Investigating a Novel CT Density-Based Approach to Regional Lung Ventilation Mapping
Abstract
Purpose
Functional avoidance radiotherapy requires accurate regional lung ventilation mapping. Jacobian-based free-breathing ventilation derived from deformable image registration (DIR) of 4DCT is established but susceptible to inaccuracies caused by errors in the displacement fields. We are developing an alternative HU-based ventilation mapping technique exploiting the multi-scan acquisition of 5DCT.
Methods
We retrospectively analyzed 204 patients who underwent 5DCT imaging (25 sequential fast-helical free-breathing CT scans per patient). Two-stage DIR (deedsBCV followed by pTV) aligned 24 scans to a selected reference image. For each lung voxel, we performed log-linear regression of HU-based tissue density against respiratory amplitude across all 25 scans, yielding HU-based ventilation as the negative slope of log-density versus amplitude. Jacobian-based ventilation was calculated conventionally from DIR displacement fields. Whole-lung validation compared patient-mean ventilation values using linear regression. Voxel-wise validation assessed spatial agreement using Spearman correlation coefficients and Dice similarity coefficients for the 20% highest-ventilation regions using each technique. Analysis included 200 patients for voxel-wise metrics.
Results
HU-based and Jacobian-based ventilation showed excellent whole-lung agreement with a slope 0.99 (95% CI: 0.96–1.03), intercept 0.01, and R-squared 0.93. Voxel-wise Spearman correlation was 0.51 ± 0.10 (mean ± SD), indicating moderate spatial agreement. Dice coefficient for high-ventilation regions was 0.50 ± 0.07, demonstrating that both methods identify similar highly ventilating regions. Visual comparison revealed concordant ventilation patterns for most patients, with HU-based maps displaying finer-scale heterogeneity compared to smoother Jacobian-based maps.
Conclusion
HU-based ventilation mapping using 5DCT shows strong agreement with Jacobian-based methods at whole-lung and regional levels. The technique derives ventilation from density changes across 25 measurements per voxel rather than displacement field derivatives, potentially offering improved spatial resolution. Future work will validate against external reference standards and evaluate clinical utility for functional avoidance treatment planning.