Spatiotemporal characterizations of thoracic four-dimensional computed tomography for functional lung avoidance radiation therapy

Abstract

Functional lung avoidance radiotherapy (FLA-RT) has emerged as a promising approach to protect lung function and reduce radiation-induced lung injury (RILI) in thoracic cancer treatment. The effectiveness of FLA-RT heavily relies on accurate functional lung imaging, which remains challenging due to the complex spatiotemporal heterogeneity of lung function. This thesis aims to improve computed tomography ventilation imaging (CTVI) techniques for FLA-RT applications by investigating the spatiotemporal heterogeneities captured by thoracic four-dimensional computed tomography (4DCT). We first introduced a novel 4DCT-based multi-phase lung ventilation imaging framework that recovers the entire ventilation process throughout the respiratory cycle, developed and validated using 4DCT scans from 15 lung or esophageal cancer patients. The framework utilizes the parameterized Integrated Jacobian Formulation concept to estimate local expansion distributions for each phase relative to the end-of-expiration (EE), generating dynamic surrogate ventilation images by warping these phase-specific distributions into their respective breathing phases. Quantitative analysis revealed significant spatiotemporal heterogeneities in ventilation distribution, with mean interphase Spearman correlation coefficients (SCC) ranging from 0.23 ± 0.20 to 0.93 ± 0.04, decreasing near the EE phase. Only 26.2% of lung voxels exhibited the same expansion/contraction pattern as the global lung. These investigations demonstrate the prevalent heterogeneity and nonlinearity of lung expansion and contraction patterns over time, revealing critical limitations in classical biphasic CTVI studies. Building on these insights, we developed a two-stage framework for robustly extracting and mapping ventilation surrogates based on the concept of subregional respiratory dynamics (SRD). This approach was validated using 50 subjects from three cohorts of the VAMPIRE challenge, containing 4DCT and reference ventilation imaging (RefVI) scans. The framework partitions lung subregions on the 4DCT EE phase using anatomically constrained simple linear iterative clustering, while performing sliding-preserved interphase image registrations. SRDs of breathing-induced volume and intensity changes were tracked for each lung subregion across phases, integrating mechanical collapsibility and physiological tissue density to construct voxel-level surrogate ventilation distribution ( ) maps. The resulting maps demonstrated improved spatial representation and imaging performance compared to classical biphasic CTVI metrics, with median SCC between and RefVI scans of 0.600, 0.582, and 0.561 for the three cohorts, versus 0.152, 0.245, and 0.365 for biphasic Jacobian maps. Median Dice similarity coefficients (DSC) against RefVI scans, showing the highest and lowest functioning lung regions’ concordances, were 0.611(0.626), 0.592(0.620), and 0.601(0.611) for maps, superior to biphasic Jacobian maps. To evaluate the clinical relevance of the proposed SRD-based ventilation analysis method, we conducted a dosimetric study comparing map-guided FLA-RT plans with conventional lung radiotherapy (ConvRT) and Jacobian-based FLA-RT plans. With 10 lung cancer patients with 4DCT scans, we retrospectively implemented contour-based (cFLA-RT) and voxel-based (vFLA-RT) FLA-RT strategies, using an automated planning framework to ensure consistency. For cFLA-RT, map-guided plans significantly reduced mean dose (0.71 Gy, p=0.004), V20 (1.76%, p=0.005), and V5 (3.75%, p=0.020) to reference avoidance structures (RefAS) compared to ConvRT. Similarly, for vFLA-RT, significant reductions were observed in RefAS mean dose (0.87 Gy, p=0.001), V20 (1.95%, p=0.046), and V5 (4.97%, p=0.008). map-guided plans also outperformed biphasic Jacobian-based FLA-RT plans, with significant improvements in RefAS dosimetric parameters for cFLA-RT and vFLA-RT strategies (p<0.05 for all). The dosimetric advantages were achieved while maintaining similar target coverage and keeping organ-at-­risk doses within acceptable limits. In conclusion, this thesis presents advancements in functional lung imaging for RT, offering new tools and methodologies that address critical gaps in current practice. The multi-phase ventilation imaging and SRD-based ventilation surrogate mapping approach provide a more comprehensive and accurate representation of lung function dynamics throughout the respiratory cycle. The dosimetric study demonstrates the potential of the improved imaging methodologies to enhance FLA-RT planning, potentially leading to reduced RILI without compromising tumor control. While further clinical validation is needed, the integration of the proposed techniques into clinical workflows could lead to more personalized and effective treatment strategies for lung cancer patients, ultimately improving their care and quality of life.