#AI #South Korea #Seoul #Lunit #Radiology

Bridging the Trust Divide between Radiologists and AI in Breast Cancer Screening

A groundbreaking study published by Lunit in the renowned journal Radiology has unveiled a critical trust gap in the collaboration between radiologists and artificial intelligence (AI) during breast cancer screening. Conducted in a real-world setting, this comprehensive research is seen as a large-scale evaluation of how radiologists interact with AI tools, particularly the Lunit INSIGHT MMG. With this study, Lunit aims to address the pressing issues surrounding AI integration in clinical workflows and diagnosis accuracy.

The Study's Framework

The research, part of the prospective ScreenTrustCAD trial, took place at Capio St Göran Hospital in Stockholm, Sweden. It involved more than 55,000 women, allowing for extensive analysis of AI's role in the screening process. Lunit INSIGHT MMG served as an independent third reader alongside two human radiologists, a setup that provided valuable insights into decision-making dynamics.

Key Findings

One of the most striking outcomes from the study was that despite AI's superior performance in identifying potential cancers, radiologists were less inclined to recall patients that AI had flagged. Specifically, when only AI indicated that a case could be cancerous, just 4.6% of those cases were recalled by radiologists. In contrast, 14.2% of cases identified solely by a radiologist were recalled. The recall rates rose to 57.2% when flagged by two radiologists, though it plummeted again to 38.6% when both AI and one radiologist pointed out a case.

These findings suggest a profound hesitation among radiologists to trust AI recommendations. Brandon Suh, CEO of Lunit, emphasized that the focus should not solely be on AI's performance but rather on the interpretations and actions of human clinicians in response to AI inputs. "The gap between AI input and human response can significantly impact real-world outcomes," he stated.

Insights from Experts

Dr. Karin Dembrower, lead author of the study and radiologist at Karolinska Institutet, echoed this sentiment by stressing the need for better integration of AI findings into clinical workflows. She remarked, "This isn't about whether AI can detect cancer; it's about how these findings are perceived and acted upon by healthcare professionals."

Evaluating Diagnostic Performance

While the interplay between radiologists and AI took center stage, the study also verified the strong diagnostic performance of Lunit's AI system. Among the women recalled after screening, 22% of those flagged by AI were diagnosed with cancer, compared to just 3.4% flagged by a single radiologist and 2.5% by two radiologists. Interestingly, when both AI and a radiologist flagged a case, the diagnosis rate rose substantially to 25%.

This variance illustrates AI's potential to identify high-risk cases that may otherwise be overlooked in traditional double-reading protocols. The discrepancies in trust and recall rates underscore a critical need for radiologists to feel more confident in AI's capabilities to enhance patient outcomes.

The Road Ahead

As Lunit continues to spearhead AI integration in healthcare, they aim to foster a more collaborative environment between AI tools and healthcare providers. The insights gleaned from this study highlight the necessity of designing solutions that not only showcase AI's advantages but also build clinician trust in these technologies.

Lunit's commitment to defeating cancer through AI is evident as they expand their reach, currently serving over 4,800 medical institutions across more than 55 countries. Their innovative approach underscores the importance of harnessing advanced technology in transforming the landscape of cancer diagnostics and treatment.

For further details, explore the full study published in Radiology.