Detecting algorithmic bias in medical AI-models