Multi-Omics Model Accurately Predicts Radiotherapy Response in SCLC Brain Metastases

Key Clinical Summary:

• Design/Population: A retrospective multi-omics analysis included 144 patients with small cell lung cancer brain metastases treated with whole-brain radiotherapy between January 2020 and June 2024. The study integrated clinical, radiomic and dosiomic factors to predict short-term response.
• Key Outcomes: Concurrent chemoradiotherapy (P = .042) and conformal boost radiotherapy (P = .027) were independent clinical predictors of response. A hybrid feature model combining clinical, radiomic, and dosiomic features achieved the highest predictive performance with AUC 0.792 in training and 0.711 in external validation. Nomogram tools based on the model showed favorable clinical applicability.
• Clinical Relevance: Integrating dosiomics with radiomics and clinical factors improves prediction of whole-brain radiotherapy in patients with small cell lung cancer brain metastases, supporting personalized radiotherapy planning and treatment selection to potentially improve patient outcomes.

According to results from a retorspecitve analysis, a multi-omics model integrating  clinical, radiomic, and dosiomic features demonstrated high predictive accuracy for response to whole-brain radiotherapy (WBRT) among patients with small cell lung cancer (SCLC) brain metastases. 

“Dosiomics and radiomics elaborate the low-and high-order features extracted from images to predict clinical outcomes,” stated Yifan Lei, MD, Kunming Medical University, Yunnan, China, and coauthors. “The study seeks to develop accurate machine learning models to predict the radiotherapy response of WBRT.”

In this study, researchers collected data from 144 patients with SCLC brain metastases who underwent WBRT at a single institution between January 2020 and June 2024. Radiomic and dosiomic features were extracted from pre-treatment CT images, and treatment planning system dose data were extracted from radiotherapy dose maps using 3D Slicer software. These features were subsequently screened using LASSO and Logistic Regression models to evaluate associations between response and WBRT.  

Patients were identified as either responders (complete or partial response; n= 74) or non-responders (stable or progressive disease; n= 70), and classification models were built incorporating combinations of clinical factors, radiomic features, dosiomics features, and their various integrations. A hybrid feature model combining all 3 domains was built to serve as the primary model. Key outcomes included assessment of predictive performance and the development of nomograms to visualize individual treatment response. Additionally, a subset of patients were analyzed in an external validation set. 

Multivariate analysis identified concurrent chemoradiotherapy (P = 0.042), conformal boost radiotherapy (P = 0.027), and select radiomic and dosiomic features as independent predictors of response to WBRT. Based on univariate analysis combined with least absolute shrinkage and selection operator regression, 3 dosiomic features and 4 radiomic features were selected from an initial set of 851 features.

The hybrid model achieved the highest predictive performance, with a mean area under the curve of 0.792 in the training cohort and 0.711 in a prospectively collected external validation cohort. Nomogram tools developed from the hybrid model demonstrated favorable clinical applicability.

 “The integration of clinical parameters with dosiomics and radiomic features in a multi-omics framework demonstrates enhanced predictive accuracy for assessing [WBRT] outcomes in [SCLC],”concluded Dr Lei et al. “This comprehensive approach may facilitate clinical decision-making by enabling more precise treatment customization and individualized therapeutic strategies.”

Source:

Lei Y, Bai H, Gong C, et al. Multi‐omics predicts radiotherapy response in small cell lung cancer patients receiving whole brain irradiation. J Appl Clin Med Phys. Published online January 22, 2026. doi:10.1002/acm2.70466