“What Would Need to Change?” Counterfactual Explanations as a Tool for Interpretable and Actionable Medical AI