Understanding human-AI interaction in medical imaging: moving towards 'appropriate reliance'
Year: 2025
Type: conference-paper
Abstract: It has been suggested that the combination of human-AI in medical decision-making could foster a ‘hybrid intelligence’, producing outcomes superior to either mind alone. However, this is dependent upon the effective integration of AI output into the decision-making process and human observers are frequently over/under-reliant upon AI. The fundamental purpose of my research is to better understand factors affecting human-AI interaction in medical imaging to help promote ‘appropriate reliance’ upon AI. My first study explored the impact of observer experience and attitudes towards AI technology upon performance with interactive AI during a pulmonary nodule detection task. The results revealed that AI use led to improvements in performance for first and third-year radiography students but had minimal impact upon the performance of reporting radiographers. Eye tracking data revealed that AI prompt salience could rapidly pull observer attention towards the nodule but failed to capture attention entirely. Observer attitudes towards AI technology did not appear to have any moderating effect upon performance with AI. My second study will explore the impact of confirmation bias upon AI-assisted decision making. Specifically, the study will seek to address how AI agreement with an observer’s prior hypothesis across the initial ten cases reviewed affects subsequent reliance upon AI.
Source: Insight (University of Cumbria)
Authors: Davies, Sarah, Donovan, Tim
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7082098826