AlexNet模型分级诊断膝关节软骨损伤  

AlexNet architecture for grading diagnosis of knee cartilage injury

作　　者：车至玮 郭冬梅[1] 张丽榕 韩云鹏[3] 任新萍 CHE Zhiwei;GUO Dongmei;ZHANG Lirong;HAN Yunpeng;REN Xinping(Department of Radiology,the Second Hospital of Dalian Medical University,Dalian 116027,China;School of Digital Art and Design,Dalian Neusoft University of Information,Dalian 116023,China;Department of Radiology,Dalian NO.3 People's Hospital,Dalian 116091,China)

机构地区：[1]大连医科大学附属第二医院放射科,辽宁大连116027 [2]大连东软信息学院数字艺术与设计学院,辽宁大连116023 [3]大连市第三人民医院放射科,辽宁大连116091

出　　处：《中国介入影像与治疗学》2023年第4期218-222,共5页Chinese Journal of Interventional Imaging and Therapy

摘　　要：目的基于3.0T MR常规序列图像及T2 mapping构建AlexNet模型,观察其用于分级诊断膝关节软骨损伤的价值。方法针对131例膝关节软骨损伤患者及25名无膝关节病变体检者共选取2500幅膝关节MRI,包括常规序列图像及T2 mapping,建立AlexNet模型;对比观察人工阅片、SqueezeNet模型及AlexNet模型用于分级诊断膝关节软骨损伤的效能。结果AlexNet模型分级诊断膝关节软骨损伤的整体效能高于人工阅片及SqueezeNet模型(P均<0.05);其分级诊断膝关节软骨损伤的准确率为97.50%,诊断Ⅰ级损伤精确度、召回率及F1-score分别为99.66%、98.67%及99.16%,Ⅱ级损伤为94.70%、95.33%及95.02%,Ⅲ级损伤为95.61%、94.33%及94.97%,Ⅳ级损伤为98.04%、100%及99.01%。结论AlexNet模型用于分级诊断膝关节软骨损伤效能较佳。Objective To establish an AlexNet architecture based on 3.0T MR conventional sequence images and T2 mapping and observe its value for grading diagnosis of knee cartilage injury.Methods A total of 2500 knee MR images acquired from 131 knee cartilage injury patients and 25 subjects who underwent health examination without knee joint disease were selected,including conventional sequence images and T2 mapping,and an AlexNet architecture was established.The values of manual interpretion,of SqueezeNet architecture and AlexNet architecture for grading diagnosis of knee cartilage injury were comparatively observed.Results The overall efficacy of AlexNet architecture for grading diagnosis of knee cartilage injury were higher than that of manual interpreting and SqueezeNet architecture(all P<0.05).The accuracy of AlexNet architecture for grading diagnosis of knee cartilage injury was 97.50%,and its accuracy,recall rate and F1-score of gradeⅠinjury was 99.66%,98.67%and 99.16%,of gradeⅡinjury was 94.70%,95.33%and 95.02%,of gradeⅢinjury was 95.61%,94.33%and 94.97%,of gradeⅣinjury was 98.04%,100%and 99.01%,respectively.Conclusion AlexNet Architecture was effective for grading diagnosis of knee cartilage injury.

关 键 词：膝关节 软骨疾病 深度学习 磁共振成像 

分 类 号：R681.3[医药卫生—骨科学] R445.2[医药卫生—外科学]