Defining the Intra-subject Variability of Whole-lung CT Densitometry in Two Lung Cancer Screening Trials

Abstract

To define a statistically based variation of individual whole-lung densitometry above which a real increase of pulmonary extent can be suspected in lung cancer screening trials. Baseline and 3-month follow-up low-dose computed tomography (LDCT) examinations of 131 smokers or former smokers recruited in the ITALUNG (32 subjects) and MILD (99 subjects) trials were compared using for each data set two different image processing tools for whole-lung densitometry. Both trials were approved by institutional review boards, and written informed consent was obtained from all participants. Assuming that no change of emphysema extent can occur in a 3-month interval, the Bland and Altman method was used to assess the agreement between baseline and follow-up LDCT examinations for lung volume, 15th percentile (Perc15) of lung density and Perc15 corrected for lung volume by application of a linear detrend on log-transformed data. Similar results were obtained in each data set using two different image processing tools. In the ITALUNG cohort the 95% limits of agreement (LoA) interval of volume corrected Perc15 was -9.7 to 10.7% using image processing method 1 and -10.3 to 11.5% using image processing method 2. In the MILD cohort, the 95% LoA interval of volume corrected Perc15 was -14.7 to 17.3% with both image processing methods. In the two considered lung cancer screening settings a range of 9.7-14.7% decrease of volume corrected Perc15 represents a statistically defined threshold to suspect a real increase of emphysema extent in serial LDCT examinations.