PET-CT-based auto-contouring in non-small-cell lung cancer correlates with pathology and reduces interobserver variability in the delineation of the primary tumor and involved nodal volumes

Int J Radiat Oncol Biol Phys. 2007 Jul 1;68(3):771-8. doi: 10.1016/j.ijrobp.2006.12.067. Epub 2007 Mar 29.


Purpose: To compare source-to-background ratio (SBR)-based PET-CT auto-delineation with pathology in non-small-cell lung cancer (NSCLC) and to investigate whether auto-delineation reduces the interobserver variability compared with manual PET-CT-based gross tumor volume (GTV) delineation.

Methods and materials: Source-to-background ratio-based auto-delineation was compared with macroscopic tumor dimensions to assess its validity in 23 tumors. Thereafter, GTVs were delineated manually on 33 PET-CT scans by five observers for the primary tumor (GTV-1) and the involved lymph nodes (GTV-2). The delineation was repeated after 6 months with the auto-contour provided. This contour was edited by the observers. For comparison, the concordance index (CI) was calculated, defined as the ratio of intersection and the union of two volumes (A intersection B)/(A union or logical sum B).

Results: The maximal tumor diameter of the SBR-based auto-contour correlated strongly with the macroscopic diameter of primary tumors (correlation coefficient = 0.90) and was shown to be accurate for involved lymph nodes (sensitivity 67%, specificity 95%). The median auto-contour-based target volumes were smaller than those defined by manual delineation for GTV-1 (31.8 and 34.6 cm(3), respectively; p = 0.001) and GTV-2 (16.3 and 21.8 cm(3), respectively; p = 0.02). The auto-contour-based method showed higher CIs than the manual method for GTV-1 (0.74 and 0.70 cm(3), respectively; p < 0.001) and GTV-2 (0.60 and 0.51 cm(3), respectively; p = 0.11).

Conclusion: Source-to-background ratio-based auto-delineation showed a good correlation with pathology, decreased the delineated volumes of the GTVs, and reduced the interobserver variability. Auto-contouring may further improve the quality of target delineation in NSCLC patients.

Publication types

  • Evaluation Study
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Artificial Intelligence*
  • Carcinoma, Non-Small-Cell Lung / diagnosis*
  • Female
  • Fluorodeoxyglucose F18
  • Humans
  • Image Enhancement / methods*
  • Imaging, Three-Dimensional / methods*
  • Lung Neoplasms / diagnosis*
  • Male
  • Observer Variation
  • Pattern Recognition, Automated / methods
  • Positron-Emission Tomography / methods*
  • Radiopharmaceuticals
  • Reproducibility of Results
  • Sensitivity and Specificity
  • Subtraction Technique
  • Tomography, X-Ray Computed / methods*


  • Radiopharmaceuticals
  • Fluorodeoxyglucose F18