Pulmonary nodules: Improved detection with vascular segmentation and extraction with spiral CT - Work in progress

Pierre Croisille, Miguel Souto, Maria Cova, Susan Wood, Yohannes Afework, Janet E. Kuhlman, Elias A. Zerhouni

Research output: Contribution to journalArticlepeer-review


PURPOSE: To determine whether extraction of pulmonary vessels from computed tomographic (CT) images with automated segmentation improves the detection of pulmonary nodules. MATERIALS AND METHODS: Simulated nodules were superimposed on normal spiral CT images. Eight patients referred for CT assessment of pulmonary nodules were selected for clinical evaluation. Vessels were extracted from both the simulation and clinical study with a three-dimensional seeded region-growing algorithm. Three experienced radiologists were asked to locate the nodules and assign a level of confidence to their findings. Sensitivity and proportion of false-positive results per case (FPC) were calculated. Observer performance was evaluated by alternate free-response receiver operating characteristic analysis. RESULTS: Extraction of vascular structures from CT scans improved sensitivity from 63% to 84% in the simulation study and from 58% to 78% in the clinical study. The proportion of FPC decreased from 52% to 24% and from 55% to 12%, respectively. Radiologists performed consistently better with the segmented images than with the original images in both the simulation (P = .006) and the clinical (P = .0013) study. CONCLUSION: Automated vessel subtraction and extraction improves detection of pulmonary nodules.

Original languageEnglish (US)
Pages (from-to)397-401
Number of pages5
Issue number2
StatePublished - Nov 1995


  • Lung neoplasms, 60.30
  • Lung neoplasms, diagnosis, 60.30
  • Lung, CT, 60.12115
  • Lung, nodule, 60.281

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging


Dive into the research topics of 'Pulmonary nodules: Improved detection with vascular segmentation and extraction with spiral CT - Work in progress'. Together they form a unique fingerprint.

Cite this