White Matter Segmentation in Brain MR Images of Premature Infants using Deep Learning Model

Abstract

White matter injury is the most common form of brain injury in premature infants, which is highly associated with poor brain development. White matter segmentation in brain MR images is very important for early detection of brain injury in premature infants. In this paper, a deep learning based approach is proposed to segment the white matter region of premature infants. We retrospectively analyzed 63 preterm infants with gestational age < 37 weeks who underwent MRI examination. Firstly, the U-Net deep learning neural network model is used to automatically segment white matter, U-Net++ and ResU-Net are used as controlled experimental groups, and the Dice coefficient and Iou coefficient are used as evaluation indicators for white matter segmentation. In the above experiments, the U-Net network has the best experimental results. Segmenting white matter on the training set with a Dice coefficient of 0.90 and an Iou of 0.82, respectively. The test set was used to validate the performance of the proposed model, with a Dice coefficient of 0.83 and an Iou of 0.71, respectively. Secondly, we performed image binarization step on the segmented white matter, and used Gaussian filtering, flood filling and connected component analysis to process the automatically segmented white matter region, and finally obtain a white matter region with clear boundaries.