Computer-aided detection of pulmonary nodules: a comparative study using the public LIDC/IDRI database

• DOI: 10.1007/s00330-015-4030-7
• Publication Date: 2015-10-06T17:46:07Z
• Authors: Colin Jacobs; Eva M. van Rikxoort; Keelin Murphy; Mathias Prokop; Cornelia M. Schae...; Bram van Ginneken
• Keywords: Nodule (geology)

To benchmark the performance of state-of-the-art computer-aided detection (CAD) pulmonary nodules using largest publicly available annotated CT database (LIDC/IDRI), and to show that CAD finds lesions not identified by LIDC’s four-fold double reading process. The LIDC/IDRI contains 888 thoracic scans with a section thickness 2.5 mm or lower. We report two commercial one academic system. influence presence contrast, thickness, reconstruction kernel on was assessed. Four radiologists independently analyzed false positive marks best updated system showed sensitivity 82 % at an average 3.1 detections per scan. Forty-five were scored as all four in our study. On reference for lung nodule chest CT, locates vast majority low rate. Potential is substantiated fact it identifies marked during extensive LIDC annotation • systems should be validated public, heterogeneous databases. excellent benchmarking CAD. can identify missed two-stage