• Author
    Jessica Tsang
  • Discovery PI

    Hannah Milch, MD

  • Project Co-Author

    Nina Capiro, MD, Sarah Ameri, MD

  • Abstract Title

    Breast radiologists’ perceptions of AI implementation for screening mammography: a qualitative focus group study

  • Discovery AOC Petal or Dual Degree Program

    Basic, Clinical, & Translational Research

  • Abstract

    Background: Growth and development of new artificial intelligence (AI) tools have the potential to address critical healthcare gaps such as physician shortages, physician burnout, patient wait-times, etc. Efficient and ethical implementation of any AI developed for use in healthcare requires a comprehensive understanding of the interaction between the specific AI system and its intended users within the real-world clinical context.

    Objective: This qualitative study is the first in the US to utilize radiologist focus groups to evaluate the user-AI interaction between breast radiologists and an AI tool developed for screening mammography in order to optimize future implementation and integration policies.

    Methods: Focus groups were conducted virtually and comprised of breast radiologists (n=31) from six sites across the US. Groups were also stratified by pre-implementation (n=16) vs post-implementation (n=15) experience with the AI tool. Sessions were moderated with semi-structured scripts and qualitative analysis of transcripts was done by a team of three (1 breast radiologist, 1 radiology fellow, and 1 medical student) using iterative immersion-crystallization method.

    Results (in finalizing stage): Several notable themes and sub-themes were identified across groups regarding perceptions of the AI on screening mammography; including 1) General attitude of cautious optimism about AI being used as a support tool; 2) Trust in the AI being buildable and maintainable with more data, time, and personal experiences; 3) Impacts to workflow, with potential for improved efficiency but concerns about increased cognitive load/burnout; 4) Ethical and medicolegal concerns, specifically regarding liability and equity in access; 5) Factors that impact the adoption of the AI, including system integration, disruptions to established workflows, cost, user-interface optimization, and patient perceptions; 6) Appropriate training needs, with suggestions for identification of specific strengths and limitations of the AI, and case-based trainings. Additionally, stratification by pre-implementation vs post-implementation allowed for comparisons between expectations vs real-world experiences: 7) Pre-implementation users were more skeptical about the AI tool than post-implementation users, who reported higher confidence specifically with low-risk cases.

Back