AB021. Validation of real-world, non-research thoracic CT scans for quantitative analysis of COPD

Background: Quantitative CT (QCT) imaging plays an important role in phenotyping COPD and uses the voxel density histogram to measure total lung volume (TLV) and emphysema surrogates: low attenuation area (LAA) and lung density (LD). LD is often volume corrected using the predicted total lung capacity (TLC) to compensate for submaximal inspiration prior to image acquisition. QCT is carried out with careful attention to quality control including scanner make/model, calibration frequency, lung volume, acquisition protocol, and the use of contrast, and bears a financial and radiation cost. We wished to determine if: (I) thoracic CT scans acquired for clinical indications on a variety of scanners from different centres with varying calibration frequency, acquisition protocols and only simple breath holding instructions could yield reproducible data; (II) volume correcting LAA and LD using the pulmonary function test (PFT) measured TLC would compensate for submaximal inspiration better than using the predicted TLC; and (III) contrast infusion causes predictable changes in the QCT metrics TLV, LAA and LD.

Methods: A total of 82 subjects (67 COPD, 15 non-COPD) from a community respirology practice had at least 2 CT scans judged free of significant infiltrates, performed on 10 different models of scanner in 7 different community hospitals or radiology centres for clinical indications within a 13-month period and had pulmonary function tests performed respecting ATS criteria within 14 months of at least 1 CT scan. Images were analysed with Airway Inspector in ITALIC FONT (airwayinspector.acil-bwh.org) for LAA [<-950 Hounsfield Unit (HU)], LD (at 15th percentile + 1,000 HU) and TLV. 46 paired non-contrast scans (NC/NC) and 42 paired contrast/non-contrast scans (C/NC, 23 CT angio with early infusion, 19 routine contrast with late infusion) were used to construct identity plots for TLV, LAA, LD, and LAA and LD corrected for both predicted TLC and PFT measured TLC. LAA was volume corrected (VC) using the formula LAA_VC = LAA × TLC/TLV and LD using LD_VC = LD × TLV/TLC, where TLC was either the predicted TLC (PTLC) or PFT measured TLC (MTLC). Regression line slopes and Pearson’s r and P values were calculated for identity plots. Paired Student t-tests were used to detect differences in group mean TLV, LAA and LD, natural log transformed if necessary, between CT scan pairs. Significance was set at P<0.05 after Bonferroni correction. The study had local IRB approval.

Results: NC/NC inter-scan and CT-PFT intervals were 5.5 months ± 3.7 SD and 4.6±3.9 respectively, and C/NC 5.9±3.9 and 5.2±4.9 respectively. NC/NC identity plot slope, r and P values, and CT scan 1 vs. 2 mean/median and Student t-test p values respectively were; TLV: 0.93, 0.97, P<0.001, 5.32±1.42 vs. 5.38L±1.48 L, 0.22; LAA: 1.01, 0.95, P<0.001, 0.03 (0.01‒0.09 IQR) vs. 0.03 (0.01‒0.08), 0.81; LAAVC-PTLC: 1.01, 0.95, P<0.001, 0.03 (0.01‒0.08) vs. 0.03 (0.01‒0.07), 0.95; LAAVC-MTLC: 1.02, 0.95, P<0.001, 0.03 (0.01‒0.10) vs. 0.04 (0.01‒0.09), 0.99; LD: 0.98, 0.96, P<0.001, 80±27 vs. 82±28 g/L, 0.15; LDVC-PTLC: 1.00, 0.97, P<0.001, 82±26 vs. 83±27, 0.30; LDVC-MTLC: 1.03, 0.97, P<0.001, 68±23 vs. 69±25, 0.38; and for C/NC; TLV: 0.90, 0.91, P<0.001, 5.04±1.41 vs. 5.52±1.41 L, P<0.001; LAA: 0.67, 0.94, P<0.001, 0.03 (0.01‒0.06) vs. 0.05 (0.01‒0.10), P<0.001; LAAVC-PTLC: 0.74, 0.94, P<0.001, 0.02 (0.01‒0.05) vs. 0.04 (0.01‒0.08), P<0.001; LAAVC-MTLC: 0.75, 0.94, P<0.001, 0.03 (0.01‒0.07) vs. 0.05 (0.01‒0.11), P<0.001; LD: 1.21, 0.96, P<0.001, 98±40 vs. 79±31, P<0.001; LDVC-PTLC: 1.12, 0.96, P<0.001, 92±27 vs. 81±23, P<0.001; LDVC-MTLC: 1.10, 0.98, P<0.001, 79±30 vs. 70±26, P<0.001. For NC/NC, there is a high degree of reproducibility between scans for both LAA and LD with regression line slopes close to unity, Pearson’s r values between 0.95 and 0.97 and P<0.001 for all. The means of the lnLAA and LD do not differ and appear to become more similar with a step-wise increase in the Student t-test p values as lung volume correction proceeds from not correcting, to correcting using PTLC, to correcting using MTLC. On the other hand, while the regression line slopes of the C/NC correlate very strongly with Pearson’s r values between 0.94 and 0.98 and P<0.001 for all, they progressively move toward unity as volume correction proceeds from not correcting, to correcting using PTLC to correcting using MTLC. The means of the lnLAA and LD remain significantly different between C/NC (P<0.001 for all) regardless of the lung volume correction method.

Conclusions: Real-world, non-contrast thoracic CT scans can provide reproducible QCT data. LAA and LD seem more reproducible when volume corrected using the PFT measured TLC than when using the predicted TLC and suggests the former better compensates for submaximal inspiration prior to image acquisition. Contrast infusion has predictable effects on QCT metrics decreasing LAA by 33% and increasing and LD by 21%. Volume correcting using the PFT measured TLC reduces the contrast effect on LAA to 25% and LD to 10%. If validated by other centres, these findings suggest the pool of observational QCT data could be vastly expanded at little dollar and no radiation cost.

Keywords: Quantitative CT (QCT); chronic obstructive pulmonary disease (COPD); pulmonary function test (PFT)