Motor imagery classification based on electroencephalography (EEG) signals is one of the most important brain-computer interface applications, although it needs further improvement. Several methods have attempted to obtain useful information from EEG signals by using recent deep learning techniques such as transformers. To improve the classification accuracy, this study proposes a novel EEG-based motor imagery classification method with three key features: generation of a topological map represented as a two-dimensional image from EEG signals with coordinate transformation based on t-SNE, use of the InternImage to extract spatial features, and use of spatiotemporal pooling inspired by PoolFormer to exploit spatiotemporal information concealed in a sequence of EEG images. Experimental results using the PhysioNet EEG Motor Movement/Imagery dataset showed that the proposed method achieved the best classification accuracy of 88.57%, 80.65%, and 70.17% on two-, three-, and four-class motor imagery tasks in cross-individual validation.