Dynamic Contrast Enhanced (DCE) MR Perfusion Imaging Harbors Aberrantly Low Specificity in Distinguishing Post-Radiation Recurrence from Radiation Necrosis in Melanoma Brain Metastasis

Introduction: Approximately 15% of brain metastases (BrM) treated with SRS progress within 12 months. Distinguishing between recurrent brain metastases (rBrM) and radiation necrosis (RN) is a key challenge in BrM care. Advanced imaging is increasingly used as a noninvasive method to differentiate these processes, which carry different prognoses and management strategies. However, the specificity of MR perfusion has not been established in melanoma, where melanocytic and hemorrhagic changes can confound traditional imaging methods. We hypothesize that rBrM exhibit paradoxically low plasma volume (Vp), heightening risk of misdiagnosis and inappropriate treatment. This study aims to assess the predictive value of Vp in distinguishing rBrM from RN in melanoma patients.

Methods: Patients with post-SRS recurrent melanoma BrM, resected between 1/2014 and 9/2024, with pre-SRS and pre-resection MR perfusion imaging were retrospectively evaluated using MR dynamic contrast-enhanced perfusion imaging. Perfusion parameters were used to classify lesions as Vp(elevated) or Vp(suppressed) by a blinded neuroradiologist; this classification was compared to surgical pathology to evaluate the predictive value of Vp. Additional quantitative and radiomic analyses are ongoing.

Results: Forty-five lesions in 43 patients were included: 36 (80%) with pathologically-diagnosed rBrM, and 9 (20%) RN. Of the 9 RN cases, 2 (22%) were Vp(suppressed), a surrogate for RN, while 7 (78%) were Vp(elevated), a surrogate for rBrM. While 19/36 (53%) tissue-confirmed rBrM were Vp(elevated), the remaining 17 (47%) were Vp(suppressed). The positive predictive value of MR perfusion for rBrM was 73%, while negative predictive value was 11%.

Conclusion : Our findings indicate that MR perfusion has a much lower predictive value in distinguishing rBrM versus RN in melanoma patients compared to historical basket cohorts. Careful multidisciplinary, radiomic, and multimodality consideration is necessary to optimize noninvasive risk stratification, as this has significant implications for patient surveillance and intervention.