Clin Transl Oncol. 2026 Mar 22. doi: 10.1007/s12094-026-04311-x. Online ahead of print.
ABSTRACT
AIM: Four-dimensional computed tomography (4D-CT) is the gold standard for radiotherapy planning in non-small cell lung cancer (NSCLC), yet its use in radiomics remains underexplored. This study proposes a reproducible, scalable methodology for assessing radiomic feature (RF) stability in 4D-CT and evaluates whether image filtering identifies additional stable RFs compared to unfiltered images.
METHODS: Early-stage NSCLC patients treated with SBRT with 4D-CT were included. Gross tumor volumes (GTVs) were re-segmented on all available phases. RFs were extracted using PyRadiomics. Features with near-zero variance in > 85% of patients were excluded. RF stability was evaluated using two complementary approaches: (i) coefficient of variation (COV), quantifying the magnitude of inter-phase variability, and (ii) repeated-measures modeling, assessing the presence of a statistically significant association between RF values and respiratory phase. RFs with COV < 5% and 5-10% were considered highly stable and stable, respectively. Repeated-measures analyses were performed separately for expiratory (0-40%) and inspiratory (50-90%) phases.
RESULTS: Seventy patients met the inclusion criteria. 1892 RFs were analyzable. Based on COV, about 21% (397/1892) of RFs were highly stable, and 18% (338/1892) were stable, while the remaining showed intermediate or high variability. The largest proportion of highly stable RFs derived from lbp-3D (25%) and log-sigma (12%) filtered images. Repeated measures analysis showed that only a limited subset of RFs had a statistically-significant dependence on respiratory phase, with 1747 and 1744 RFs remaining time-independent across expiratory and inspiratory phases, respectively.
CONCLUSION: Radiomic features extracted from 4D-CT images in early-stage NSCLC patients show heterogeneous stability across respiratory phases. Radiomic features extracted from 4D-CT images in early-stage NSCLC exhibit heterogeneous quantitative variability across respiratory phases. However, only a minority of features show statistically significant time dependence. The study provides a reproducible methodological framework to identify stable radiomic features from 4D-CT, enabling their more reliable use in lung cancer radiomic studies.
PMID:41865337 | DOI:10.1007/s12094-026-04311-x
