Commentary: CT-based radiomics integrated model for brain metastases in stage III/IV ALK-positive lung adenocarcinoma patients

Yun QiaoNan Wang^*

• Changzhi People's Hospital, Changzhi, China

IntroductionWe read with great interest the article by Wang et al. (2025) entitled “CT-based radiomics integrated model for brain metastases in stage III/IV ALK-positive lung adenocarcinoma patients” [1]. The study represents a valuable effort in exploring radiomics as a non-invasive biomarker for stratifying brain metastasis (BM) risk in a clinically challenging subset of non-small cell lung cancer (NSCLC) patients with ALK rearrangements. The integration of radiomics features with clinicoradiological parameters into a nomogram is methodologically robust and offers promise for early identification of high-risk individuals. The use of external validation and decision curve analysis (DCA) enhances the generalizability and clinical relevance of the findings. Despite these strengths, several important methodological and conceptual issues merit discussion to improve future research in this domain.DiscussionOne major concern lies in the small sample size, especially in the external validation cohort (n=34). While the inclusion of an independent dataset is commendable, the limited number of BM+ patients (n=16) reduces statistical power and may explain the inconsistent prognostic performance of the nomogram in this group (P=0.130). Previous radiomics studies have emphasized the importance of large, multicenter datasets to reduce overfitting and improve model stability and reproducibility [2].Furthermore, although the study employed regularization techniques such as LASSO and SMOTE, the final model retained only two radiomics features out of 1834, raising concerns about overfitting and the biological interpretability of the model. Notably, the selected features were not thoroughly discussed in terms of their imaging phenotype or potential pathophysiological basis. Interpretable radiomics is increasingly advocated in precision oncology to facilitate clinical adoption [3].Another critical issue relates to the use of CT rather than MRI for radiomics analysis. While CT is widely available and standardized, it may be less sensitive for characterizing parenchymal or subtle changes predictive of early BM, particularly in ALK+ populations known to have high intracranial tropism. A multimodal radiomics approach incorporating MRI could enhance predictive performance, as shown in recent brain metastasis studies [4].The decision to stratify risk using X-tile-derived Rad_score thresholds, although practical, may not be optimal for generalizability. Thresholds derived from internal data may not translate across institutions due to variations in imaging protocols or population characteristics. Calibration techniques or external derivation of cutoffs could mitigate this limitation [5].Finally, while the authors correctly identified lymphadenopathy and carcinomatous lymphangitis as clinical predictors of BM, these factors are relatively late manifestations of systemic spread and may limit the utility of the model in early intervention. Future studies might incorporate molecular data (e.g., TP53 co-mutations, ALK resistance mutations) to improve preclinical prediction, as integrative radiogenomic approaches are gaining traction in NSCLC [6].ConclusionIn summary, Wang et al. provide a promising framework for non-invasive prediction of brain metastases in ALK-positive lung adenocarcinoma patients. However, to enhance clinical translatability, future research should prioritize larger multicenter cohorts, biological interpretability of features, multimodal imaging integration, and external validation of risk thresholds.