Two-step adversarial debiasing with partial learning -- medical image\n case-studies

Abstract

The use of artificial intelligence (AI) in healthcare has become a very\nactive research area in the last few years. While significant progress has been\nmade in image classification tasks, only a few AI methods are actually being\ndeployed in hospitals. A major hurdle in actively using clinical AI models\ncurrently is the trustworthiness of these models. More often than not, these\ncomplex models are black boxes in which promising results are generated.\nHowever, when scrutinized, these models begin to reveal implicit biases during\nthe decision making, such as detecting race and having bias towards ethnic\ngroups and subpopulations. In our ongoing study, we develop a two-step\nadversarial debiasing approach with partial learning that can reduce the racial\ndisparity while preserving the performance of the targeted task. The\nmethodology has been evaluated on two independent medical image case-studies -\nchest X-ray and mammograms, and showed promises in bias reduction while\npreserving the targeted performance.\n

Ähnliche Arbeiten