作者： Richard J Chen , Ming Y Lu , Tiffany Y Chen , Drew FK Williamson , Faisal Mahmood
DOI:
关键词:
摘要: Background: While artificial intelligence (AI) applications in medicine and healthcare have exploded in the decade, so has increased concern over the vulnerabilities of the software and the regulatory challenges. The training data used for the AI algorithms have come under scrutiny for having large sample-selection biases, such as only having patients from the same race or ethnicity groups or acquiring medical images using only one equipment type, that prevent the model from being successfully deployed in settings sufficiently different from those in which the trained data were acquired. Thus, large, heterogenous, and diverse datasets are thus necessary to develop and refine best practices in evidence-based medicine involving AI. To overcome the paucity of annotated medical data in real-world settings, synthetic data are being increasingly used. Generative adversarial networks (GANs) are a type of generative …