DBSegment: Fast and robust segmentation of deep brain structures -- Evaluation of transportability across acquisition domains

Segmenting deep brain structures from magnetic resonance images is important\nfor patient diagnosis, surgical planning, and research. Most current\nstate-of-the-art solutions follow a segmentation-by-registration approach,\nwhere subject MRIs are mapped to a template with well-defined segmentations.\nHowever, registration-based pipelines are time-consuming, thus, limiting their\nclinical use. This paper uses deep learning to provide a robust and efficient\ndeep brain segmentation solution. The method consists of a pre-processing step\nto conform all MRI images to the same orientation, followed by a convolutional\nneural network using the nnU-Net framework. We use a total of 14 datasets from\nboth research and clinical collections. Of these, seven were used for training\nand validation and seven were retained for independent testing. We trained the\nnetwork to segment 30 deep brain structures, as well as a brain mask, using\nlabels generated from a registration-based approach. We evaluated the\ngeneralizability of the network by performing a leave-one-dataset-out\ncross-validation, and extensive testing on external datasets. Furthermore, we\nassessed cross-domain transportability by evaluating the results separately on\ndifferent domains. We achieved an average DSC of 0.89 $\\pm$ 0.04 on the\nindependent testing datasets when compared to the registration-based gold\nstandard. On our test system, the computation time decreased from 42 minutes\nfor a reference registration-based pipeline to 1 minute. Our proposed method is\nfast, robust, and generalizes with high reliability. It can be extended to the\nsegmentation of other brain structures. The method is publicly available on\nGitHub, as well as a pip package for convenient usage.\n