Automation Bias May Impair Radiologist Performance on Mammograms
New research finds that incorrect suggestions from an AI-based clinical decision support system could impair radiologist accuracy on mammograms.
Source: Getty Images
By - Incorrect advice from artificial intelligence (AI)-based decision support systems may negatively impact the performance and accuracy of radiologists at all levels of expertise when reading mammograms, according to a study published last week in Radiology.
In recent years, healthcare stakeholders have posited that AI can significantly improve medical imaging. Applications for AI in radiology vary, but the study indicates that AI-based mammographic support systems have demonstrated some of the most promising of these.
Despite this, challenges in applying AI to medical imaging, such as automation bias, can limit the use of these technologies.
The study defines automation bias as “the propensity for humans to favor suggestions from automated decision-making systems” and indicates that the phenomenon is a known source of error in human-AI interactions. However, the role of automation bias in AI-aided mammography have not been adequately studied, the researchers stated.
To address this research gap, the researchers sought to investigate how automation bias affects inexperienced, moderately experienced, and very experienced radiologists when reading mammograms with the support of an AI system.
The research team tasked 27 radiologists with reading 50 mammograms, after which they provided corresponding Breast Imaging Reporting and Data System (BI-RADS) assessments, an approach used to describe breast imaging findings.
The mammograms presented were split into two randomized sets: a training set of 10 mammograms in which the AI suggested the correct BI-RADS categories, and a second set with 40 mammograms, 12 of which contained incorrect BI-RADS categories purportedly suggested by the AI.
From there, the research team evaluated the radiologists’ performance at assigning correct BI-RADS scores.
Overall, radiologists demonstrated significant dips in performance when the purported AI suggested incorrect BI-RADS categories.
inexperienced radiologists assigned the correct BI-RADS category nearly 80 percent of the time when the AI suggested the correct BI-RADS category, but their accuracy fell below 20 percent when the AI suggested the wrong BI-RADS score.
Experienced radiologists also saw significant performance dips, but to a lesser extent: their accuracy went from approximately 82 percent to 45.5 percent when the AI suggested the incorrect score.
“We anticipated that inaccurate AI predictions would influence the decisions made by radiologists in our study, particularly those with less experience,” said study lead author Thomas Dratsch, MD, PhD, from the Institute of Diagnostic and Interventional Radiology, at University Hospital Cologne in Cologne, Germany, in the press release. “Nonetheless, it was surprising to find that even highly experienced radiologists were adversely impacted by the AI system’s judgments, albeit to a lesser extent than their less seasoned counterparts.”
Their findings indicate that factors like automation bias must be carefully considered when combining clinicians’ expertise with AI-aided decision support systems, the researchers stated.
“Given the repetitive and highly standardized nature of mammography screening, automation bias may become a concern when an AI system is integrated into the workflow,” Dratsch said. “Our findings emphasize the need for implementing appropriate safeguards when incorporating AI into the radiological process to mitigate the negative consequences of automation bias.”
The research team indicated that teaching users about the reasoning process of the AI system, providing them with confidence levels for each of the system’s outputs, and ensuring that clinicians using an AI system feel accountable for their own decisions could be possible safeguards against automation bias.
Moving forward, the researchers will leverage eye-tracking technology to further investigate the decision-making processes of radiologists using AI and explore different methods of presenting AI outputs to see which are most effective at avoiding automation bias.
