TCANet for motor imagery EEG classification

TCANet: A Temporal Convolutional Attention Network for Motor Imagery EEG Decoding [Paper]

core idea: Multi-scale CNN + TCN + multi-head self-atttention

Our research builds upon and improves the CTNet and MSCFormer.

Abstract:

Decoding motor imagery electroencephalogram (MI-EEG) signals is fundamental to the development of brain–computer interface (BCI) systems. However, robust decoding remains a challenge due to the inherent complexity and variability of MI-EEG signals. This study proposes the Temporal Convolutional Attention Network (TCANet), a novel end-to-end model that hierarchically captures spatiotemporal dependencies by progressively integrating local, fused, and global features. Specifically, TCANet employs a multi-scale convolutional module to extract local spatiotemporal representations across multiple temporal resolutions. A temporal convolutional module then fuses and compresses these multi-scale features while modeling both short- and long-term dependencies. Subsequently, a stacked multi-head self-attention mechanism refines the global representations, followed by a fully connected layer that performs MI-EEG classification. The proposed model was systematically evaluated on the BCI IV-2a and IV-2b datasets under both subject-dependent and subject-independent settings. In subject-dependent classification, TCANet achieved accuracies of 83.06% and 88.52% on BCI IV-2a and IV-2b respectively, with corresponding Kappa values of 0.7742 and 0.7703, outperforming multiple representative baselines. In the more challenging subject-independent setting, TCANet achieved competitive performance on IV-2a and demonstrated potential for improvement on IV-2b.

Overall Framework:

Dataset & prepare processing

the same as CTNet

Experimental Setup:

The original training set was split into training and validation subsets with a ratio of 8:2. Data augmentation was performed to increase the training set size by 100%.

Comparison of Subject-specific classification accuracy (in %) and kappa on the BCI IV-2a dataset.

Method \ Subject	A01	A02	A03	A04	A05	A06	A07	A08	A09	Average
ShallowConvNet	82.64	56.94	90.97	68.06	71.18	57.64	75.69	82.29	75.69	73.46
DeepConvNet	79.17	51.74	87.85	75.69	76.39	60.07	93.06	79.51	84.38	76.43
EEGNet	85.76	65.28	88.89	69.79	71.18	57.29	74.65	80.90	84.72	75.38
EEGInception	70.14	53.82	70.49	68.40	73.26	53.82	68.75	72.22	68.75	66.63
TSception	62.15	39.58	73.26	54.86	64.93	47.22	59.72	63.19	63.54	58.72
EEGTCNet	79.51	65.97	92.36	69.44	73.96	60.42	85.07	81.94	76.04	76.08
ADFCNN	88.19	60.07	92.01	78.82	70.49	65.97	83.68	84.03	81.60	78.32
MSCFormer	86.46	61.46	93.75	80.90	77.78	69.44	91.32	83.68	78.47	80.36
TCANet (proposed)	88.89	70.14	92.71	79.86	77.78	74.31	92.71	85.76	85.42	83.06

Note: Comparison of experimental results when data augmentation only generates 1 times the original training sample size

Comparison of Subject-specific classification accuracy (in %) and kappa on the BCI IV-2b dataset.

Method \ Subject	A01	A02	A03	A04	A05	A06	A07	A08	A09	Average
ShallowConvNet	71.88	64.29	83.75	96.88	92.81	84.69	90.94	90.94	85.62	84.64
DeepConvNet	79.06	65.71	82.19	97.50	95.31	80.62	91.25	92.19	89.06	85.88
EEGNet	75.94	66.07	85.31	98.44	94.38	84.38	91.25	94.69	87.50	86.44
EEGInception	77.81	66.07	85.62	98.12	98.12	87.81	90.31	95.31	90.00	87.69
TSception	75.31	63.57	75.31	95.00	90.31	74.38	84.38	90.00	80.62	80.99
EEGTCNet	76.88	65.00	84.69	96.88	89.69	86.56	91.88	94.69	85.94	85.80
ADFCNN	79.38	62.86	82.50	97.19	95.31	84.38	91.25	92.50	87.50	85.87
MSCFormer	75.00	68.57	80.00	98.44	95.94	85.00	93.75	94.38	88.44	86.61
TCANet (proposed)	82.50	70.71	86.88	97.81	94.69	87.19	92.50	95.94	88.44	88.52

Comparison of cross-subject classification accuracy (in %) and kappa on the BCI IV-2a & IV-2b datasets.

Method \ Subject	BCI IV-2a	BCI IV-2b
ShallowConvNet	58.64	74.92
DeepConvNet	61.86	76.10
EEGNet	62.44	75.92
EEGInception	58.66	74.64
TSception	50.70	71.67
EEGTCNet	57.91	75.80
ADFCNN	60.40	76.24
MSCFormer	59.39	74.92
TCANet (proposed)	60.98	74.61

Citation

Hope this code can be useful. I would appreciate you citing us in your paper. 😊

Zhao, W., Lu, H., Zhang, B. et al. TCANet: a temporal convolutional attention network for motor imagery EEG decoding. Cogn Neurodyn 19, 91 (2025). https://doi.org/10.1007/s11571-025-10275-5