Fetal Brain Component Segmentation Using 2-Way Ensemble U-Net

Fetal brain segmentation has been a field of interest since a long time. However, it is a challenging task as well for reasons, like blurred images due to fetal motion. Recently, deep learning has been successful in performing this task with good accuracy. In this paper, we developed 2-way Ensemble U-Net model, a convolutional neural network architecture for performing segmentation on the fetal brain image to divide it into its seven components: intracranial space and extra-axial cerebrospinal fluid spaces, gray matter, white matter, ventricles, cerebellum, deep gray matter, and brainstem and spinal cord. The fetal brain image can be obtained by segmenting it from the fetal magnetic resonance images using any of the previous works on fetal brain segmentation, which presents our work as an extension of the already existing segmentation works. The Jaccard similarity and Dice score for this task are 83% and 88%, respectively. This is higher than that returned by any of the previous models, when trained for the same task, thus showing the potential of our model in segmentation related tasks.