Artificial Intelligence (Al) has brought significant breakthroughs to the field of Medical Imaging, thanks to the high quantity of clinical data generated by different institutions and the inductive ability of Convolutional Neural Networks (CNNs). Nevertheless, Medical Imaging is a domain with a large variability degree (for example, different acquisition protocols, diverse health centers, etc.), often leading to poor generability when deploying algorithms to slightly different domains from their training distribution. The Medical Image Computing division in DKFZ, led by Prof. Klaus Maier-Hein, has developed an algorithm for vessel occlusion detection from contrast-enhanced Computer Tomography (CT) images known as "angiographies" (https://www.nature.com/articles/s41467-023-40564-8). Several clinics in Germany have been involved in the project, enabling access to a main dataset and two smaller sets from external clinics. The algorithm provides good performance in the external sets, but still, a degradation in performance occurs when moving from internal to external test sets. Within my Ph.D., I want to investigate techniques that help guarantee model robustness and generalization while identifying out-of-distribution samples (domain adaptation, active learning, foundation model fine-tuning, etc.). Data from more clinics will be made available in the near future, so having a robust algorithm against potential domain shifts is crucial. Additionally, I also want to validate my approaches on other medical image datasets.