https://atlas.rsna.org/schemas/2025-11/model.json

Peri- and Intratumoral Radiomic Features on Baseline CT Scans Predicts Chemotherapy Response in Lung Adenocarcinoma

https://dx.doi.org/10.1148/ryai.2019180012

1.0

Monica Khunger, ude.esac@067kxm (as listed)

Supported by the U.S. Department of Defense, the National Cancer Institute, the National Institute of Diabetes and Digestive and Kidney Diseases, the National Center for Research Resources, the Ohio Third Frontier Technology Validation Fund, the Wallace H. Coulter Foundation Program in the Department of Biomedical Engineering, and the Clinical and Translational Science Award Program at Case Western Reserve University.

HIPAA-compliant, IRB-approved study with waiver of informed consent (Cleveland Clinic IRB #14-562).

Quadratic Discriminant Analysis (QDA) classifier for response prediction; LASSO Cox proportional hazards models to derive radiomics risk-score signatures for TTP and OS.

Noninvasive prediction of chemotherapy response and risk stratification for time to progression and overall survival in NSCLC from baseline CT; potential to identify high-risk patients who may benefit from more intensive follow-up or alternative strategies.

clinical decision support systems

Radiomics TTP risk-score cutoff of 0.1 used to stratify patients into high- and low-risk groups (selected to maximize χ2 in training).

Patients with advanced NSCLC (adenocarcinoma; stage IIIB/IV) undergoing first-line pemetrexed-based platinum doublet chemotherapy; uses baseline non–contrast-enhanced chest CT to predict response and assess risk for TTP and OS.

Baseline non–contrast-enhanced thoracic CT images; intratumoral and peritumoral (up to ~15 mm) radiomic texture and shape features.

Manually segment index lung lesion; define peritumoral ring by radial dilation to ~15 mm; extract stable, reproducible radiomic features (ICC ≥ 0.8 via RIDER test-retest and inter-reader analysis); select features via minimum redundancy maximum relevance; train QDA on balanced training cohort; evaluate on independent validation set. Construct LASSO Cox radiomics signatures for TTP/OS and stratify by risk-score cutoff.

Single-institution retrospective study with relatively small sample size; variability in CT acquisition parameters (e.g., section thickness) with noted performance decrease at thicker sections; dependence on manual lesion annotation; lack of actionable mutation and PD-1/PD-L1 data; generalizability not yet established; did not explicitly evaluate effects of other acquisition parameters.

Further large-scale, multisite validation is needed before clinical deployment.

Feature reproducibility assessed using RIDER lung CT test-retest (ICC ≥ 0.8 threshold) and inter-reader stability across three radiologists; segmentation agreement quantified by Dice similarity and over-/under-segmentation errors; consistent classifier AUCs across readers (validation AUCs 0.76–0.78).