骨质疏松性椎体压缩骨折MRI精准辅助诊断模型的研究  

An intelligent diagnosis model of osteoporotic vertebral compression fracture based on MRI scans

作　　者：严瀚 刘文锋 吴梦林 张广滔 廉宪坤 俞祝良[2] Yan Han;Liu Wenfeng;Wu Menglin;Zhang Guangtao;Lian Xiankun;Yu Zhuliang(The Second Hospital Affiliated to South China University of Technology(The First People's Hospital of Guangzhou),Guangzhou 510180,China;South China University of Technology,Guangzhou 510641,China)

机构地区：[1]华南理工大学附属第二医院(广州市第一人民医院),广州510180 [2]华南理工大学,广州510641

出　　处：《中华创伤骨科杂志》2023年第1期64-69,共6页Chinese Journal of Orthopaedic Trauma

摘　　要：目的拟开发一种基于MRI深度学习的自动精准检测骨质疏松性椎体压缩骨折(OVCF)的诊断模型。方法回顾性收集2019年1月至2021年10月广州市第一人民医院诊断为OVCF的500例患者资料。男396例,女204例;年龄(74.5±6.0)岁;骨密度T值为-2.9±0.8;骨折节段:L1128例,L_(2)113例,L_(3)109例,L_(4)115例,L_(5)108例。选择多模态分层融合网络进行训练、测试及验证,应用grad-cam可视化方法,构建基于脊柱MRI图像的深度学习模型。随机抽取30例诊断为OVCF患者的MRI图像,比较深度学习的精准辅助诊断模型与高年资脊柱外科医师对OVCF的诊断价值。结果建立的基于MRI图像深度学习的精准辅助诊断模型对OVCF的诊断准确度为96.7%,灵敏度为93.5%,特异度为88.9%,阳性预测值为100.0%,阴性预测值为76.9%,均高于2名高年资脊柱科医师(70.0%、72.7%、28.6%、82.1%、28.6%),差异均有统计学意义(P<0.05)。结论本研究成功建立了的基于MRI图像的深度学习的OVCF精准辅助诊断模型,其诊断效能高于脊柱外科医师。Objective To develop a deep learning model which can automatically and accurately detect osteoporotic vertebral compression fractures(OVCF)based on artificial intelligence.Methods MRI images of 500 patients diagnosed with OVCF at The First People's Hospital of Guangzhou from January 2019 to October 2021 were collected retrospectively.There were 396 males and 204 females,with an age of(74.5±6.0)years.The T value of bone mineral density was-2.9±0.8.The fracture segments were L1 in 128 cases,L_(2) in 113 cases,L_(3) in 109 cases,L_(4) in 115 cases,and L_(5) in 108 cases.The multimodal layered converged network was used to train,test,and verify the robustness and generalization ability of a deep learning model based on MRI images of OVCF.The grad-cam was applied to visualize the results.The diagnostic value of the model for OVCF was assessed by comparing the diagnoses between the artificial intelligence model and 2 senior spinal surgeons on the MRI images of 30 OVCF patients randomized from the 500 ones.Results Of the precise auxiliary diagnosis model for OVCF based on MRI images,the diagnostic accuracy was 96.7%,the sensitivity 93.5%,the specificity 88.9%,the positive predictive value 100.0%,and the negative predictive value 86.6%,all significantly higher than those of the 2 senior spinal surgeons(70.0%,72.7%,28.6%,82.1%,and 28.6%)(P<0.05).Conclusion The present study has successfully established a deep learning model which can automatically and accurately diagnose OVCF based on MRI images,showing a high diagnostic efficiency than human spinal surgeons.

关 键 词：脊柱 骨质疏松 骨折 磁共振成像 人工智能 

分 类 号：R683[医药卫生—骨科学] R445.2[医药卫生—外科学] R580[医药卫生—临床医学]