摘要
目的:初步探究使用最新的人工智能(AI)检测方法检测乳腺X线病灶,包括肿块,乳腺内淋巴结和钙化,为进一步的乳腺钼靶X线AI智能系统应用提供初步验证。方法:使用深度学习目标检测Faster R-CNN算法,训练人工标注的1892例乳腺钼靶X线数据集,在400例测试数据集上验证AI病灶检测的性能。结果:AI智能检测出肿块526个(共689个),乳腺内淋巴结912个(共1098个),圆形钙化52个(共73个),环形钙化519个(共692个),粗糙钙化353个(544个),其敏感度分别为76.4%,83.1%,71.2%,75.0%,64.9%,假阳率分别为35.7%,38.6%,0.9%,0.6%,18.4%。结论:AI能较好地检测出乳腺钼靶X线影像中的肿块,淋巴结和钙化,为更深入的AI智能检测系统研究提供初步验证。
Objective:The current study aims to develop a detection model for lesions including masses,lymph nodes,and calcification in mammography with the latest artificial intelligence(AI)algorithms.Our findings provide preliminary verification for the application of deep learning in multi-category lesion detection in mammography.Methods:This study used a deep learning object detection algorithm called faster R-CNN to detect different lesions.The training dataset composed of 1892 mammography studies was manually labelled by experienced radiologists.A testing dataset of 400 mammography studies was used to evaluate the performance of AI.Results:In the test dataset,our AI algorithm detected 526 masses(689 in total),912 lymph nodes(1098 in total),52 round calcifications(73 in total),519 rim calcifications(692 in total),coarse calcification 353(544 in total),and sensitivities of each lesion category was 76.4%,83.1%,71.2%,75.0%and 64.9%,respectively.And false positive rates of each lesion were 35.7%,38.6%,0.9%,0.6%and 18.4%,respectively.Conclusion:Our AI lesion detection algorithm can detect masses,lymph nodes and calcifications mammography,and lays a research foundation for further multi-category AI detection system for mammography.
作者
李欣
梁森
黄正南
夏晨
张荣国
吴孝掌
赖原仲
LI Xin;LIANG Shen;HUANG Zheng-nan(Department of Imaging,Tangshan Gongren Hospital,Hebei 063000,China)
出处
《放射学实践》
北大核心
2018年第10期1029-1032,共4页
Radiologic Practice
关键词
人工智能
学习
乳腺肿瘤
放射摄影术
Artificial intelligence
Learning
Breast neoplasms
Radiography
