TowardsTrustworthyAutomaticDiagnosisSystemsby EmulatingDoctors'ReasoningwithDeep ReinforcementLearning