Traditional task-specific computational pathology models require a substantial labeled dataset for training to perform various tasks, while foundation models can be trained on large-scale, unlabeled ...