A Non-random Spatial Distribution of Brain Metastases

Abstract

We sought to quantitatively describe the spatial distribution of brain metastases (BM) in a single institution experience treating metastatic disease to the brain, with the alternative hypothesis that BM spatial frequency does not follow a stochastic distribution. It has been demonstrated that metastatic disease infrequently occurs in the hippocampus, leading to a low risk of failure within the region when using hippocampal-sparing whole brain radiotherapy protocols. Probabilistic mapping of BM to other regions of the brain, however, has not been previously quantified. We retrospectively analyzed patients with newly diagnosed and recurrent brain metastases from 2006-2016. Each lesion was manually defined on MRI with T1 gadolinium-enhanced imaging. Standard brain atlases dividing the brain into 52 anatomical regions were used to identify the frequency of metastasis in each region. This was compared to the expected probability of occurrence, which was assumed to be a random distribution based on region volume. The quantified BM maps were analyzed across different primary cancers and histologic subtypes. Results indicate a non-random distribution of brain metastases that agree with prior reports of a low risk of developing disease in the hippocampi. Analyses revealed additional regions – the frontal pole, lateral occipital cortex, supramaginal gyrus, and temporal pole – had a lower incidence of BM than predicted. We also demonstrate an increase in risk of involvement for several regions, notably the precentral gyrus and paracingulate gyrus relative to what would be expected based on brain volume (p<0.01, corrected for multiple comparisons). Clinically, a more granular understanding of BM distribution may permit sparing of additional brain regions with partial brain radiotherapy. This analysis forms the framework for a deterministic model to approximate the non-stochastic process of BM formation. Further analyses correlating prognostic factors such as age, histology, previous therapeutic interventions and additional imaging data (e.g. cerebral blood flow mapping) with our current findings are necessary to further develop this model.