摘要
目的将基于深度学习卷积神经网络的改进目标检测算法CF-SSD应用于肝囊型包虫病超声图像处理中,以实现肝包虫病病灶的自动检测,提高临床诊断效率。方法首先,用ResNet-50替换SSD骨干网络VGG-16、增加输出特征层的预选框数量;其次,对ResNet-50的前三层进行特征融合,将低层的细节特征与高层的语义特征进行结合;最后,在残差网络模块中增加坐标注意力机制模块,增强高层特征图语义信息。结果在肝包虫超声图像上进行实验,与SSD原算法比较,CFSSD算法检测的平均精度均值与速度分别为85.64%与50.34 Fps,提升了1.43%与22.90 Fps。结论结果表明,与SSD原算法对比,CF-SSD模型可实现更高效的检测,并确定肝囊型包虫病超声图像中的病灶位置和类别,满足临床要求,可为后期研究提供新思路。
Objective To realize the automatic detection of liver hydatid lesions and improve the efficiency of clinical diagnosis,the object detection of CF-SSD algorithm based on deep learning convolution neural network is applied to the ultrasonic image processing of liver cystic echinococcosis.Methods Firstly,the SSD backbone network VGG-16 was replaced by ResNet-50,and the number of preselected frames of the output feature layer was increased;secondly,the first three layers of ResNet-50 were fused to combine the low-level detail features with the high-level semantic features;finally,the attention mechanism module coordinate attention was added to the residual network module to enhance the semantic information of the high-level feature map.Results Compared with the original SSD algorithm,the mean average precision and speed of CF-SSD algorithm were 85.64%and 50.34 Fps,respectively,which increased by 1.43%and 22.90 Fps.Conclusion The experimental results show that compared with the original SSD algorithm,the CF-SSD model is more efficient in detecting and classifying the location and category of lesions in the ultrasonic image of hepatic cystic echinococcosis and can meet the requirements of clinical instantaneity,which provides a new idea for later research.
作者
米吾尔依提·海拉提
热娜古丽·艾合麦提尼亚孜
王正业
严传波
MIWUERYITI·Hailati;RENAGULI·Aihemaitiniyazi;WANG Zhengye;YAN Chuanbo(College of Public Health,Xinjiang Medical University,Urumqi Xinjiang 830011,China;College of Medical Engineering Technology,Xinjiang Medical University,Urumqi Xinjiang 830011,China)
出处
《中国医疗设备》
2023年第3期66-71,共6页
China Medical Devices
基金
国家自然科学基金(地区项目)(81560294)
省部共建中亚高发病成因与防治国家重点实验室项目(SKL-HIDCA-2020-YG)
大学生创新创业训练计划(202110760006)。
关键词
肝囊型包虫病
深度学习
目标检测
SSD算法
注意力机制
特征融合
liver cystic echinococcosis
deep learning
automatic detection
SSD algorithm
attention mechanism
feature fusion
