ESTRO 2025 - Abstract Book

S3822

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

3427

Digital Poster Temporal analysis of 4DCT subregional respiratory dynamics based on machine learning for lung function assessment Zihan Li 1 , Yu-Hua Huang 1 , Zhi Chen 1 , Bing Li 2 , Hong Ge 2 , Jing Cai 1 , Ge Ren 1 1 Department of Health Technology and Informatics, The Hong Kong Polytechnic University, Hong Kong SAR, Hong Kong. 2 Department of Radiation Oncology, The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, Zhengzhou, China Purpose/Objective: Algorithms based on subregional respiratory dynamics (SRD) have been proposed to effectively capture spatiotemporal heterogeneity in the ventilation process, though relied on empirical modelings to map surrogate ventilation from SRD measurements. Given that avoidance of normal lung tissue during radiotherapy reduces the risk of radiation-induced injury, this study investigated the feasibility of a machine learning approach for temporally analyzing SRD extracted from four-dimensional computed tomography (4DCT) scans, utilizing a dual-path recurrent neural network (DPRNN) that integrates local and global cyclical respiratory patterns for function assessment. retrospectively collected from the VAMPIRE challenge, and split into training (n=35) and testing (n=11) cohorts. Lung parenchyma was partitioned into anatomically constrained subregions on the end-expiratory phase image. Respiratory dynamics were characterized through SRD extraction, capturing both local subregional changes and global whole-lung patterns across breathing phases. DPRNN was designed to classify subregions as normal or defective functions, with separate paths processing intensity and volume changes, enhanced by positional encodings representing respiratory cycle information. Data augmentation through cyclic shifting was implemented to capture phase-invariant features in the respiratory cycle. The ventilation distribution from DPRNN (V DPRNN ) was generated by interpolating model probabilities of subregion-wise classification. Both V DPRNN and empirical SRD based ventilation maps (V ESRD ) were created for the testing cohort and evaluated against V ref through voxel-wise correlation analysis and functional region overlap metrics. Material/Methods: 46 lung cancer patients with paired 4DCT and nuclear medicine-based ventilation reference (V ref ) were

The overall workflow.

Results: DPRNN showed promising performance in distinguishing functional from defective lung subregions, with the area under the receiver operating characteristic curve of 0.709 and accuracy of 0.727 for the training cohort, and 0.695 and 0.696 for the testing cohort, respectively. The generated V DPRNN ventilation maps showed significantly improved correlations with V ref scans compared to V ESRD (Spearman =0.61±0.11 vs V ESRD , p=0.0004). When evaluating spatial concordance of high-functioning (>66th percentile) and low-functioning (<33rd percentile) regions, V DPRNN showed significantly improved Dice coefficients of 0.62±0.05 and 0.63±0.07, respectively (p=0.0277 and 0.0054 vs V ESRD ).