Due to the intra- and inter-individual variability of the electroencephalography (EEG) signals, brain-computer interfaces (BCI) require a daily user-specific calibration. This offline calibration step is necessary to set feature extraction, classification and pre-processing parameters. Yet, it is time consuming and might cause fatigue before the actual use of the BCI. Our goal is to reduce this time with a self-supervised classification method that achieves good detections with minimal calibration trials, for use in a motor imagery (MI)-based BCI that aims to enhance the rehabilitation of stroke patients. To process a small amount of labeled data, self-supervised learning (SSL) is currently the state-of-the-art method in the fields of vision and natural language processing [1], which makes it interesting to explore for EEG data.