Researchers at Herlev and Gentofte Hospital in Copenhagen reported that doctors still outperformed artificial intelligence when interpreting chest X-rays. The findings appeared in the journal Radiology, offering important context for how AI can support clinical work without replacing expert judgment.
The study assessed more than 2,000 X-ray images analyzed by 72 radiologists alongside four commercial AI systems. The scans were collected across four Danish hospitals over a two-year period. Across this dataset, AI precision varied from 62% to 95%, with higher sensitivity reducing missed disease cases. Yet radiologists demonstrated higher overall accuracy, particularly in avoiding false positives. The AI systems struggled most with identifying airspace diseases of the lungs, showing positive predictive values in the 40% to 50% range in that category.
Unnecessary imaging due to false positives carries risks and costs, including patient radiation exposure and higher healthcare expenses. As a result, the study suggests that AI should remain as an assistive tool rather than a standalone detector for common lung conditions.
The researchers also noted that earlier work claiming AI superiority often relied on image review alone, without considering a patient’s medical history or prior imaging. In real clinical practice, radiologic interpretation typically integrates multiple data sources beyond the image itself.
Earlier discussions have explored how socioeconomic factors influence health outcomes. For example, studies have highlighted that low-income individuals may have a greater chance of living longer when certain social determinants are addressed, underscoring the broader context in which diagnostic tools are used (attribution: Radiology).