基于神经网络的乳腺全容积超声图像质控方法  

Neural Network-Based Quality Control Method for Breast Full Volume Ultrasound Images

作　　者：庄洋 王金宏 ZHUANG Yang;WANG Jin-hong(Guangdong Artificial Intelligence and Modern Ultrasound Engineering Technology Research Center,Guangdong Shantou 515063;Shantou Chaonan Minsheng Hospital,Guangdong Shantou 515000)

机构地区：[1]广东省人工智能与现代超声工程技术研究中心,广东汕头515063 [2]汕头潮南民生医院,广东汕头515000

出　　处：《中国医疗器械信息》2024年第19期56-58,共3页China Medical Device Information

摘　　要：乳腺容积超声成像作为一种非侵入性、无辐射的乳腺检查技术,在乳腺疾病的早期筛查与诊断中扮演了越来越重要的角色。然而,乳腺容积超声成像图像在采集和处理过程中易受到多种干扰因素的影响,如设备噪声、患者呼吸运动、组织界面反射等,导致图像中出现伪像,降低了图像质量。同时,乳头位置的准确识别对于乳腺疾病的精确定位和诊断也至关重要。文章提出了一种基于神经网络的乳腺全容积超声数据图像质控方法,利用YOLO目标检测网络,分别实现了干扰伪像的检测和乳头位置的精确检测。通过实验验证,该方法在提高乳腺容积超声成像图像质控效率和准确性方面表现出色。As a non-invasive,radiation-free breast examination technology,automated breast ultra sound(ABUS)has played an increasingly important role in the early screening and diagnosis of breast diseases.However,ABUS images are susceptible to a variety of interference factors during acquisition and processing,such as equipment noise,patient respiratory movement,tissue interface reflection,etc.,which lead to artifacts in the image and reduce the image quality.At the same time,accurate identification of the nipple position is also crucial for the precise positioning and diagnosis of breast diseases.This paper proposes a deep learning-based breast full-volume ultrasound data image quality control method.Using the YOLO target detection network,the detection of interference artifacts and the precise detection of the nipple position are realized respectively.Through experimental verifications,this method performs well in improving the efficiency and accuracy of ABUS image quality control.

关 键 词：自动乳腺全容积超声成像 神经网络 伪像检测 乳头检测 YOLO 

分 类 号：R445.1[医药卫生—影像医学与核医学]