EEG-based brain-computer interfaces (BCIs) are making strides in neurorehabilitation and assistive technologies by creating pathways for communication with external devices. A key challenge in these systems is decoding motor imagery (MI) from EEG signals, which are highly complex and noisy. This study explores the potential of two deep learning models, EEGNet and MS-CRANet, to address this challenge. EEGNet is a lightweight neural network designed specifically for EEG data, leveraging temporal and spatial filters to classify signals efficiently. On the other hand, MS-CRANet, originally developed for video analysis, uses advanced mechanisms like multi-scale convolution, recurrence, and attention to extract both short- and long-term patterns from the data. We tested these models on publicly available MI datasets and found them both to be effective, with MS-CRANet showing better adaptability to complex EEG patterns. EEGNet achieved 68% accuracy, while MS-CRANet slightly outperformed it at 71%. These findings highlight the promise of deep learning in advancing EEG-based BCI technologies.