_trial + squeeze(Signal_trials.eye(i,:,:));
  413. temp_label = [0,1,0,0,0,0,0];
  415. elseif art == 3 && masks.muscle(i)
  416. mixed_trial = mixed_trial + squeeze(Signal_trials.muscle(i,:,:));
  417. temp_label = [0,0,1,0,0,0,0];
  419. elseif art == 4 && masks.heart(i)
  420. mixed_trial = mixed_trial + squeeze(Signal_trials.heart(i,:,:));
  421. temp_label = [0,0,0,1,0,0,0];
  423. elseif art == 5 && masks.line(i)
  424. mixed_trial = mixed_trial + squeeze(Signal_trials.line(i,:,:));
  425. temp_label = [0,0,0,0,1,0,0];
  427. elseif art == 6 && masks.channel(i)
  428. mixed_trial = mixed_trial + squeeze(Signal_trials.channel(i,:,:));
  429. temp_label = [0,0,0,0,0,1,0];
  431. elseif art == 7 && masks.other(i)
  432. mixed_trial = mixed_trial + squeeze(Signal_trials.other(i,:,:));
  433. temp_label = [0,0,0,0,0,0,1];
  434. end
  436. if sum(temp_label) == 0
  437. temp_label = [1,0,0,0,0,0,0];
  438. end
  440. trial_mix_data = cat(1, trial_mix_data, reshape(mixed_trial, [1, size(mixed_trial,1), size(mixed_trial,2)]));
  441. trial_mix_label = [trial_mix_label; temp_label];
  442. end
  443. end
  445. end
  447. function [trial_mix_data, trial_mix_label] = gen_allArtifacts_AC(EEG_epochs, Signal_trials, masks)
  448. % thresh_use==false, trial_mixing==0, mix_artifact==true
  450. nTrials = size(EEG_epochs,1);
  451. trial_mix_data = [];
  452. trial_mix_label = [];
  454. for i = 1:nTrials
  455. mixed_trial = squeeze(EEG_epochs(i,:,:));
  457. mixed_trial = mixed_trial + squeeze(Signal_trials.eye(i,:,:));
  458. mixed_trial = mixed_trial + squeeze(Signal_trials.muscle(i,:,:));
  459. mixed_trial = mixed_trial + squeeze(Signal_trials.heart(i,:,:));
  460. mixed_trial = mixed_trial + squeeze(Signal_trials.line(i,:,:));
  461. mixed_trial = mixed_trial + squeeze(Signal_trials.channel(i,:,:));
  462. mixed_trial = mixed_trial + squeeze(Signal_trials.other(i,:,:));
  464. temp_label = [0, masks.eye(i), masks.muscle(i), masks.heart(i), masks.line(i), masks.channel(i), masks.other(i)];
  465. if sum(temp_label) == 0
  466. temp_label = [1,0,0,0,0,0,0];
  467. end
  469. trial_mix_data = cat(1, trial_mix_data, reshape(mixed_trial, [1, size(mixed_trial,1), size(mixed_trial,2)]));
  470. trial_mix_label = [trial_mix_label; temp_label];
  471. end
  473. end

SSDLabeler.m at commit 10158c5, under CC-BY-SA-4.0 · at the source

Overview

Authors: Taketo Akama1, Akima Connelly1,2, Shun Minamikawa1, Natalia Polouliakh1
  1. Sony Computer Science Laboratories, Inc., Tokyo, Japan
  2. Institute of Science Tokyo, Tokyo, Japan
Institutions: Sony Computer Science Laboratories (Japan); Institute of Science Tokyo (Japan)
Journal: Scientific reports, volume 16, issue 1, article 25980
Dates: received 3 March 2026; accepted 28 May 2026; published online 8 June 2026
Type: Research article · Language: English
License: CC BY-NC-ND
Identifiers: DOI 10.1038/s41598-026-56070-y · PMID 42259834 · PMCID PMC13487299 · OpenAlex W4417142610
Open access: gold, a free copy (OpenAlex)
Preprint: arxiv:2512.05500
Status: code verified
Categories: EEG (modality), human (organism)
Methods: Spectral & time-frequency, Preprocessing, Connectivity, Smoothing, state filtering, decompositions, Machine learning, Physiology & signal measures
Keywords: EEG, Semi-synthetic data, Artifact, Multi-label classification, Biological techniques, Computational biology and bioinformatics, Engineering, Mathematics and computing, Neuroscience
MeSH: Artifacts*, Electroencephalography*, Signal Processing, Computer-Assisted*, Algorithms, Humans (* major topic)
Topic: EEG and Brain-Computer Interfaces (Cognitive Neuroscience, Neuroscience), according to OpenAlex
Citations: not cited yet (Europe PMC); 32 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 8 matches between paragraphs and lines of code.

AkimaCon/SSDLabeler

License: CC-BY-SA-4.0
State: the link answers, verified on 27 September 2026
Evidence: files inventoried
Commit: 10158c5989af6d7f8abba8d067b715f667bede6b, 5 June 2026
Languages: Python (11), MATLAB (2), Jupyter (1), Shell (1)
Size: 25 files, 15 scripts
Software Heritage: not archived
Found in: “Code availability”
Holds: README, license file, environment (EEG_Artifact_Classification/requirements.txt), 1 notebook
Not found: CITATION.cff, tests, continuous integration, documentation
Tools: PyTorch (7 files), NumPy (6 files), pandas (4 files), scikit-learn (3 files), PyTorch Lightning (2 files), Matplotlib (2 files), EEGLAB (1 file), ICLabel (1 file), Signal Processing Toolbox (1 file), Statistics and Machine Learning Toolbox (1 file), seaborn (1 file), statsmodels (1 file)
Availability: 1 check, the latest on 27 September 2026: the link answers
  • 27 September 2026: the link answers
17 files
At the source: github.com/AkimaCon/SSDLabeler

Code availability statement

The paper has a code availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

Read it in the paper: doi.org/10.1038/s41598-026-56070-y.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

  • 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 15 scripts, each with its path and the digest of its content;
  • 8 matches between paragraphs of the paper and lines of the code (method lexical-v1);
  • neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

Datasets cited

Data availability statement

The paper has a data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

Read it in the paper: doi.org/10.1038/s41598-026-56070-y.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 4 authors, 9 keywords, 5 MeSH terms, 31 references.

Cite

This paper

Akama, T., Connelly, A., Minamikawa, S., & Polouliakh, N. (2026). SSDLabeler: realistic semi-synthetic data generation for multi-label artifact classification in EEG. Scientific reports, 16(1), 25980. https://doi.org/10.1038/s41598-026-56070-y

BibTeX

@article{akama2026ssdlabeler,
author = {Akama, Taketo and Connelly, Akima and Minamikawa, Shun and Polouliakh, Natalia},
title = {{SSDLabeler: realistic semi-synthetic data generation for multi-label artifact classification in EEG}},
journal = {Scientific reports},
year = {2026},
month = jun,
volume = {16},
number = {1},
pages = {25980},
publisher = {Nature Publishing Group},
issn = {2045-2322},
doi = {10.1038/s41598-026-56070-y},
url = {https://doi.org/10.1038/s41598-026-56070-y},
pmid = {42259834},
pmcid = {PMC13487299}
}

RIS

TY - JOUR
AU - Akama, Taketo
AU - Connelly, Akima
AU - Minamikawa, Shun
AU - Polouliakh, Natalia
TI - SSDLabeler: realistic semi-synthetic data generation for multi-label artifact classification in EEG
T2 - Scientific reports
J2 - Sci Rep
PY - 2026
DA - 2026/06/08
VL - 16
IS - 1
SP - 25980
SN - 2045-2322
PB - Nature Publishing Group
DO - 10.1038/s41598-026-56070-y
UR - https://doi.org/10.1038/s41598-026-56070-y
LA - en
ER -

CSL-JSON

{
"id": "10.1038/s41598-026-56070-y",
"type": "article-journal",
"title": "SSDLabeler: realistic semi-synthetic data generation for multi-label artifact classification in EEG",
"container-title": "Scientific reports",
"author": [
{
"family": "Akama",
"given": "Taketo"
},
{
"family": "Connelly",
"given": "Akima"
},
{
"family": "Minamikawa",
"given": "Shun"
},
{
"family": "Polouliakh",
"given": "Natalia"
}
],
"container-title-short": "Sci Rep",
"volume": "16",
"issue": "1",
"page": "25980",
"DOI": "10.1038/s41598-026-56070-y",
"PMID": "42259834",
"PMCID": "PMC13487299",
"ISSN": "2045-2322",
"publisher": "Nature Publishing Group",
"URL": "https://doi.org/10.1038/s41598-026-56070-y",
"language": "en",
"issued": {
"date-parts": [
[
2026,
6,
8
]
]
}
}

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.3390/s26134019 [code]
NeuroStat: An Open-Source EEG Connectivity Platform for Randomised Controlled Trials.
Journal: Sensors (Basel, Switzerland)
In common: ICLabel, pandas, Matplotlib, 1 other tool, physionet.org/content/eegmmidb, EEG, 3 references
[2] doi:10.1002/hbm.70528 [code]
Explainable AI Insights Into EEG Classification and Its Alignment to Neural Correlates.
Journal: Human brain mapping
In common: ICLabel, PyTorch, seaborn, 4 other tools, EEG, 3 references
[3] doi:10.7554/elife.107088 [code]
Development of auditory and spontaneous movement responses to music over the first postnatal year.
Journal: eLife
In common: ICLabel, EEGLAB, Signal Processing Toolbox, 6 other tools, EEG, 1 reference
[4] doi:10.1038/s41597-026-07807-x [code]
A Multimodal fNIRS-EEG Dataset for Unilateral Limb Motor Imagery.
Journal: Scientific data
In common: ICLabel, EEGLAB, Signal Processing Toolbox, 3 other tools, EEG, 2 references
[5] doi:10.1038/s42003-026-10957-8 [code]
Brain defence by the extracellular matrix protein Cochlin.
Journal: Communications biology
In common: PyTorch Lightning, Signal Processing Toolbox, statsmodels, 7 other tools
[6] doi:10.1016/j.isci.2026.115375 [code]
Bursts of regional cortical inhibition during smartphone use.
Journal: iScience
In common: ICLabel, EEGLAB, Signal Processing Toolbox, 5 other tools, 2 references
[7] doi:10.1093/bioinformatics/btag169 [code]
Bidirectional Mamba-2 boosts EEG super-resolution via regression and diffusion.
Journal: Bioinformatics (Oxford, England)
In common: PyTorch Lightning, PyTorch, scikit-learn, 3 other tools, physionet.org/content/eegmmidb, EEG
[8] doi:10.1002/mds.70348 [code]
Electroencephalography-Based Clustering Reveals Robust Neurophysiological Subtypes in Parkinson's Disease.
Journal: Movement disorders : official journal of the Movement Disorder Society
In common: ICLabel, EEGLAB, statsmodels, 5 other tools, EEG, 1 reference
[9] doi:10.3390/bios16070394
Hybrid Edge-Cloud Asymmetric Analytics for Portable Multimodal BCI Biosensors.
Journal: Biosensors
In common: physionet.org/content/eegmmidb, EEG, 4 references
[10] doi:10.1162/imag.a.1269 [code]
From early to contemporary normative modeling: Mapping individual differences in neurophysiological signals.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: ICLabel, statsmodels, seaborn, 4 other tools, EEG, 2 references

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.