Text this: From Accuracy to Vulnerability: Quantifying the Impact of Adversarial Perturbations on Healthcare AI Models