Geometric Prognostic Markers Of Brain Metastasis Based On Voxelwise Analysis In MRI

PurposeNoninvasive prognostic prediction of brain metastasis(BM)remains highly challenging due to the diverse biological heterogeneity.The present study aims to investigate the prognostic imaging markers of BM based on voxelwise geometric analysis.Methods:1102 patients with a total of 4365 BMs in our institution from January 2012 to June2017 were reviewed.Three MRI geometric features,the number,volume and location,were evaluated.These features were statistically analyzed by age,sex,survival time,primary malignancies,and the Ki-67 index of primary tumors and BMs,and further visualized by heatmaps.Cox proportional hazard regression was used to analyze independent prognostic factors.Results:The parietal,insular and left occipital lobes and the cerebellum were more vulnerable to BM.Lesions from lung squamous cell carcinoma(LSC)had a larger normalized volume but occurred less frequently than those from lung adenocarcinoma.Infratentorial BM was associated with young age,male sex,lung neuroendocrine carcinoma,LSC and high BM Ki-67 index.Cox regression indicated that old age(p<.001,hazard ratio(HR)=1.283,95% confidence interval(CI)=1.119-1.471),male sex(p<.001,HR=1.414,95% CI=1.222-1.636),melanoma BM(p=.032,HR=2.264,95% CI=1.072-4.782)and ≥ 3 BMs(p<.001,HR=1.330,95% CI=1.142-1.549)independently predicted poor prognosis.For solitary BMs,male sex(p<.001,HR=1.550,95% CI=1.246-1.928)was the only independent risk factor,while solitary BMs in the parietal(p=.006,HR=0.633,95% CI=0.457-0.875)and occipital(p=.033,HR=0.712,95% CI=0.520-0.973)lobes and the limbic system(p=.018,HR=0.504,95% CI=0.286-0.889)could be independent determinants for a better prognosis.Conclusion:Imaging markers based on voxelwise geometric analysis may provide valuable insights for prognostic prediction for patients with BM.