Artificial Intelligence in Radiology: Ally or Obstacle?
Artificial Intelligence (AI) is here to stay, but we decide how to welcome it, and how to implement it. For providers of intelligent solutions like Eden, it is crucial to consider how radiologists interact with these tools in their workspaces. This can make the difference between an AI that works with the physician and one that works against them.
The field of radiology employs a variety of intelligent tools for image analysis and the generation of diagnostic impressions. These programs are capable of analyzing large amounts of information in seconds, thereby reducing the physician’s workload, speeding up emergency diagnosis, and even revealing findings previously invisible to the radiologist.
But like any other technology, AI is fallible. According to a 2023 study based on knee osteoarthritis diagnoses, the error rate of an AI is generally equivalent to that of a human radiologist. However, depending on the interaction with the radiologist, these incidents can increase significantly and lead to incorrect decisions. Risk factors for medical error caused by AI include:
- Lack of understanding of how the algorithm works (also known as the “black box” effect)
- Absence of ROI or other visual AI indicators to flag its findings on the study image
- Automatic insertion of AI-generated findings into the study report
These results point to an implementation of AI in which the radiologist’s agency is not prioritized. Raymundo González, CTO at Eden, believes this should be a crucial consideration for any intelligent tool in the sector: “everything must be designed to give the physician complete control.”
Raymundo leads the development of Eden Creator, a language intelligence model that provides precise diagnostic conclusions based on the findings of a study. The goal of this tool is not to replace the physician’s interpretation, but to complement their written report, a task that “does not always require the radiologist’s skill and expertise.”
Creator is designed to empower the user, who has the ability to review the suggested conclusions, edit them, accept them, or simply reject them. With each instance, the algorithm learns from these expert decisions to refine its predictions further. This helps avoid the risk of an erroneous conclusion resulting from AI use. In addition, this tool can save the radiologist up to an hour of work, reducing the fatigue and eye strain that can also lead to medical error.
As intelligent tools become increasingly accessible worldwide, it is essential to explore implementation approaches that take the radiologist’s working conditions into account and mitigate the risk of medical error. As the American College of Radiology (ACR) states, “AI is never “an alternative” to the radiologist,” but rather an ally at their service.