How comfortable are radiologists and neurologists in incorporating artificial intelligence (AI) for the triage of brain magnetic resonance imaging (MRI) scans?
In a new study, recently published in the European Journal of Radiology, researchers conducted a survey of 55 United Kingdom-based neuroradiologists, 41 general radiologists and 37 neurologists to assess clinician perceptions and comfort level with the use of AI for brain MRI triage. The study authors noted that 61 percent of the respondents had not used AI in clinical practice. They added that the median age of the survey respondents ranged between 35 to 44.
The researchers found that 74 percent of survey respondents noted a need for improvement in the prioritization of brain MRI scans with 71 percent preferring the use of AI-assisted triage over current chronological reporting within patient classes.
A majority of the respondents indicated a willingness to accept AI-based triage decisions for brain MRI scans without radiologist input. The researchers noted that 47 percent were comfortable, and 11 percent were very comfortable with trusting AI triage decisions. According to the survey results, 23 percent were neutral, and 14 percent were uncomfortable.
However, 82 percent of survey respondents emphasized the importance of seeing the rationale behind an AI tool’s triage decision. Sixty percent said the inclusion of an AI-generated heat map to highlight abnormality detection reinforced their confidence in the AI software.
“The findings suggest that clinical AI triaging tools should maintain the concept of explainability as a core priority throughout their development and implementation process to ensure that clinicians feel confident in its application in patient care,” wrote lead study author Munaib Din, M.D., who is affiliated with the School of Biomedical Engineering and Imaging Sciences at King’s College London in the United Kingdom, and colleagues.
(Editor’s note: For additional content on MRI-related news and research, click here.)
The researchers noted a higher degree of confidence in AI triage capabilities among neuroradiologists in comparison to neurologists but also pointed out that neuroradiologists and general radiologists had greater familiarity with AI imaging tools than neurologists.
Three Key Takeaways
1. Preference for AI in brain MRI triage. A significant portion (71 percent) of surveyed clinicians expressed a preference for AI-assisted triage over current chronological reporting systems for prioritizing brain MRI scans.
2. Explainability enhances confidence. Most respondents (82 percent) emphasized the importance of understanding the rationale behind AI decisions with 60 percent noting that visual tools like AI-generated heat maps improve their confidence in the AI's outputs.
3. Radiologists are more comfortable than neurologists with AI. Neuroradiologists and general radiologists showed higher confidence in AI triage capabilities compared to neurologists, likely due to greater familiarity and exposure to AI tools in clinical imaging.
“It is plausible that the discrepancy results from neurologists having less exposure than their radiologist colleagues to imaging AI tools or having fewer concerns regarding reporting backlogs. It is uncertain whether this difference in perspective between radiologists and neurologists would impact the clinical translation of this particular AI application, as both groups work closely together in facilitating the patient pathway,” added Din and colleagues.
For the utilization of AI into radiology workflow, 50 percent of the survey respondents preferred employing AI as a first reader for detecting possible abnormalities with 38 percent seeing AI as a second reader for post-radiologist review, according to the study authors.
(Editor’s note: For related content, see “Brain MRI Study Assesses Impact of AI in Differentiating Radionecrosis from Neoplastic Progression of Metastasis,” “AI Software for Brain MRI Gets Expanded FDA Clearance for Multiple Sclerosis Assessment” and “FDA Clears AI-Powered MRI Software for ARIA Detection in Patients with Alzheimer’s Disease.”)
In regard to study limitations, the authors acknowledged that processes for AI approval and implementation into workflows differ across countries, noting that the survey respondents were comprised entirely of clinicians based in the United Kingdom. The researchers also noted the survey utilized for the study did not explore medicolegal considerations with AI implementation.
