바로가기 메뉴 본문 바로가기 주메뉴 바로가기
  • 06-1Have you prepared measures to mitigate bias due to human and physical elements in data collection?
    Determine applicability: Consider this question if you are collecting and creating datasets for AI training, and determine if the requirement has been satisfied.

    • Bias due to human elements is a result of conscious or unconscious bias on certain information.
    ✓ Human bias: Includes automation bias, group attribution bias, implicit bias, and in-group bias.

    • In order to avoid human bias, specific collection and inspection criteria for data collection should be established to prevent bias in data attributes by data collectors or rectify bias in the review by having a diverse and sufficient number of reviewers.

    • Physical bias may occur due to problems in the equipment that collects medical data. For instance, imaging data with physical limitations, such as certain colors, brightness, and resolutions, may be collected for each patient depending on the different set values for each X-ray imaging machines. Thus, an error may be present in the future when using datasets collected with other equipment, even if the algorithm was trained using datasets collected with specific equipment.

    • Because this can lead to training biased toward certain races or certain equipment, it is best to remove various human and physical biases and reinforce diversity by using different medical devices in numerous regions and races as much as possible.