A semantic segmentation algorithm supported by image processing and neural network

The accurate segmentation of the lesion area is of great significance to the actual medical treatment. However, the segmentation results of the current segmentation network are not accurate enough to provide guidance for actual medical treatment. To solve this problem, a improved U-Net segmentation...

Full description

Saved in:

Bibliographic Details
Published in:	2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT) pp. 937 - 941
Main Authors:	Liu, Liwei, Qu, Daming, Hou, Alin
Format:	Conference Proceeding
Language:	English
Published:	IEEE 22-11-2021
Subjects:	attention mechanism deep learing Feature extraction image processing Image segmentation Lesions Liver Medical treatment neural network Neural networks residual optimization module Semantics
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The accurate segmentation of the lesion area is of great significance to the actual medical treatment. However, the segmentation results of the current segmentation network are not accurate enough to provide guidance for actual medical treatment. To solve this problem, a improved U-Net segmentation network is proposed. Firstly. The residual module and new attention mechanism are introduced to optimize the encoder, and 2×2 convolution is used instead of pooling operation, which can refine and extract features while retaining spatial feature information. Secondly, the attention mechanism is introduced before the upsampling jump connection, so that the network pays attention to the spatial information of the low-level feature map. The improved U-Net segmentation network was evaluated on the LiTS datasets. Compared with the traditional If-Net, the Dice coefficient and recall rate are increased by 5.6% and 3.03 % respectively in the liver segmentation task, the Dice coefficient and recall rate are increased by 7.51% and 8.8% respectively in the liver tumor segmentation task.
DOI:	10.1109/ICESIT53460.2021.9696835