Feasibility of AI as First Reader in the 4-IN-THE-LUNG-RUN Lung Cancer Screening Trial: Impact on Negative-Misclassifications and Clinical Referral Rate
Authors
Anna N H Walstr, Harriet L Lancaster, Marjolein A Heuvelmans, Carlijn M van der Aalst, Juul Hubert, Dana Moldovanu, Sytse F Oudkerk, Daiwei Han, Jan Willem C Gratama, Mario Silva, Harry J de Koning, Matthijs Oudkerk
The paper published by iDNA evaluates the feasibility of using AI software (AVIEW-LCS) as a first reader in the 4-IN-THE-LUNG-RUN trial. This study compared the performance of AI and radiologists in identifying negative cases and their impact on clinical referral rates. Among 3,678 participants, AI recorded 31 (0.8%) negative misclassifications (NM), outperforming radiologists, who recorded 407 (11.1%) NMs. Additionally, the NM referral rate for AI was significantly lower at 2.9%, compared to 11.8% for radiologists. AI demonstrated the potential to reduce workload by 71.2% while maintaining diagnostic safety. The study highlights AI’s potential to improve diagnostic workflows and alleviate radiologists’ burden.
Here, NM refers to false-negative classifications at the first reading, while NM referral represents false-negative classifications at the final review. The paper also discusses positive misclassification (PM), noting that AI's PM rate (5.7%) was higher than that of radiologists (0.5%). However, it concludes that the additional burden from AI’s PM is minimal while emphasizing the need for further research.