MRI-based intratumoral and peritumoral heterogeneity for predicting intracranial progression in EGFR-mutant lung adenocarcinoma patients with treatment-naive brain metastases receiving first-line third-generation EGFR-TKIs.

Abstract

Purpose: To develop and externally test a multiparametric MRI habitat-based framework integrating intratumoral heterogeneity (ITH), peritumoral heterogeneity (PTH), and clinical characteristics for predicting intracranial progression in patients with EGFR-mutant lung adenocarcinoma and de novo brain metastases receiving first-line third-generation (3rd-gen) EGFR-TKI monotherapy.

Materials and methods: This retrospective multicenter study included patients with EGFR-mutant lung adenocarcinoma and de novo brain metastases who received first-line 3rd-gen EGFR-TKI monotherapy. Patients were assigned to an internal training cohort (n = 90) and an external testing cohort (n = 51). The binary prediction outcome was intracranial progression within 20 months after treatment initiation. Pretreatment contrast-enhanced T1-weighted imaging, T2-weighted imaging, and diffusion-weighted imaging were analyzed. The index brain metastasis and a 6-mm peritumoral region were segmented, followed by voxel-wise habitat clustering. Habitat-specific radiomic features were extracted and selected to construct ITH, PTH, and combined intratumoral-peritumoral habitat (PITH) signatures. Clinical predictors were evaluated using univariable and multivariable logistic regression. Support vector machine models were evaluated using receiver operating characteristic, decision curve, calibration, and Kaplan-Meier analyses.

Results: A total of 141 patients were analyzed, including 75 with progressive disease (PD) and 66 without PD (non-PD). Multivariable logistic regression identified EGFR mutation subtype (odds ratio [OR] = 2.390, 95% confidence interval [CI]: 1.187-4.811, P = 0.015) and the maximum diameter of the index brain metastasis as independent predictors (OR = 2.048, 95% CI: 1.204-3.483, P = 0.008). Models based on imaging heterogeneity features outperformed the clinical model overall. The clinical model achieved an AUC of 0.662, while the ITH, PTH, and PITH models achieved AUCs of 0.715, 0.724, and 0.758, respectively. The combined model showed favorable performance (AUC 0.795, accuracy 76.5%, sensitivity 70.4%, and specificity 83.3%). Patients classified by the combined model as high risk had significantly shorter intracranial progression-free survival (iPFS), than those classified as low risk (log-rank P = 0.019).

Conclusion: The proposed framework integrating ITH, PTH, and clinical features may provide a noninvasive approach for predicting intracranial progression risk and stratifying prognosis in patients with EGFR-mutant lung adenocarcinoma with newly diagnosed BMs receiving first-line 3rd-gen EGFR-TKI monotherapy.