Brain tumor segmentation based on improved U-Net

Abstract

Background: Automatic segmentation of brain tumors using deep learning algorithms is one of the research hotspots in the field of medical image segmentation at this stage. An improved u-net network is proposed to segment brain tumors in order to improve the segmentation effect of brain tumors.Methods: In order to solve the problems that other brain tumor segmentation models such as U-Net have insufficient ability to segment edge details, poor extraction of location information and the commonly used Binary Cross-Entropy and Dice loss are often ineffective when used as loss functions for brain tumor segmentation models, we propose a serial encoding-decoding structure, which achieves improved segmentation performance by adding Hybrid Dilated Convolution (HDC) modules and the connections between each module of the two serial networks, in addition, we propose a new loss function in order to make the model more focused on samples that are difficult to segment and classify. We compared the results of our proposed model and the commonly used segmentation models under IOU, PA, Dice, Precision, Hausdorf95, and ASD metrics.Results: The performance of the proposed method outperforms other segmentation models in each metric. In addition, the schematic diagram of segmentation results shows that the segmentation results of our algorithm are closer to ground truth, showing more details of brain tumors, while the segmentation results of other algorithms are more smooth.Conclusions: Our algorithm has better semantic segmentation performance, compared with other commonly used segmentation algorithms. The technology we proposed can be used in the diagnosis of brain tumors to provide better protection for patients' later treatment.

Defect Detection of Solar Panel Based on DenseNet Network

“…This dense connection mode enables each layer of the network to learn the characteristics of the previous multiple layers and reuse the characteristics of each layer. This not only strengthens the propagation of characteristics within each layer of the network, but also greatly slows down the gradient disappearance problem of the deep network [13]. In the initial stage of the convolutional neural network, if there are L layers, there will be L connections, but in DenseNet, there will be [L(L+1)]/2 connections.…”

Bilateral FS-UNet: a logging image feature segmentation model based on depth learning

“…However, the above methods are not suitable in the field of logging images with complex features because of the disadvantages of not making full use of contextual information and insufficient processing of detailed features. The improved U-Net network based on full convolutional network (FCN) has good prospects for application in the field of small sample image segmentation [13], and since the sample size of each category in the logging image dataset used in this paper is small, it can complete the model training and achieve better image segmentation with less sample size [15], which is very suitable for the characteristics of the dataset used in this paper.…”