Ultrasound image segmentation algorithm for hepatic cystic echinococcosis based on improved DeepLabV3+
10.3969/j.issn.1005-202X.2024.06.007
- VernacularTitle:基于改进DeepLabV3+的囊型肝包虫病超声图像分割算法
- Author:
Miwueryiti HAILATI
1
;
Renaguli AIHEMAITINIYAZI
;
Li LI
;
Chuanbo YAN
Author Information
1. 新疆医科大学公共卫生学院,新疆乌鲁木齐 830011
- Keywords:
hepatic cystic echinococcosis;
deep learning;
DeepLabV3+;
MobileNetV2;
efficient channel attention
- From:
Chinese Journal of Medical Physics
2024;41(6):702-709
- CountryChina
- Language:Chinese
-
Abstract:
Objective To apply the improved DeepLabV3+based image semantic segmentation algorithm to the ultrasound image processing for hepatic cystic echinococcosis,thereby achieving automatic segmentation and detection of hepatic echinococcosis lesions,and improving clinical diagnostic efficiency.Methods DeepLabV3+based image semantic segmentation network was employed as the basic method,and the following improvements were made.To address the issues of high computational complexity,high memory consumption,difficulty in deploying on embedded platforms with limited computing power,and difficulty in fully utilizing multi-scale information when extracting image feature information,the original backbone network Xception of the model was replaced with MobileNetV2 for obtaining a lightweight model framework.Additionally,efficient channel attention was applied to underlying features for reducing computational complexity and improving the clarity of target boundaries;and finally,Dice Loss was introduced into the model to alleviate the problem of the model focusing more on the background area and ignoring the foreground area containing the target.Results Validation was conducted on 5 lesion types in the self-built VOC2007 dataset of hepatic cystic echinococcosis.Experimental results showed that the improved model achieved a mean intersection over union of 73.8 and a mean pixel accuracy of 83.5,indicating that the model can predict more precise semantic segmentation results and effectively optimize model complexity and segmentation accuracy.
