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)

  • 16691

    +

    P1.03-054 - Quantitative Accuracy and Lesion Detectability of Low-Dose FDG-PET for Lung Cancer Screening (ID 3839)

    14:30 - 14:30  |  Author(s): A.T. Sjoeholm
    • Abstract

    Background:
    Low-dose computed tomography (CT) screening for high-risk patients can reduce lung cancer mortality, but false-positivity rates are high. Positron emission tomography (PET)/CT is more accurate compared to CT alone, but typically is associated with a higher radiation exposure. We investigate a low radiation dose PET/CT solution without compromising quality.
    
    Methods:
    Twenty lung cancer patients were scanned with PET/CT after an uptake period of 60 min, following injection of 5.9±0.14 mCi 18F-Fluorodeoxyglucose. All were scanned with 2 beds covering the lungs at 10 min each, resulting in 120±25 x106 mean true coincident counts per bed. Reduced doses were simulated by randomly discarding events in the PET list mode according to 9 predefined true count levels, from 20 to 0.25 x10[6]. For each patient & simulated dose, the highest possible number of independent realizations was generated & reconstructed, up to 50. The reconstruction algorithm was OP-OSEM, using TOF and PSF, with 2 iterations, 21 subsets & 5mm smoothing, producing 400x400 image matrices with voxel size 2.04x2.04x2.03mm. At each simulated dose, lesions consistent with those of early lung cancer were identified & classified by metabolic volume, signal-to-background contrast, mean & max SUV, lesion-to-background SNR, & Hotelling observer SNR. Bias & stability of the lesion activity measurements were evaluated across all simulated dose levels, and detectability was determined by various human-trained, numerical observer models.
    
    Results:
    Twelve isolated lung lesions (mean volume 2.61±2.86 cm[3] on CT) were studied in detail. Analyses of bias & reproducibility in the lesion activity values showed that measurements were stable until the count levels approached extreme conditions. Bias in the lesion VOI mean & max SUV were relatively negligible until true count level was decreased to 1 million. Variance on reproducibility of lesion values showed a more dramatic trend, but standard deviation was still around 10% at 5 million counts.
    
    Conclusion:
    We show that simulated images with accurate lesion characteristics can be obtained at 1/12 of a typical radiotracer dose.