The onset detection of epileptic seizures from multivariate Electroencephalogram (EEG) data is a challenging task. The variation in seizure patterns across patients and epilepsy types makes it particularly difficult to create a generic solution. Existing approaches indicate low recall due to their inability to capture complex seizure onset patterns. In this paper, we propose VISOR – a novel approach to detect the onset of epileptic seizures based on novel patient profiles and visual, personalized feature representations. VISOR leverages a vision transformer model to learn the spatio-temporal relationships between features, capture individual seizure propagation patterns, and perform seizure onset detection in a heterogeneous multi-patient dataset. Evaluation on a real-world dataset demonstrates that VISOR outperforms state-of-the-art baselines by at least 5% points for seizure onset detection in terms of the F1 score and indicates higher effectiveness for more complex patterns of propagating seizures.