OPTIMUS-Prime: Minimal and Sufficient Concept Explanations for Deep Vision Models