Author of

• 20192

  +

  P3.13 - Radiology/Staging/Screening (ID 729)

  • Event: WCLC 2017
  • Type: Poster Session with Presenters Present
  • Track: Radiology/Staging/Screening
  • Presentations: 1
  • Moderators:
  • Coordinates: 10/18/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)

  • 24156

    +

    P3.13-004 - Prospective Study of Sequential Ultra-Low then Standard Dose 18F-FDG PET/CT Scans for Lung Lesion Detectability  (ID 8102)

    09:30 - 09:30  |  Author(s): J.D. Schaefferkoetter
    • Abstract

    Background:
    Lung cancer screening with low-dose computed tomography (CT) is better than chest X-rays but is non-specific. Accuracy is improved with positron emission tomography (PET), at a cost of additional radiation. We had previously reported on simulated low-dose PET imaging and demonstrated that 10x10[6] net true counts is sufficient to generate images with acceptable diagnostic quality. We now hypothesize that we can maintain image quality with a 92% reduction of fluorodeoxyglucose (FDG) tracer activity from 6 mCi to 0.5 mCi.
    
    Method:
    Nine patients have been scanned with two sequential PET/CT scans on the same day. The patient is first scanned with 0.5 mCi FDG and a low-dose CT protocol, followed by a routine PET/CT with 6 mCi FDG. PET data from the standard-dose scan were manipulated to emulate various noise (dose) levels, corresponding to nine pre-defined true count levels. Data were matched to the level of the low-dose scan, to compare noise statistics to a ground truth and to directly validate our methods. The data were reconstructed, with many independent noise realizations, and the images were reviewed. Ten lesions, in seven patients, were identified as having the size and uptake consistent with those found in early disease. For a given count level, the corresponding images were determined to be acceptable if lesion detectability was comparable to that found in the full-statistic image set. Detection performance was determined automatically by machine learning, namely, convolution neural networks trained by 4 previous observer responses.
    
    Result:
    Lesion detection accuracy was evaluated in 4458 total image sub-volumes. Regions containing both target lesions (2627 samples) and healthy lung background (1831 samples) were used to assess sensitivity and specificity for the task at all noise levels. The table shows data across the 4 observer models.

    Mean Sensitivity & Specificity by True Count Groups

    True count/ millions	<0.5	0.5-1	1-2	2-5	5-10	10-20	>20
    Mean sensitivity/%	0.35	18.85	62.35	85.40	95.73	96.23	96.42
    Mean Specificity/%	6.64	6.93	20.43	66.54	93.55	96.87	98.25

    
    
    Conclusion:
    Low-dose PET can provide good performance for lesion detection within the true count range 5-10×10[6].