Author of

• 16637

  +

  P1.03 - Poster Session with Presenters Present (ID 455)

  • Event: WCLC 2016
  • Type: Poster Presenters Present
  • Track: Radiology/Staging/Screening
  • Presentations: 1
  • Moderators:
  • Coordinates: 12/05/2016, 14:30 - 15:45, Hall B (Poster Area)

  • 16700

    +

    P1.03-063 - Quantitative Imaging Features Predict Incidence Lung Cancer in Low-Dose Computed Tomography (LDCT) Screening (ID 5669)

    14:30 - 14:30  |  Author(s): L. Hall
    • Abstract

    Background:
    Although the NLST demonstrated a benefit of LDCT screening for reducing lung cancer and all-cause mortality, LDCT screening identifies large numbers of indeterminate pulmonary nodules and there are limited clinical decision tools that predict probability of cancer development. Using data and images from the NLST, we extracted quantitative imaging features from nodules of baseline positive screens (T0) and delta features from T0 to first follow-up (T1) and performed analyses to identify imaging features that predict incidence lung cancer. Analyses were stratified by nodule size since guidelines in the U.S. have increased the size threshold for positivity to 6 mm.
    
    Methods:
    We extracted 438 features from T0 nodules, and delta features from T0 to T1, including size, shape, location, and texture information. Nodules were identified for 170 cases that were diagnosed with incidence lung cancer at the first (T1) or second (T2) follow-up screen and for 328 controls that had three consecutive positive screens (T0 to T2) not diagnosed as lung cancer. The cases and controls were split into a training cohort and a testing cohort and classifier models (Decision tree-J48, Rule Based Classier-JRIP, Naive Bayes, Support Vector Machine, Random Forests) that were stratified by nodule size (< 6 mm, 6 to 16 mm, ≥ 16 mm) were used to identify the most predictive feature sets.
    
    Results:
    The training cohort consisted of 83 cases and 172 controls and a testing cohort of 77 cases and 135 controls. Within and across each cohort, there were no significant differences in demographic and clinical covariates. Training and testing was first performed using three difference nodule size groups (< 6 mm, 6 to 16 mm, and ≥ 16 mm) which revealed for < 6 mm a final model of 5 features (AUROC=0.75), 6 to 16 mm a final model with 10 features (AUROC=0.73), and ≥ 16 mm a final model with 10 features (AUROC=0.83). Finally, we combined the two larger groups and found for ≥ 6 mm a final model with 10 features with an AUROC of 0.84 (95% CI 0.8-0.87), Sensitivity of 60% (95% CI 0.50-0.71), and Specificity of 95% (95% CI 0.92-0.99).
    
    Conclusion:
    In this analysis we revealed a set of highly informative imaging features that predicts subsequent development of incidence lung cancer among individuals presenting with a ≥ 6 mm nodule. These imaging features could be scored in the clinical setting to improve nodule management of current size-based screening guidelines.