Figures & diagnostics gallery
Every figure, straight from the reports
53 publication figures reproduced unmodified from reports/. Filter by dataset and topic; open any figure for a zoomable view with its scientific caption. Keyboard: ←→ to step, Esc to close.
Showing 53 of 53 figures
- MindWaveSignal Processing & Quality
Butterworth 1–40 Hz filter response
Magnitude (dB) and phase across 0–60 Hz for the 4th-order zero-phase SOS bandpass. −3 dB cut-offs at 1 Hz and 40 Hz; DC drift suppressed by > 40 dB at 0.2 Hz and mains harmonics by > 30 dB at 50/60 Hz. Identical relative bandwidth at 512 Hz and the decimated 256 Hz rate.
- MindWaveSignal Processing & Quality
Raw vs. filtered FP1 traces
Representative 5-second waveforms for Rest, Medium Math and Hard Math. Filtering removes low-frequency baseline wander without distorting oscillatory bursts such as resting alpha.
- MindWaveSignal Processing & Quality
Per-subject signal-quality summary
Percentage of rejected 1-s chunks per participant under the screening rules. Over 90% of subjects kept a rejection rate below 1%.
- MindWaveNeurophysiology & Behavior
Mean power spectrum per task
Grand-average PSDs (thick) over individual spectra (thin) from 1–45 Hz. Rest shows an 8–12 Hz alpha peak from eyes-closed idling; both math tasks show alpha suppression and elevated theta/beta power.
- MindWaveNeurophysiology & Behavior
Key biomarkers by task
Distributions of six key features across Rest, Medium and Hard Math: strong Rest-vs-Math separation, minimal Medium-vs-Hard separation.
- MindWaveNeurophysiology & Behavior
The eyes-closed alpha confound
Panel A: relative alpha collapses from eyes-closed Rest to eyes-open Medium Math (Δ = −0.1886, d_z = −1.378) and then flatlines between Medium and Hard (Δ = −0.0029, p = 0.2553). Panel B: non-alpha markers (rel_theta, hjorth_complexity) rise progressively with demand.
- MindWaveMachine Learning & Ablation
Confusion matrices — Rest/Math, Medium/Hard, 3-class
Rest vs Math (GroupKFold RF): Rest 62.6% correct, Math 86.8% correct. Medium vs Hard: 43.9% and 64.5% correct — heavy cross-confusion. The 3-class model confuses Medium and Hard extensively.
- MindWaveMachine Learning & Ablation
Data-leakage demonstration
Random chunk-level KFold (leaked) vs subject-grouped GroupKFold (honest). Leakage inflates RF balanced accuracy from 71.78% to 83.12% (+11.33 pp) and LogReg from 71.69% to 76.37% (+4.68 pp).
- MindWaveMachine Learning & Ablation
Within-subject permutation null
200 within-subject label permutations: null mean 0.5084 (SD 0.0039, 95th percentile 0.5145) vs observed 0.7131. Empirical p = 1/201 = 0.0050.
- MindWaveMachine Learning & Ablation
Personal-calibration sample efficiency
First k contiguous chunks per task used for calibration, evaluation strictly on later chunks. k ≤ 20 s degrades performance (−2.4 to −13.6 pp); only k = 40 s yields a net gain (+2.39 pp).
- MindWaveMachine Learning & Ablation
TreeSHAP feature importance
Global ranking over 1,000 evaluation chunks: theta_alpha_beta, theta_beta, rel_theta and hjorth_complexity push toward high load; rel_alpha carries negative SHAP during math (desynchronisation); spec_entropy decreases.
- PhysioNetSignal Processing & Quality
Timestamp interval distribution
Clean 256 Hz raster quantised to alternating 3.0 ms and 4.0 ms intervals, plus the periodic Δt = 0 ms duplicate spike present in A3, A5 and A10.
- PhysioNetSignal Processing & Quality
Participant recording timelines
Continuous EEG coverage across the ~85-minute protocol for all 10 participants: session blocks, 3- and 6-minute rest breaks, manual tags and the reconstructed A8 Session 2 onset.
- PhysioNetSignal Processing & Quality
Session & event alignment error
Sub-sample stimulus-to-EEG alignment error (< 2.0 ms) across participants, and the manual-tag lead (11.2 s) and lag (16.8 s) relative to E-Prime boundary stimuli.
- PhysioNetSignal Processing & Quality
Trial event coverage & epoch usability
Per-participant epoch usability: strictly clean epochs, single-duplicate-sample epochs (A3, A5, A10) and scalp-channel artifact exclusions.
- PhysioNetSignal Processing & Quality
4-channel traces with event markers
Continuous Muse EEG (TP9, AF7, AF8, TP10) with synchronised target/non-target stimulus onsets, response triggers and the 2.0-s epoch window.
- PhysioNetSignal Processing & Quality
Raw vs. Track A / Track B filtered traces
Continuous raw signal against Track A (1–40 Hz, cognitive bandpower) and Track B (0.5–30 Hz, ERP/time-domain) outputs.
- PhysioNetSignal Processing & Quality
Dual-track SOS filter responses
Magnitude and phase for both tracks, with single-pass (~48 dB/oct) and forward-backward (~96 dB/oct) roll-off per transition edge.
- PhysioNetSignal Processing & Quality
Continuous PSD before/after filtering
Welch PSD of the continuous recording showing steep stopband attenuation outside the passband.
- PhysioNetSignal Processing & Quality
Channel quality & fallback gains
Usable rates by channel and the gain from frontal single-channel fallback across the cohort.
- PhysioNetSignal Processing & Quality
Cohort channel usability (97.8% frontal yield)
Retention per scalp channel and overall frontal epoch yield. Mastoid channels suffer temporalis tension; frontal fallback restores 97.8% (13,772 / 14,080) usable trials.
- PhysioNetSignal Processing & Quality
Participant data availability
Total extracted trials (1,408), usable fallback trials and bilateral-intact trials for each participant.
- PhysioNetSignal Processing & Quality
Structural feature missingness
Availability matrix over 500 sampled trials across A1–A10 — e.g. frontal alpha asymmetry is structurally NaN where only one frontal electrode is usable.
- PhysioNetSignal Processing & Quality
200 µV vs 300 µV threshold sensitivity
Relaxing the cut-off from 200 µV to 300 µV adds only 124 trials (+0.88%) and shifts effect sizes by < 0.01 d_z.
- PhysioNetSignal Processing & Quality
Cohort sensitivity: N = 10 vs N = 7 effect sizes
Excluding the duplicate-glitch participants (A3, A5, A10) preserves 100% of effect directions, with |Δd_z| < 0.18.
- PhysioNetSignal Processing & Quality
Cohort sensitivity: overlaid distributions
Feature distributions of the primary (N = 10) and sensitivity (N = 7) cohorts overlaid.
- PhysioNetNeurophysiology & Behavior
Accuracy by music condition (fixed order)
Cohort accuracy across the four sessions: 36.51% → 40.99% → 45.26% → 51.79%. Because every participant ran the same order, this rise is collinear with practice and cannot be attributed to music.
- PhysioNetNeurophysiology & Behavior
Reaction time by music condition
Correct-response latency stays roughly stable (~382–397 ms) across the four conditions; music main effect on RT n.s. (p = 0.1513).
- PhysioNetNeurophysiology & Behavior
1-back vs 3-back behavioural contrast
Participant-level paired lines: d′ drops from 1.817 to 1.110 (F(1,9) = 17.14, p = 0.0025, η²p = 0.656) and accuracy from 48.89% to 38.38% under 3-back load.
- PhysioNetNeurophysiology & Behavior
Participant-wise performance
Individual accuracy across all four conditions — mean accuracy ranges from 3.85% (A6) to 90.77% (A1), highlighting cohort heterogeneity.
- PhysioNetNeurophysiology & Behavior
Speed–accuracy trade-off
Reaction time vs accuracy across conditions and participants.
- PhysioNetNeurophysiology & Behavior
Omission rates & RT distribution
Per-participant omission rates (A4 75.9%, A6 96.2%, A10 95.7%) and the aggregate RT histogram over all valid presses.
- PhysioNetNeurophysiology & Behavior
Practice curve vs flat EEG across sessions
Behavioural accuracy rises across the four fixed-order sessions while EEG rhythms stay flat (no condition difference survives FDR, all q ≥ 0.90) — the signature of a practice effect, not a music effect.
- PhysioNetNeurophysiology & Behavior
Paired theta rise, 1-back → 3-back
Participant-level paired slopes for EEG features. TP10_rel_theta d_z = +1.26 (p = 0.0039 uncorrected). Exploratory: zero of 42 features survive BH-FDR (q ≥ 0.164).
- PhysioNetNeurophysiology & Behavior
Frontal theta vs working-memory sensitivity
Between-subject scatter: mean frontal theta vs d′, Spearman ρ = −0.8788 (p = 0.0008, N = 10). Trait-level association only — within-subject fluctuations do not track performance (|r_rm| < 0.20).
- PhysioNetNeurophysiology & Behavior
Frontal EEG features by music condition
Boxplots of frontal features across the four auditory conditions. No spectral difference survives FDR correction.
- PhysioNetNeurophysiology & Behavior
Frontal bandpower by music condition
Frontal bandpower distributions under No Music, Relaxing, Stressing and AI Music.
- PhysioNetNeurophysiology & Behavior
Frontal bandpower by N-back level
Frontal theta, alpha and beta across 0-, 1-, 2- and 3-back levels.
- PhysioNetNeurophysiology & Behavior
Stimulus-locked epochs, 1-back vs 3-back
Representative 2.0-s stimulus-locked waveforms under low vs high working-memory load.
- PhysioNetNeurophysiology & Behavior
Pre- vs post-stimulus features
Feature distributions for the active pre-stimulus window (−500–0 ms, anticipation under continuous music — not a resting baseline) vs the post-stimulus window.
- PhysioNetNeurophysiology & Behavior
Individual differences in frontal rhythms
Frontal theta and alpha distributions per participant (A1–A10) — the between-subject spread that defeats generic models.
- PhysioNetNeurophysiology & Behavior
Frontal feature correlation matrix
Spearman rank correlations across 17 core frontal rhythm features.
- PhysioNetMachine Learning & Ablation
Phase 3C generic model performance
Dummy, Logistic Regression and Random Forest balanced accuracy under participant-grouped GroupKFold, hovering at the 50% chance line.
- PhysioNetMachine Learning & Ablation
Phase 3C confusion matrices (held-out participants)
Confusion matrices on held-out participants under GroupKFold.
- PhysioNetMachine Learning & Ablation
Generic vs personalized accuracy
Generic All (51.2%), Generic Condition-Matched (50.0%) and Personalized (55.6%) RF balanced accuracy on identical held-out blocks (p = 0.0025, d_z = 1.31, Holm p_adj = 0.0075).
- PhysioNetMachine Learning & Ablation
Personalization gain by participant
Grouped bars per participant — 9 of 10 improve under personalized calibration.
- PhysioNetMachine Learning & Ablation
Per-participant calibration delta
Waterfall of ΔBalAcc = personalized − generic per participant (largest: A7 +9.99, A4 +8.19, A1 +7.50, A3 +6.65 pp; smallest: A2 +0.38 pp).
- PhysioNetMachine Learning & Ablation
Calibration budget scaling
Accuracy across 10/20/30/50% budgets under original partitions and on the fixed common evaluation set (Sessions 3–4): 52.19% → 54.68% → 55.77% → 57.06%. An observed trend, not a proven optimum.
- PhysioNetMachine Learning & Ablation
Calibration trials vs held-out accuracy
Sample-efficiency curve of calibration trial count against held-out accuracy on the common evaluation set.
- PhysioNetMachine Learning & Ablation
Feature-group ablation
Ranking of feature subsystems under personalized calibration: band_ratios first (56.25%), erp last at chance (50.26%).
- PhysioNetMachine Learning & Ablation
Normalization strategy comparison
Raw (53.5%) vs z-score (55.7%) vs MinMax (56.0%), with scalers fit strictly on calibration blocks.
- PhysioNetMachine Learning & Ablation
Feature-selection stability
How often each top feature is selected across participant calibration sets.
- PhysioNetMachine Learning & Ablation
Phase 4 confusion matrices
Side-by-side confusion matrices for generic vs personalized classifiers on held-out blocks.
No figures match these filters.