AI system could reduce radiologists' workload for reporting chest radiographs

A recent multicenter study, published in European Radiology on May 17, has demonstrated that a commercially available artificial intelligence (AI) system can effectively identify normal chest radiographs. The use of such an AI system may reduce the workload for radiologists. This innovative approach holds promise for enhancing efficiency in radiology departments in times of increasing workloads.

The study, conducted by Steven Schalekamp at the imaging department of Radboudumc (Nijmegen) and Jeroen Bosch Hospital (‘s Hertogenbosch), evaluated the potential of the Lunit INSIGHT CXR3 AI system to identify normal chest radiographs. Researchers retrospectively analyzed 1670 consecutive chest radiographs.

The AI system achieved an area under the Receiver Operating Characteristic (ROC) curve (AUC) of 0.918 in detecting normal radiographs. The ROC curve is a graph that shows how well a system can tell the difference between two outcomes, like normal and abnormal X-rays, as the criteria for making decisions change. The AUC, or area under the curve, is a number that represents the system's overall accuracy. No significant performance difference was noted between the two hospitals involved, highlighting the AI's reliability.

The system also showed promise in reducing radiologists' workloads, identifying 53% of normal chest radiographs at a conservative threshold, which could potentially reduce workload by 15%. Additionally, the AI system exhibited high sensitivity for urgent and critical findings, ensuring no critical cases were missed, and achieved a remarkable negative predictive value (NPV) of 98% for these findings.

The ability of AI to reliably identify normal chest radiographs can streamline radiology workflows by eliminating unnecessary reporting of normal images. This is particularly relevant in an era where radiology departments face increased pressures to maintain efficiency and manage growing workloads. However, the AI product is not yet approved for autonomous use and should be supervised by a radiologist, which limits the potential workload reduction.

While the study's findings are promising, further research is necessary to validate the AI system's performance on larger and more diverse datasets. The imaging department of the Radboudumc started a prospective study to evaluate the effectiveness of the AI system in current clinical practice.

-

Read the full study here: Performance of AI to exclude normal chest radiographs to reduce radiologists’ workload | European Radiology (springer.com)

Schalekamp, S., van Leeuwen, K., Calli, E. et al. Performance of AI to exclude normal chest radiographs to reduce radiologists’ workload. Eur Radiol (2024). https://doi.org/10.1007/s00330-024-10794-5