<p>Due to the limitations in the accuracy and robustness of current EEG classification methods, applying motor imagery for practical Brain-Computer Interface applications remains challenging. Therefore, an EEG classification method with high accuracy and strong robustness is of significant importance. This paper proposed a method called 3D CNN and LSTM for Motor Imagery (3D-CLMI), which combines 3D CNN and LSTM network with attention to classify MI-EEG signals. This method combined MI-EEG signals from different channels into 3D features and extracted spatial features through convolution operations with multiple 3D convolutional kernels of different scales. At the same time, in order to ensure the integrity of the extracted temporal features of the MI-EEG signal, 3D-CLMI adopted a parallel structure to obtain spatial and temporal features respectively, and then combined the obtained features for classification. Experimental results showed that this method achieved a classification accuracy of 92.7% and an F1-score of 0.91 on BCI Competition IV 2a, which were both higher than the state-of-the-art methods in the field of MI tasks. Additionally, 12 participants were invited to complete a four-class MI task, and experiments on the collected dataset showed that our method also maintained the highest classification accuracy and F1-score. Our proposed method achieved the best results on both datasets, and we then demonstrated the effectiveness of each part of the proposed method through ablation experiments. Additionally, we designed a rehabilitation application system in a VR environment based on the proposed method, and the experimental results validated that it could assist patients with impaired hand motor function.</p>