An evaluation of the impact of different types of AI on clinical decision making when reviewing medical images

AI is being used increasingly in image interpretation tasks. There are challenges for its optimal use in reporting environments. Human reliance on technology and bias can cause decision errors. Previous studies have demonstrated that trust issues exist amongst radiologists and radiographers in both over-reliance (automation bias) and reluctance in AI use for decision support.

AI feedback can be provided in many different forms, such as heat maps, AI binary, textual description, regions of interest, the AI quality (percentage) or the AI interpretation with clinical codes to allow for automated patient journey. These many forms of AI feedback exist and there remains a variation in the use and provision of the forms in clinical departments. Understanding which form of AI feedback proves most useful will promote good decision hygiene and the use of AI in the clinical setting.  

This proposed study aims to determine the most useful/preferred format of AI feedback and to provide this to aid the image interpretation process. This aims to improve the integration of AI into the decision making process and to optimise the performance of reporting clinicians.

Aim: To explore the most useful forms of AI to aid the clinical decision making process.

Objectives of the research: To determine the impact of EUI and the type of AI information provided on

-the decision switching and thought processes.

-the accuracy and automation bias.

-the user performance and confidence of prediction.

To investigate the user preference of AI type and explanation user interface (EUI).

To co-design, develop and test the use of an EUI.

Methods:

Ethical approval will be sought from Ulster University Research and Ethics Filter Committee. Participants will be asked to interpret images with the assistance of a form of AI feedback to determine its use and impact on the clinical decision making process.