Fig 1.
An overview of proposed Dual-Branch Deepwalk-Transformer Spatiotemporal Fusion Network (Deepwalk-TS) for seizure detection.
The upper shows the spatial branch of Deepwalk-based GCN used to reconstruct the channel graph. The lower shows the temporal branch of Guided-CNN Transformer for sequence feature extraction.
Fig 2.
An illustration of the Deepwalk algorithm operation involves a random walk on the graph, resulting in a “walk” sequence, which is then utilized to construct a graph.
Fig 3.
An illustration of the data preprocessing workflow.
Table 1.
Results of CHB-MIT single patient experiments.
Table 2.
Results of single-patient experiments on the Sina dataset.
Table 3.
Comparison with existing methods on the CHB-MIT dataset.
Fig 4.
ROC curve comparing our method with advanced performance methods.
Fig 5.
Example of the process of constructing a map of multichannel EEGs.
Fig 6.
t-SNE visualization of the original and model-embedded distributions of the test data.
Different blue and orange represent the characteristic sample points of normal and epileptic seizures, respectively. (a) Original data distribution. (b) Distribution of feature embeddings for model prediction results.
Fig 7.
Accuracy and loss curves for DeepWalk-TS model during the training and validation process.
Fig 8.
Visual histogram of ablation test performance.