Text this: Assessing the adversarial robustness of multimodal medical AI systems: insights into vulnerabilities and modality interactions