E1605. Satisfaction with AI: A New Source of Interpretation Error
Authors
Li-Hsiang Yen;
University of Rochester
Background
Multiple sources of radiologic interpretation errors have been described in literature. With the recent advances in artificial intelligence (AI) algorithm, a new potential source of error is introduced. This educational exhibit will show several cases that were missed by the AI algorithm and raise radiologist's awareness for this potential error.
Educational Goals / Teaching Points
A new potential source of radiologic interpretation error, satisfaction with AI, is being introduced. When the AI algorithm interprets a study as negative, it is very important for the interpreting radiologist to know that AI algorithm may miss findings and avoid being mislead by the AI algorithm.
Key Anatomic/Physiologic Issues and Imaging Findings/Techniques
Commonly used AI algorithms from Aidoc include detection of intracranial hemorrhage, cervical spine fracture, pulmonary embolism, rib fracture, and intraabdominal free air. This educational exhibit will showcase findings that were missed by the AI algorithm.
Conclusion
Most developers of AI algorithms advertise the subtle findings the algorithm could detect. However, there are also subtle findings that may be missed by the AI algorithm. It is crucial for the interpreting radiologist to recognize this potential pitfall and avoid the "satisfaction with AI" error.
