Article
Non-invasive prediction of the BRAF mutation status in patients with melanoma brain metastases using MRI radiomics
MRT Radiomics zur nicht-invasiven Vorhersage des BRAF-Mutationsstatus in Patienten mit zerebralen Melanommetastasen
Search Medline for
Authors
Published: | June 4, 2021 |
---|
Outline
Text
Objective: Malignant melanomas frequently develop brain metastases. Targeted therapies such as BRAF inhibitors have shown relevant intracranial response rates in initial trials. The most frequent BRAF V600E mutation occurs in up to 60% of melanomas. However, recent studies reported a significant discrepancy with respect to the mutational status of the extracranial melanoma and melanoma brain metastases. We evaluated the potential of MR-radiomics for the non-invasive diagnosis of BRAF status in brain metastases.
Methods: Forty-four patients with 48 brain metastases secondary to malignant melanoma underwent brain tumor surgery with subsequent molecular tissue analysis of the BRAF mutational status. All patients received a preoperative MRI including contrast-enhanced (CE) T1- and T2-weighted sequences. Brain metastases were manually segmented based on CE T1-weigthed MRI. After image preprocessing 2632 radiomics features were extracted using pyradiomics. Patients were randomly assigned to a training and a test dataset in a ratio of 50/50. To test predictive performance multiple algorithms were trained based on 5-fold cross validation. Finally, the best performing model was applied to the test dataset to evaluate model generalizability.
Results: Twenty-one patients (48%) showed a BRAF V600E mutation. In the test dataset, the radiomics model using a support vector machine classifier yielded an area under the receiver operating characteristic curve (AUC) of 0.8 (accuracy, 77%; sensitivity, 70%; specificity, 83%) for prediction of the BRAF mutation status.
Conclusion: MRI radiomics allows a non-invasive prediction of the BRAF V600E mutation status in patients with melanoma brain metastases. Further evaluation of the developed model is ongoing.