What is the clinical evidence from this study?

Study design: Other. Population: Stroke (n=8). Intervention: EEGNet vs. Customized CNN and DeepConvNet. Primary outcome: Classification of eight distinct normal and abnormal motor activities (F1-score).

What question did this study set out to answer?

This study evaluates deep learning methods to classify normal and abnormal gait activities in stroke patients using EEG data.

June 19, 2026ComputersOpen Access

EEG-Based Gait Classification in Stroke Patients Using Deep Learning

View Full Paper

Ask AI

Bookmark

Key result

EEGNet achieves a top F1-score of 0.81 for classifying motor activities in stroke patients.

Why the study?

Analyzing complex, high-dimensional EEG signals to detect gait anomalies in stroke rehabilitation remains challenging.

Comparison

Customized CNN vs DeepConvNet vs EEGNet across 32, 22, and 15 EEG channel configurations

Design

Exploratory pilot study

Authors

SKSarunya KanjanawattanaSuranaree University of Technology ISIsaman SangbamrungSuranaree University of Technology DWDulyawat WiriyaphongSuranaree University of Technology

Discussion

Member takes

Implication

Should not yet guide stroke rehabilitation decisions; leaves open EEGNet utility for gait classification pending larger prospective validation.

Key Points

This study evaluates deep learning methods to classify normal and abnormal gait activities in stroke patients using EEG data.
Used EEG signals from eight stroke patients to train and evaluate models: DeepConvNet and EEGNet.
Investigated different channel configurations (32, 22, 15 channels) for optimal performance.
Employing Leave-One-Out Cross-Validation (LOOCV) for model evaluation with seven patients.
EEGNet achieved an average F1-score of 0.810 and an F1-score of 0.915 on an unseen patient.
Exhibited a low false positive rate of 0.175, reducing false alarms.
The 32-channel setup demonstrated highest consistency in gait classification.

Structured PICO

Population

8 stroke patients evaluated using multi-class deep learning frameworks for classifying normal and abnormal motor activities using EEG data.

Exposure

Multi-class deep learning frameworks (customized Convolutional Neural Network [CNN], DeepConvNet, and EEGNet) using EEG data with varying channel reduction configurations (32, 22, and 15 channels)

Outcome

Classification performance of eight distinct normal and abnormal motor activities (measured by F1-score and false positive rate)

EEGNet demonstrated superior performance in differentiating complex gait patterns from EEG signals in stroke patients, highlighting its potential for real-time, non-invasive monitoring in neurorehabilitation.

Limitations

limited cohort

Cite This Study

Kanjanawattana et al. (2026) studied Stroke (n=8). EEGNet vs. Customized CNN and DeepConvNet was evaluated on Classification of eight distinct normal and abnormal motor activities (F1-score). EEGNet achieved the highest descriptive average F1-score of 0.810 for classifying normal and abnormal motor activities in stroke patients.

synapsesocial.com/papers/6a35982fdd3be7785e70ecd8 https://doi.org/10.3390/computers15060392

View Full Paper

Ask AI

Bookmark

Implication

Should not yet guide stroke rehabilitation decisions; leaves open EEGNet utility for gait classification pending larger prospective validation.

EEG-Based Gait Classification in Stroke Patients Using Deep Learning

Discussion

Member takes

Implication

Key Points

Structured PICO

Limitations

Cite This Study

EEG-Based Gait Classification in Stroke Patients Using Deep Learning

Discussion

Member takes

Implication

Key Points

Structured PICO

Limitations

Cite This Study

Also Consider

Also Consider