An experimental study to investigate the impact of different forms of AI feedback on automation bias and decision switching on diagnostic radiographers.

Artificial intelligence (AI) continues to assert itself in healthcare, with applications including diagnosis of pathology on radiographic images. The NHS recognises AI technologies in their Long-Term Plan (2019) and as a way to ensure the effective continuation of diagnostic provision (NHS (Richards Report) 2021). Computer technology processing ability has enabled application of increased sophistication of AI in healthcare mimicking the way the human brain functions through ‘learning’ from experience. These systems have proven to have high accuracy in detection of abnormality on radiographic images ((Islam et al. 2017, Qin et al. 2018, Guan et al, 2018), but are not currently fully utilised in imaging due, largely, to trust issues with the system in a landscape where error carries significant weight. A balance between due caution and optimal use is not yet established. Through increasing integration into the clinical setting, awareness of unique AI biases are becoming clear, such as the human reliance on the machine causing the user to question their initial decision or change their mind (Goddard et al., 2014; Bond et al., 2018) This study aims to investigate the impact of binary and visual feedback (heatmaps) on radiographers of different ages, familiarity with technology and confidence in diagnosis by determining the propensity to change their mind from their initial interpretation. 212 wds