中枢神经系统淋巴瘤血浆ctDNA阳性相关的MRI特征

RATIONALE AND OBJECTIVES: Peripheral-blood circulating tumor DNA (ctDNA) detectability is inconsistent in central nervous system lymphoma (CNSL), so negative plasma findings are hard to interpret. This exploratory, hypothesis-generating study tested whether routinely available, reproducible multiparametric Magnetic resonance imaging (MRI) features are associated with plasma ctDNA positivity across repeated paired MRI-plasma assessments.

MATERIALS AND METHODS: In this retrospective single-center study, 25 patients with diffuse large B-cell lymphoma involving the CNS (primary or secondary) contributed 46 paired plasma ctDNA and contrast-enhanced brain MRI evaluations. MRI features were prospectively defined and independently annotated by two blinded radiologists; features were retained only with substantial inter-reader agreement (kappa/weighted kappa ≥0.60) or good ICC (≥0.75). Associations with ctDNA positivity (variant allele frequency ≥0.5%) were screened using logistic generalized estimating equations (GEEs) for within-patient correlation, prioritized by patient-level bootstrap stability selection with L1-penalized logistic regression, and summarized in adjusted multivariable models (age at CNS diagnosis and disease type).

RESULTS: Among 31 reproducible MRI-derived features, edema crossing the midline had the highest selection frequency (55.8%). In the primary adjusted multivariable model, edema crossing the midline was linked to markedly lower odds of plasma ctDNA positivity (Firth-penalized OR 0.068). To address within-patient correlation from repeated evaluations, uncertainty was estimated by patient-level cluster bootstrap (B = 1000; percentile method), giving a 95% confidence interval of 0.00031-0.236 with complete directional consistency across resamples (Pr[OR < 1] = 1.000; bootstrap failure rate 0.3%). Sensitivity analyses indicated that high non-contrast-enhancing tumor (nCET) burden (≥34%) was associated with higher ctDNA positivity odds, while mass effect was associated with lower odds, though both showed sparse-data patterns.

CONCLUSION: Reproducible MRI features, especially edema crossing the midline, were inversely associated with plasma ctDNA positivity across repeated evaluations. Patient-level cluster-bootstrap inference supported the robustness of the observed association. External validation in larger cohorts is needed before clinical implementation.