Atrial fibrillation is a cardiac arrhythmia most often originating in the left atrium, causing anarchic electrical activity and thus a loss of atrial contraction. This increases the risk of stroke through clot formation in the atrium, but also of heart failure. Atrial fibrillation is a major cause of stroke, accounting for more than 25% of all strokes. In addition, a quarter of ischemic strokes remain without an obvious cause at the end of hospitalization, and it is recommended that atrial fibrillation be detected intensively with long-term heart rhythm recording. Implantable loop recorders can detect 30% of atrial fibrillation cases over the 3-year battery life of these devices, after a stroke of undetermined origin. However, these devices require a small operation to implant them under the skin, and they are expensive. The hypothesis of this study is that MRI imaging of the left atrium would enable better selection of patients to receive an implantable loop recorder. MRI can quantify the proportion of the left atrium with scar tissue, which is likely to favour the onset of atrial fibrillation. If the results confirm this hypothesis, the number of patients requiring an implantable loop recorder could be reduced, and perhaps an anticoagulation strategy based on MRI data could be introduced. In addition to the usual follow-up by cardiologists and neurologists, participation in this study involves a cardiac MRI (with contrast agent) within 3 months of the stroke.