IMPORTANCE Brain metastases are a common source of morbidity for patients with cancer, and limited data exist to support the local therapeutic choice between surgical resection and stereotactic radiosurgery (SRS). OBJECTIVE To evaluate local control of brain metastases among patients treated with SRS vs surgical resection within the European Organization for the Research and Treatment of Cancer (EORTC) 22952-26001 phase 3 trial.

PurposeStereotactic radiosurgery (SRS) alone is an increasingly common treatment strategy for brain metastases. However, existing prognostic tools for overall survival (OS) were developed using cohorts of patients treated predominantly with approaches other than SRS alone. Therefore, we devised novel risk scores for OS and distant brain failure (DF) for melanoma brain metastases (MBM) treated with SRS alone.Methods and materialsWe retrospectively reviewed 86 patients treated with SRS alone for MBM from 2009-2014. OS and DF were estimated using the Kaplan-Meier method. Cox proportional hazards modeling identified clinical risk factors. Risk scores were created based on weighted regression coefficients. OS scores range from 0-10 (0 representing best OS), and DF risk scores range from 0-5 (0 representing lowest risk of DF). Predictive power was evaluated using c-index statistics. Bootstrapping with 200 resamples tested model stability.ResultsThe median OS was 8.1 months from SRS, and 54 (70.1 %) patients had DF at a median of 3.3 months. Risk scores for OS were predicated on performance status, extracranial disease (ED) status, number of lesions, and gender. Median OS for the low-risk group (0-3 points) was not reached. For the moderate-risk (4-6 points) and high-risk (6.5-10) groups, median OS was 7.6 months and 2.4 months, respectively (p < .0001). Scores for DF were predicated on performance status, ED status, and number of lesions. Median time to DF for the low-risk group (0 points) was not reached. For the moderate-risk (1-2 points) and high-risk (3-5 points) groups, time to DF was 4.8 and 2.0 months, respectively (p < .0001). The novel scores were more predictive (c-index = 0.72) than melanoma-specific graded prognostic assessment or RTOG recursive partitioning analysis tools (c-index = 0.66 and 0.57, respectively).ConclusionsWe devised novel risk scores for MBM treated with SRS alone. These scores have implications for prognosis and treatment strategy selection (SRS versus whole-brain radiotherapy).

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

PurposeState‐of‐the‐art automated segmentation methods achieve exceptionally high performance on the Brain Tumor Segmentation (BraTS) challenge, a dataset of uniformly processed and standardized magnetic resonance generated images (MRIs) of gliomas. However, a reasonable concern is that these models may not fare well on clinical MRIs that do not belong to the specially curated BraTS dataset. Research using the previous generation of deep learning models indicates significant performance loss on cross‐institutional predictions. Here, we evaluate the cross‐institutional applicability and generalzsability of state‐of‐the‐art deep learning models on new clinical data.MethodsWe train a state‐of‐the‐art 3D U‐Net model on the conventional BraTS dataset comprising low‐ and high‐grade gliomas. We then evaluate the performance of this model for automatic tumor segmentation of brain tumors on in‐house clinical data. This dataset contains MRIs of different tumor types, resolutions, and standardization than those found in the BraTS dataset. Ground truth segmentations to validate the automated segmentation for in‐house clinical data were obtained from expert radiation oncologists.ResultsWe report average Dice scores of 0.764, 0.648, and 0.61 for the whole tumor, tumor core, and enhancing tumor, respectively, in the clinical MRIs. These means are higher than numbers reported previously on same institution and cross‐institution datasets of different origin using different methods. There is no statistically significant difference when comparing the dice scores to the inter‐annotation variability between two expert clinical radiation oncologists. Although performance on the clinical data is lower than on the BraTS data, these numbers indicate that models trained on the BraTS dataset have impressive segmentation performance on previously unseen images obtained at a separate clinical institution. These images differ in the imaging resolutions, standardization pipelines, and tumor types from the BraTS data.ConclusionsState‐of‐the‐art deep learning models demonstrate promising performance on cross‐institutional predictions. They considerably improve on previous models and can transfer knowledge to new types of brain tumors without additional modeling.