In this study, we perform a comparative analysis of automated image segmentation of subcortical structures in the elderly brain. Manual segmentation is very time-consuming and automated methods are gaining importance as a clinical tool for diagnosis. The two most commonly used software libraries for brain segmentation -FreeSurfer and FSL- are put to work in a large dataset of 4,028 magnetic resonance imaging (MRI) scans collected for this study. We find a lack of linear correlation between the segmentation volume estimates obtained from FreeSurfer and FSL. On the other hand, FreeSurfer volume estimates tend to be larger thanFSL estimates of the areas putamen, thalamus, amygdala, caudate, pallidum, hippocampus, and accumbens. The characterization of the performance of brain segmentation algorithms in large datasets as the one presented here is a necessary step towards partially or fully automated end-to-end neuroimaging workflow both in clinical and research settings.

在这项研究中，我们对老年人大脑皮层下结构的自动图像分割进行了比较分析。手动分割非常耗时，自动化方法作为临床诊断工具越来越重要。用于大脑分割的两个最常用的软件库 - FreeSurfer 和 FSL - 用于为本研究收集的 4,028 个磁共振成像 (MRI) 扫描的大型数据集。我们发现从 FreeSurfer 和 FSL 获得的分割体积估计之间缺乏线性相关性。另一方面，FreeSurfer 体积估计值往往大于壳核、丘脑、杏仁核、尾状核、苍白球、海马体和伏隔核区域的 FSL 估计值。