Real-world sensitivity of AI CADe for lung cancer detection on CT: 72% overall, but only 38% for ground-glass nodules

In a real-world lung cancer screening cohort (n=129), AI computer-aided detection (CADe) had 72% sensitivity for malignant nodules on CT, vs 92% for vessel-suppressed images. CADe sensitivity was 38% for ground-glass and 13% for cystic nodules (P<0.001). Subpleural/paramediastin…

• Retrospective analysis of 129 patients with lung cancer in a screening program after deployment of ClearRead VIS/CADe.
• Vessel-suppressed images showed high sensitivity (92%) across nodule sizes and morphologies, while CADe performance varied significantly.
• After excluding masses, subpleural and paramediastinal locations were independently associated with lower CADe sensitivity (P<0.05).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.