Is Dataset Quality Still a Concern in Diagnosis Using Large Foundation Model?