cs.LGOct 20, 2026

EEG-Based Classification of Hypnotic Susceptibility Using Two-Dimensional Convolutional Neural Networks

Authors: Krzysztof Wrobel, Grzegorz M. Wojcik, Filip Postepski, Andrzej Kawiak, Katarzyna Zemla, Grzegorz Sedek

Organizations: Department of Neuroinformatics and Biomedical Engineering, Institute of Computer Science, Maria Curie-Skłodowska University, Akademicka 9, 20-031 Lublin, Poland · Institute of Psychology, SWPS University, Chodakowska 19/31, 03-815 Warsaw, Poland

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Guided Imagery (GI) is a mind-based relaxation technique widely used in clinical and therapeutic settings. Objectively assessing whether a patient has entered a relaxed state remains a challenge for clinicians. This study investigates electroencephalography (EEG)-based classification of hypnotic susceptibility using two-dimensional Convolutional Neural Networks (2D CNNs) applied to minimally preprocessed signals, without manual feature extraction. Data were collected from 62 male participants (32 Control, 30 Hypno) using a 256-channel dense array amplifier during Guided Imagery or cognitive task conditions. A sliding-window framework evaluated up to 868 independent models across seven anatomically defined brain regions and 124 ten-second time windows. Segment-level random-split screening was followed by Leave-One-Subject-Out (LOSO) cross-validation to obtain subject-independent performance estimates, and an ensemble approach combined predictions across K = 10 windows. The best single-window model achieved accuracy of 0.82 on the random split. Under LOSO, anterior regions generalised better across subjects: the frontal region achieved mean accuracy of 0.646 (AUC: 0.692). An ensemble of K = 10 windows further improved the prefrontal region to mean accuracy of 0.738. The occipital region, dominant in random-split screening, collapsed to near-chance under LOSO, revealing subject-specific EEG fingerprinting rather than genuine group differences. Bandpass filtering and z-score normalisation suffice for competitive EEG classification without manual feature extraction. Anterior cortical regions are identified as the most generalisable neural markers of hypnotic susceptibility.

Figures & tables

Figure 1: Top: overview of the EEG Laboratory at UMCS, Lublin, Poland. Bottom: EGI 256-channel EEG cap.
RegionNNElectrode identifiers (sorted)
Prefrontal22E2, E3, E10–E12, E18–E20, E25–E27, E31–E34, E37–E39, E46–E48, E222
Frontal20E4–E6, E13–E15, E21–E23, E28–E30, E35, E36, E40, E41, E214, E215, E223, E224
Temporal72E1, E49, E54–E58, E61–E64, E67–E75, E82, E83, E91–E95, E102–E104, E173, E178, E180, E181, E189–E195, E200–E204, E208–E213, E216–E221, E225–E229, E232, E233, E250–E256
Central25E7–E9, E16, E17, E24, E42–E44, E50–E52, E59, E65, E182–E184, E186, E195–E198, E205–E207
Parietal41E45, E53, E60, E66, E77–E81, E84–E90, E96–E101, E129–E132, E141–E144, E152–E155, E161–E164, E170, E171, E179
Occipital40E105–E119, E121–E128, E138–E140, E149–E151, E158–E160, E167–E169, E175–E177, E188, E199
Table 1: Electrode subsets per brain region in the 256-channel EGI GSN 130 cap. Electrode identifiers follow the EGI convention (E n ); consecutive identifiers are shown as ranges. NN is the number of electrodes per region.
Figure 2: 2D-CNN architecture used for HIPNO/CONTROL EEG classification.
Brain RegionDiscriminative windowsProportion (%)Peak ACCMean ACC
Occipital112/12490.30.80260.7206
Parietal95/12476.60.77300.7087
Prefrontal24/12419.40.82240.7086
Central20/12416.10.80260.6998
Frontal13/12410.50.78290.7105
Temporal0/1240.0——
Table 2: Classification performance of the CNN model across EEG brain regions. Discriminative windows are those exceeding the 0.67 accuracy threshold. Total number of analysed windows per region: 124.
Figure 3: Number of discriminative windows (ACC >0.67>0.67 ) per brain region out of 124 total analysed windows.
Figure 4: Peak and mean classification accuracy per brain region for regions with at least one discriminative window.
Figure 5: Proportion (%) of discriminative windows (ACC >0.67>0.67 ) out of 124 total windows per brain region.
RegionWindow (s)ACCF1PrecisionRecallAUC
Occipital771–7810.7660.7600.7890.7320.848
Parietal1121–11310.7990.8000.8030.7970.899
Prefrontal1101–11110.7530.7670.7320.8040.829
Central1101–11110.6880.6780.7040.6540.757
Frontal1121–11310.7660.7490.8150.6930.855
Table 3: Classification metrics for the best-performing window per discriminative region (HIPNO as positive class). Values are obtained by re-training the identified best window with a fixed random seed; run generate_best_metrics.py to reproduce.
Figure 6: Normalised confusion matrices for the best-performing window per discriminative brain region. Values represent row-normalised proportions; raw counts are shown in parentheses. Rows correspond to true labels; columns to predicted labels.
Figure 7: ROC curves for the five discriminative brain regions, computed from the best-performing window per region. Line styles distinguish regions for monochrome reproduction. AUC values are shown in the legend.
RegionWindow (s)Mean ACC ±\pm StdF1PrecisionRecallAUC
Occipital771–7810.453±0.2350.453\pm 0.2350.4340.4350.4340.429
Parietal1121–11310.631±0.2280.631\pm 0.2280.6270.6130.6410.680
Prefrontal1101–11110.624±0.3530.624\pm 0.3530.6160.6080.6240.681
Central1101–11110.591±0.2430.591\pm 0.2430.5820.5760.5880.599
Frontal1121–11310.646±0.3020.646\pm 0.3020.6380.6330.6410.692
Table 4: LOSO cross-validation results for the best window per discriminative region. ACC is mean ±\pm std across 62 folds; F1, Precision, Recall, and AUC are computed on concatenated fold predictions (HIPNO as positive class).
Figure 8: Per-subject accuracy distributions from LOSO cross-validation for each discriminative brain region. Each box represents the distribution of 62 fold accuracies. The dashed line marks the 0.67 discriminative threshold.
Figure 9: Aggregated normalised confusion matrices from LOSO cross-validation. Values are computed from the concatenated predictions of all 62 folds. Row-normalised proportions are shown; raw counts in parentheses.
Figure 10: Aggregated ROC curves from LOSO cross-validation, computed from concatenated fold predictions. AUC values are shown in the legend.
RegionSingle ACCMean ACC ±\pm StdΔ\Delta ACCF1PrecisionRecallAUC
Prefrontal0.6240.738±0.3610.738\pm 0.361++ 0.1140.7220.7440.7010.761
Frontal0.6460.667±0.3470.667\pm 0.347++ 0.0210.6350.6750.6000.723
Central0.5910.629±0.3380.629\pm 0.338++ 0.0380.5840.6370.5390.629
Parietal0.6310.609±0.3520.609\pm 0.352−- 0.0220.5640.6120.5220.639
Occipital0.4530.462±0.3520.462\pm 0.352++ 0.0090.4290.4410.4170.472
Table 5: Ensemble LOSO results with K = 10 windows (soft voting). Single-window LOSO accuracy is included as reference. Δ\Delta ACC is relative to the single-window baseline. ACC is mean ±\pm std across 62 folds; F1, Precision, Recall, and AUC are computed on concatenated fold predictions (HIPNO as positive class).
Figure 11: Per-subject accuracy distributions from ensemble LOSO cross-validation (K = 10) for each discriminative brain region. The dashed line marks the 0.67 discriminative threshold.
Figure 12: Aggregated normalised confusion matrices from ensemble LOSO cross-validation (K = 10), computed from concatenated predictions across all 62 folds. Row-normalised proportions are shown; raw counts in parentheses.
Figure 13: Aggregated ROC curves from ensemble LOSO cross-validation (K = 10), computed from concatenated fold predictions. AUC values are shown in the legend.
More in this topic

Classification

Classification maps observations to discrete labels, including familiar, unknown, hierarchical, or multiple categories.

See all 413 papers

Explore similar work

CardsList
Oct 20, 2026

Region-Specific EEG Classification of Hypnotic Susceptibility Using a Sliding-Window 2D CNN: A Subject-Independent Evaluation

We investigate whether a 2D Convolutional Neural Network (CNN) trained on minimally preprocessed EEG signals can classify hypnotic susceptibility across distinct brain regions. Sixty-two male participants (32 Control, 30 Guided Imagery / Hipno group) underwent a 20-minute EEG recording using a 256-channel EGI cap. A sliding-window approach evaluates all 10-second windows within the recording for each of seven predefined brain regions (prefrontal, frontal, temporal, central, parietal, occipital, and facial), yielding up to 868 independently trained models. Initial screening with a segment-level 90/10 random split showed that five regions exceeded the 0.67 accuracy threshold, with the occipital region achieving the highest mean accuracy (0.721) across 90.3% of windows. Subject-independent evaluation via Leave-One-Subject-Out (LOSO) cross-validation, however, substantially revised this picture: occipital performance collapsed to near-chance (mean ACC 0.453, AUC 0.429), revealing that high random-split accuracy was driven by subject-specific EEG fingerprints rather than genuine group differences. In contrast, the frontal and prefrontal regions maintained above-chance LOSO accuracy (0.646 and 0.624, respectively), demonstrating genuine generalisability. These results underscore the necessity of subject-independent validation in EEG-based classification studies and highlight anterior cortical regions as more reliable neural markers of hypnotic susceptibility.

Krzysztof Wrobel, Grzegorz M. Wojcik

Department of Neuroinformatics and Biomedical Engineering, Institute of Computer Science, Maria Curie-Skłodowska University, Lublin, Poland

ClassificationCross-Subject EEG DecodingEEG Decoding
AskPDF
May 5, 2026

Spatiotemporal Convolutions on EEG signal -- A Representation Learning Perspective on Efficient and Explainable EEG Classification with Convolutional Neural Nets

Classification of EEG signals using shallow Convolutional Neural Networks (CNNs) is a prevalent and successful approach across a variety of fields. Most of these models use independent one-dimensional (1D) convolutional layers along the spatial and temporal dimensions, which are concatenated without a non-linear activation layer between. In this paper, we investigate an alternative encoding that operates a bi-dimensional (2D) spatiotemporal convolution. While 2D convolutions are numerically identical to two concatenated 1D convolutions along the two dimensions, the impact on learning is still uncertain. We test 1D and 2D CNNs and a CNN+transformer hybrid model in a low-dimensional (3-channel) and a high-dimensional (22-channel) BCI motor imagery classification task. We observe that 2D convolutions significantly reduce training time in high-dimensional tasks while maintaining performance. We investigate the root of this improvement and find no difference in spectral feature importance. However, a clear pattern emerges in representational similarity across models: 1D and 2D models yield vastly different representational geometries. Overall, we suggest an improved model with a 2D convolutional layer for faster training and inference. We also highlight the importance of architecturally-driven encoding when processing complex multivariate signals, as reflected in internal representations rather than purely in performance metrics.

Laurits Dixen, Stefan Heinrich, Paolo Burelli

IT University of Copenhagen Denmark

Efficient Neural Network InferenceMI ClassificationEEG DecodingConvolutional Neural Networks
AskPDF
May 19, 2026

Classification of IED-free EEG Responses for Assisted Epilepsy Diagnosis

Diagnosing epilepsy is challenging when routine EEGs lack interictal epileptiform discharges (IEDs). Intermittent photic stimulation (IPS) and hyperventilation (HV) can increase diagnostic yield, but their interpretation is subjective. We propose a reproducible pipeline that classifies EEG recordings acquired during stimulation procedures, using machine-learning features spanning temporal, spectral, wavelet, and connectivity domains, and a stacked ensemble to combine complementary feature sets. Performance is evaluated with leave-one-subject-out (LOSO) cross-validation on the TUH Epilepsy Corpus and a clinical Erasmus MC (EMC) cohort, including IED-free analyses on TUH. On TUH, ensembles achieve up to 97.8% AUC / 93.1% BAC on IED-free resting-state EEG and 94.1% AUC / 86.8% BAC on IED-free IPS. On EMC, IPS provides the strongest discrimination (79.4% AUC / 73.9% BAC), while HV performance benefits from stratifying subjects by responsiveness. These results indicate that stimulation-evoked activity, particularly IPS, contains meaningful discriminative information for IED-free epilepsy classification and that multi-domain ensembling improves robustness.

Giacomo Zanardini, Ryan Moesman, Paul van der Kleij +2

Signal Processing Systems, Delft University of Technology, Delft, The Netherlands · Department of Neurology, Erasmus Medical Center, Rotterdam, The Netherlands

ElectroencephalographyEEG DecodingBrain Disorder Classification
AskPDF