Artificial intelligence model boosts lung nodule diagnostic accuracy in clinical trial

DeepFAN is a transformer-based model trained on more than 10,000 pathology-confirmed lung nodules and evaluated in a multireader, multicase clinical trial (Chinese Clinical Trial Registry ChiCTR2400084624; primary publication: Nature Cancer, DOI 10.1038/s43018-026-01147-w). The model reached an AUC of 0.939 (95% CI 0.930-0.948) on an internal test set and 0.954 (95% CI 0.934-0.973) on the trial dataset of 400 cases drawn from three independent institutions. Across 12 junior radiologists, model assistance improved average AUC by 10.9% (95% CI 8.3-13.5%), accuracy by 10.0%, sensitivity by 7.6% and specificity by 12.6% (all P<0.001), while interreader agreement rose from fair to moderate (κ 0.313 versus 0.421; P=0.019). The trial measured the accuracy of radiological READINGS, not patient outcomes — whether this translates into earlier lung cancer diagnoses or fewer unnecessary follow-up scans remains untested.