Prostate Segmentation using 2D Bridged U-net

Abstract

In this paper, we focus on three problems in deep learning based medical image segmentation. Firstly, U-net, as a popular model for medical image segmentation, is difficult to train when convolutional layers increase even though a deeper network usually has a better generalization ability because of more learnable parameters. Secondly, the exponential ReLU (ELU), as an alternative of ReLU, is not much different from ReLU when the network of interest gets deep. Thirdly, the Dice loss, as one of the pervasive loss functions for medical image segmentation, is not effective when the prediction is close to ground truth and will cause oscillation during training. To address the aforementioned three problems, we propose and validate a deeper network that can fit medical image datasets that are usually small in the sample size. Meanwhile, we propose a new loss function to accelerate the learning process and a combination of different activation functions to improve the network performance. Our experimental results suggest that our network is comparable or superior to state-of-the-art methods.

SE-UXNet: A Medical Imaging Segmentation Network Based on Attention Mechanism

“…All experiments were done in python, using Keras [15] with a Tensorflow [16] backend. The Stochastic Gradient Descent optimization function is used to optimize the network parameters, and the cos dice loss [4] is used to measure the classification error of the network. The initial learning rate is set to 0.001, and the batch size for training and testing was kept at 8.…”

“…However, due to the defects of the network structure itself, it is difficult for U-Net and U-Net++ to construct a well-behaved deep network. Compared to a single U-Net, Wanli Chen et al [4] proposed that the use of a bridge structure helps to solve this problem by expanding model capacity. As an important attention structure, SE block [5] can be easily embedded in skip connections.…”

A Multiple Layer U-Net, Un-Net, for Liver and Liver Tumor Segmentation in CT

“…Several methods of automatic segmentation of liver and lesion have been proposed, consisting of level set parameter [8], [9], fast fuzzy c-means and adaptive watershed [10], [11], fully convolutional networks (FCNs) [12]- [15], segnet [16], encoder-decoder structure [17]- [23]. The most popular encoder-decoder architecture is the U-Net model [24] that has been modified to implement a lot of applications on medical image segmentation such as ischemic stroke lesion [25], pancreas [26], [27], retina vessel [28], [29], prostate [30], colorectal tumor [31], and brain tumor [32], etc.…”

M-Net: A Novel U-Net With Multi-Stream Feature Fusion and Multi-Scale Dilated Convolutions for Bile Ducts and Hepatolith Segmentation

“…Recently, more and more researchers have introduced deep learning into medical image segmentation [6]- [20] due to its excellent ability of self-learning from a large amount of data through its special convolutional structures. Among them, U-Net [6], [7], [9], [10], [15]- [19], [22]- [24], an end-toend full convolutional neural network, is a most promising network for medical image segmentation. Thanks to its skipconnection at different resolutions, more image details can be involved in the decoder process, resulting in better image segmentation [6].…”

“…Thanks to its skipconnection at different resolutions, more image details can be involved in the decoder process, resulting in better image segmentation [6]. A variety of researchers employed U-Net to segment tissues/organs or lesions such as brain tumor, left ventricle, prostate [7], [9], [10]. U-Net has an inherent architecture of pooling in the encoder process and interpolation in the decoder process, which will influence segmentation performance.…”