Fig 1.
Progression of modeling techniques for epilepsy prediction over the past decade.
Table 1.
EEG lead information.
Table 2.
Comprehensive details of EEG signal characteristics in the CHB-MIT dataset.
Table 3.
Detailed information of EEG signals in SWEC-ETHZ EEG dataset.
Fig 2.
Workflow diagram of the CGTNet model.
Table 4.
EEG data training process.
Fig 3.
CHB-MIT waveforms, illustrating the process of capturing EEG signals through a win-dow of 64 seconds size moving in steps of 32 seconds using the STFT algorithm.
Table 5.
Experimental hardware and software environment.
Table 6.
K-fold cross-validation table.
Fig 4.
Structure of GRU gate control unit.
Fig 5.
Sparse Transformer encoder structure.
Fig 6.
CGTNet model loss curves.
Table 7.
Single-patient 50% discount cross-validation experiments for the CGTNet model based on the CHB-MIT dataset.
Table 8.
Single-patient 50% discount cross-validation results of the CGTNet model using the SWEC-ETHZ dataset.
Table 9.
Comparative analysis of CGTNet and other models on the same task metrics.
Fig 7.
Comparison radar chart between CGTNet and the same task model indicators, where 17 and 18 represent the metric results of CGTNet applied to the CHB-MIT and SWEC-ETHZ datasets, respectively.
Table 10.
Comparison of computational time and performance across different models on the CHB-MIT dataset.
Table 11.
Performance comparison of sparse transformer and traditional transformer on the CHB-MIT dataset.
Fig 8.
Graph of single patient training process and indicator changes.
Fig 9.
Single patient epilepsy prediction confusion matrix results.
Table 12.
CGTNet ablation experiment.