A recent study, published in the journal Radiology, has demonstrated that assistance from a high diagnostic accuracy artificial intelligence (AI) algorithm can significantly enhance radiologists’ ability to detect lung cancers on chest X-rays.
The study also found that radiologists were more willing to accept AI suggestions when the AI system’s diagnostic performance was high.
The Study: AI and Radiologists Working Together
Researchers at Seoul National University conducted a retrospective study involving 30 readers (20 thoracic radiologists and 10 radiology residents) who analyzed 120 chest X-rays without AI assistance initially.
In a second session, they reassessed the X-rays, this time with the aid of either a high- or low-accuracy AI. The readers were not aware that two different AI systems were in play.
The Findings: Accuracy Matters
The study found that the use of high-accuracy AI notably improved readers’ detection performance compared to low-accuracy AI.
High-accuracy AI also led to more frequent changes in reader determinations, a phenomenon known as susceptibility.
In other words, the radiologists were more likely to modify their assessments based on the high-accuracy AI’s recommendations.
Trust in AI
Study lead author, Dr. Chang Min Park, emphasized the impact of trust in AI on human susceptibility to AI suggestions.
When radiologists were provided with suggestions by high-performance AI, they were more likely to consider and incorporate these suggestions into their assessments.
The findings highlight the crucial role of trust and performance in human-AI collaboration.
AI’s Performance Impact
Compared to the first reading session, the use of high diagnostic accuracy AI in the second session resulted in an increase in per-lesion sensitivity (0.63 versus 0.53), and specificity (0.94 versus 0.88).
In contrast, when low diagnostic accuracy AI was used, there was no improvement in these measurements.
Contextualizing AI Use
While the study underscores the benefits of high diagnostic performance AI, Dr. Park noted that what constitutes “high diagnostic performance AI” could vary depending on the task and the clinical context.
An AI model that can detect all abnormalities may seem ideal, but such a model might have limited practical use in certain clinical settings, like mass screening for pulmonary tuberculosis.
Dr. Park stressed the need for both the development of high-performance AI models and a thoughtful consideration of the clinical setting in which the AI will be utilized for the best results.
Looking Ahead
The researchers plan to extend their investigation of human-AI collaboration to the detection of other abnormalities on chest X-rays and CT images.
This research highlights the potential for AI to enhance diagnostic accuracy in radiology when the technology is appropriately developed and applied.
If you care about lung health, please read studies about why Viagra may be useful in treating lung diseases, and scientists find herbal supplement to treat lung cancer.
For more information about health, please see recent studies that coconut sugar could help reduce artery stiffness, and results showing fungi causing deadly lung infections found throughout the US.
The study was published in Radiology.
Copyright © 2023 Knowridge Science Report. All rights reserved.
