A110-16 Repeatability and Short-Term Change of Deep Learning-Based Measurement of Lung Fibrosis on Computed Tomography

Abstract Rationale Automated quantification of pulmonary fibrosis on computed tomography (CT) is increasingly used to evaluate patients with interstitial lung disease (ILD). Assessing repeatability and short-term changes in quantitative metrics is essential for determining their sensitivity and clinical utility. Methods Patients with idiopathic pulmonary fibrosis (IPF) or progressive pulmonary fibrosis were enrolled from our institution’s ILD clinic at the time of a clinically ordered CT performed using a standardized protocol. Three months later, each patient underwent two repeat CT scans acquired within a short interval using the same protocol to assess repeatability. Fibrosis extent was quantified as a percentage of lung volume on each scan using the well-validated, deep learning-based data-driven textural analysis (DTA) method. Same-day scans with lung volume differences larger than 15% were excluded. DTA repeatability on test-retest scans was evaluated using intraclass correlation coefficient (ICC), Bland-Altman analysis, and repeatability coefficient (RC). Change in DTA was calculated as the average of the test-retest scans minus the baseline value. Association between baseline DTA and 3-month change in DTA was assessed using multivariable linear regression adjusted for age, sex, body mass index, smoking history, and IPF diagnosis. Results Fifty-two participants completed baseline and 3-month follow-up CT with test-retest scans. Average age was 74 years, 36 (69%) were male, and IPF was the primary diagnosis in 21(40%) participants. Median baseline DTA score was 25.7 (IQR 15.9-35.4). Two test-retest pairs were excluded due to large lung volume differences. Among the remaining 50 same-day pairs, DTA scores showed minimal mean differences (average −0.05 units; 95% CI −0.26 to 0.17), with a standard deviation (SD) of differences of 0.74 units. Bland-Altman plot (Figure 1) confirmed absence of bias in test-retest scores and limits of agreement were −1.50 to + 1.41. The within-subject SD was 0.52 and the RC was 1.45 units, supporting a cutoff of 1.5 as a minimal detectable change in DTA at ∼95% confidence. Using this threshold, 32 of 50 participants (64%) showed DTA progression at 3 months. In adjusted analysis, baseline DTA was independently associated with greater 3-month increase in DTA (β = 0.10, 95% CI 0.02-0.18, p = 0.021). Conclusions DTA showed excellent repeatability and sensitivity to short-term change. Higher baseline scores were associated with greater 3-month progression. Combined with prior validation against physiologic and survival outcomes, these findings further support the clinical utility of DTA. This abstract is funded by: This was an independent, investigator initiated study supported by Boehringer Ingelheim Pharmaceuticals, Inc. (BIPI). BIPI had no role in the design, analysis or interpretation of the results in this study; BIPI was given the opportunity to review the publication for medical and scientific accuracy as it relates to BIPI substances, as well as intellectual property considerations.