改进型YOLOv5s网络在胆囊超声图像检测中的应用  

Application of Improved YOLOv5s Network in Gallbladder Ultrasound Image Detection

作　　者：俞永伟 YU Yongwei(Department of General Surgery,Anhui Zhongke Gengjiu Hospital,Hefei Anhui 230092,China)

机构地区：[1]安徽中科庚玖医院普外科,安徽合肥230092

出　　处：《中国医疗设备》2023年第5期99-104,共6页China Medical Devices

摘　　要：目的解决临床过程中医生利用胆囊超声图像进行结石和息肉鉴别时费时费力的问题。方法选取超声科采集的200张临床胆囊超声图像作为初始数据集,通过引入BiFPN结构和EIOU损失函数对YOLOv5s模型进行改进。首先对初始数据集进行数据增强,然后将增强后的数据集送入改进后的YOLOv5s模型中进行训练。结果经过300次迭代,改进后的YOLOv5s模型在测试集的平均精度均值达到了89.79%,与同类型模型相比有明显提升。结论改进后的YOLOv5s模型有效克服了原模型对中小目标检测精度差的问题,且敏感度明显提升,有助于医生进行胆囊超声图像中结石和息肉的识别定位。Objective To solve the problem of time-consuming and labor in identification of stones and polyps by the use of ultrasound images of the gallbladder for doctors in the clinical process.Methods Using 200 images collected by of Department of Ultrasound as the initial data set,the initial data set was first enhanced,the YOLOv5s model was improved by BiFPN structure and EIOU loss function.Firstly,the initial data set was enhanced,and then the enhanced data set was sent to the improved YOLOv5s model for training.Results After 300 iterations,the average accuracy of the improved YOLOv5s model in the test set reached 89.79%,which was significantly improved compared with the same type of model.Conclusion The improved YOLOv5s model proposed in this paper effectively overcomes the problem of poor detection accuracy of the original model for small and medium targets,and significantly improves its sensitivity.It can help doctors identify and locate stones and polyps in gallbladder ultrasound images.

关 键 词：目标检测 胆囊结石 胆囊息肉 YOLOv5s 

分 类 号：R575.62[医药卫生—消化系统] TP391.41[医药卫生—内科学]