Manual Delineation approaches for direct imaging of the subcortex

The growing interest in the human subcortex is accompanied by an increasing number of parcellation procedures to identify deep brain structures in magnetic resonance imaging (MRI) contrasts. Manual procedures continue to form the gold standard for parcellating brain structures and is used for the validation of automated approaches. Performing manual parcellations is a tedious process which requires a systematic and reproducible approach. For this purpose, we created a series of anatomical protocols for the delineation of 21 individual subcortical structures. These protocols are augmented with three example MRI datasets combined with their manual delineations. The intelligibility of the protocols was assessed by calculating Dice similarity coefficients which showed that manual parcellations created using these protocols can provide high quality training data for automated algorithms. The protocols can be applied to create high quality training data for automated parcellation procedures, as well as for further validation of existing procedures and are shared without restrictions with the research community.

随着人类大脑皮层下结构研究热度持续攀升，用于在磁共振成像（MRI）对比度图像中识别深部脑结构的脑分区（parcellation）方法也日益丰富。人工脑区分割方法至今仍是脑结构分区的金标准，常被用于自动化分割方法的性能验证。人工脑区分割工作繁琐耗时，需要采用系统化且可重复的操作流程。为此，我们针对21个独立皮层下结构的影像轮廓勾画，制定了一系列解剖学规程。该套规程配套了3例带有手动勾画结果的磁共振成像数据集作为示例。我们通过计算戴斯相似系数（Dice similarity coefficient）评估了该套规程的可操作性，结果显示，依据本规程完成的人工脑区分割结果，可为自动化算法提供高质量的训练数据。本套规程既可用于生成自动化脑分区方法所需的高质量训练数据，也可用于现有自动化分割流程的进一步验证，且已无限制地向全球科研社区开放共享。