Data For Predicting Treatment Response Using a Machine Learning Model Based on EEG and Clinical Features

Hypothesis:
The data aimed to develop a machine learning model integrating electroencephalogram (EEG) biomarkers and clinical features to predict the therapeutic response to repetitive transcranial magnetic stimulation (rTMS) in tinnitus patients.
Data Description:
"Supplementary Table 1.xlsx" contains demographic information, event-related potentials (ERPs), frequency band power, and microstate parameters for all participants.
Demographic information: Clinical indicators (age, sex, PTA, tinnitus laterality, duration, VAS, THI, etc.) are included.
Event-related potentials (ERPs): Latencies and amplitudes of the P1, N1, P2, N2, P3 and N3 components were extracted at Fz in response to a 2000 Hz pure tone stimulus under the active attention condition. Latency was defined as the time from stimulus onset to the peak of each maximal deflection. The peak-to-trough amplitude was measured, as well as the average amplitude between the peak and trough and the adjacent peaks and troughs.
Frequency band power: Channels were grouped into seven regions based on brain areas: frontal pole (FP1, FP2, FPz, AF3, AF4), frontal lateral (F1, F2, F3, F4, F5, F6, F7, F8, Fz, F11, F12), fronto-temporal (FC1, FC2, FC3, FC4, FC5, FC6, FT11, FT12), central (Cz, C1, C2, C3, C4, C5, C6), temporal (T7, T8, TP7, TP8), parietal (Pz, P1, P2, P3, P4, P5, P6, P7, P8, CPz, CP1, CP2, CP3, CP4, CP5, CP6), and occipital (O1, O2, Oz, PO3, PO4, PO7, PO8, POz). The following frequency bands were analyzed: delta (0.5-4 Hz), theta (4-8 Hz), alpha1 (8-10 Hz), alpha2 (10-12 Hz), beta1 (12-18 Hz), beta2 (18-25 Hz), beta3 (25-30 Hz), gamma1 (30-49 Hz), and gamma2 (51-80 Hz).
Microstate parameters: The k-means clustering method was used to classify these topographies into four categories (A, B, C, D), representing different EEG activity patterns. Various parameters were calculated and analyzed, including the duration of each microstate (in ms), the occurrence probability of each microstate per second (events/s), the coverage percentage of each microstate across the total time (%), the transition probabilities between microstate categories (TP/%), and the Global Explained Variance (GEV), which represents the percentage of the total EEG signal explained by a particular microstate.
Notable Findings:
Statistical analyses showed significant differences between responders and non-responders in terms of P3 latency (p < 0.001) and delta band power in the temporal lobe during eyes-open resting state (p = 0.009). The machine learning model identifies prolonged P3 and N2 latencies, abnormal low-frequency neural oscillations and pre-treatment THI functional scores as core biomarkers, offering a reliable tool for personalized therapy.