AI-powered volumetric and radiomic MRI brain analysis for prognostication and treatment response patterns in ALK+ NSCLC brain metastases: Comparative outcomes of brigatinib versus crizotinib.

Abstract

e20520 Background: Brigatinib (brig) has demonstrated improved intracranial disease control compared to crizotinib (criz) in ALK+ non-small cell lung cancer (NSCLC). In this study we provide detailed characterization into brain metastases response patterns using volumetric and radiomic analysis as well as predicting intracranial progression-free survival (iPFS), offering a promising, non-invasive approach using standard of care (SOC) imaging. Methods: We retrospectively analyzed 98 ALK+ NSCLC patients with brain metastases from ALTA-1L trial with baseline and first follow-up MRI (T1W post-contrast and T2W/FLAIR). Viable tumor volume, necrosis, and edema were quantified; 225 radiomic features were extracted per scan. Linear mixed-effects models and multivariate survival analyses were employed to correlate radiomics and volumetric change with iPFS, controlling for treatment and patient effects. Lesion-level features were aggregated to the patient level using volume-weighted averaging or by selecting features from the largest lesion. Model performance was assessed using C-index and AUC via cross-validation. Results: Of evaluable patients, 30 received brigatinib (429 lesions) and 40 received crizotinib (752 lesions). Brig treated patients exhibited significantly greater intracranial tumor burden reduction (median -61.6% vs -17.7%) and improved iPFS at 12 months (56.7% vs 7.5%) compared to criz. Brig achieved consistent tumor volume reduction regardless of baseline lesion size, while criz showed only stabilization. Early changes in specific T1PC texture radiomic features (MaxGrad_IH and HGLDE_NGLDM) correlated with long-term iPFS in brigatinib-treated lesions. Baseline radiomic features provided moderate prognostic value (C-index 0.68; AUCs 0.75 for 6-month and 0.72 for 12-month iPFS), with performance enhanced by incorporating longitudinal delta features (C-index 0.80; AUCs 0.84 and 0.87 for 6- and 12-month iPFS, respectively), particularly those derived from FLAIR sequences and aggregated across all lesions. The exploratory lesion-level volumetric response model showed moderate discrimination (AUC 0.70), though impacted by class imbalance. Conclusions: This study highlights the value of AI-powered volumetric and radiomic MRI analysis for non-invasive prognostication and monitoring, using SOC imaging. The study reiterated superior intracranial efficacy of brig over criz across various volumetric response patterns. Longitudinal delta radiomic features, significantly enhance prediction of iPFS compared to baseline features alone in ALK+ NSCLC patients receiving TKI therapy. Brig not only achieves superior intracranial disease control but also enables identification of radiomic markers predictive of long-term benefit, supporting its use as a first-line therapy. Clinical trial information: NCT02737501 .