(주) 뷰노

ECR 2026

High-Suspicion Pulmonary Nodule Detection on Chest Radiographs: Single-Center Retrospective Performance and Additional Detection Analysis of an AI System

  • Mar. 2026
  • by Y. Hwang et. al.

Single-Center Retrospective Performance and Additional Detection Analysis of an AI System


Purpose

To evaluate an artificial intelligence (AI) system for detecting suspicious pulmonary nodules in a single-center retrospective cohort and to assess its additional detection relative to radiologists' CXR-based CT triage decision ("CT call").


Methods

This study analyzed retrospective CXR data with the VUNO Med Chest X-ray AI system, comparing its detection performance against radiologists' CT triage decisions. Performance metrics of image-level and lesion-level false positive analysis were conducted to understand AI system behavior in clinical practice.


Findings

Among 310 CT-call CXRs by radiologist's visual assessment alone, 54 had high-suspicion nodules on CT; AI detected 46/54, yielding a sensitivity of 85.2%. In the 226 Non-call CXRs, AI flagged 20.8% (47/226). These AI flagged cases were reviewed by a radiologist with paired-CT. In the non–CT-call cases (n=226), lesion-level FP analysis (n=58) showed non-neoplastic lung conditions (post-inflammatory fibrosis (15.5%), interstitial markings (10.3%), focal atelectasis (6.9%)), and vascular markings (6.9%), while a small subset (e.g., Sclerotic bone metastasis from primary lung cancer) may represent clinically relevant incidental findings.

Tag in

#medical_image

#VUNO Med®-Chest X-ray™