3D-DenseUNet: A New Model for BrainSegmentation in Magnetic Resonance Images

Abstract Recent deep learning models have attracted substantial attention in infant brainanalysis. These models have performed state-of-the-art performance, such as semisupervised techniques (e.g., Temporal Ensembling, mean teacher). However, thesemodels depend on an encoder-decoder structure with stacked local operators togather long-range information, and the local operators limit the efficiency and effectiveness. Besides, the MRI data contain different tissue properties (T P s) suchas T1 and T2. One major limitation of these models is that they use both dataas inputs to the segment process, i.e., the models are trained on the dataset once,and it requires much computational and memory requirements during inference. In this work, we address the above limitations by designing a new deep-learningmodel, called 3D-DenseUNet, which works as adaptable global aggregation blocksin down-sampling to solve the issue of spatial information loss. The self-attentionmodule connects the down-sampling blocks to up-sampling blocks, and integrates the feature maps in three dimensions of spatial and channel, effectively improvingthe representation potential and discriminating ability of the model. Additionally,we propose a new method called Two Independent Teachers (2IT), that summarizes the model weights instead of label predictions. Each teacher model is trainedon different types of brain data, T1 and T2, respectively. Then, a fuse model isadded to improve test accuracy and enable training with fewer parameters andlabels compared to the Temporal Ensembling method without modifying the network architecture. Empirical results demonstrate the effectiveness of the proposedmethod. The code is publicly available at: https://github.com/AfifaKhaled/Two-Independent-Teachers-are-Better-Role-Model