基于改进U-Net和X线片的脊柱侧弯Cobb角自动测量算法研究  

Research on the Automatic Measurement Algorithm of Scoliosis Cobb Angle Based on Improved U-Net and X-Rays

作　　者：禤浚波[1,2,3] 梁英豪 梁淑慧 张绿云 胡巍[5,6] 柯宝毅 马文宇 李成 XUAN Junbo;LIANG Yinghao;LIANG Shuhui;ZHANG Luiyun;HU Wei;KE Baoyi;MA Wenyu;LI Cheng(Guangxi Key Lab of Multi-source Information Mining&Security,Guangxi Normal University,Guilin,541004;School of Artificial Intelligence,Naning College for Vocational Technology,Nanning,530008;College of Computer Science and Engineering,Guangxi Normal University,Guilin,541004;School of Big Data and Computer Science,Hechi University,Yizhou,546300;Department of Spine and Osteopathy Surgery,Guilin People's Hospital,Guilin,541002;Department of Spine and Surgery,Liuzhou People's Hospital,Liuzhou,545006)

机构地区：[1]广西师范大学广西多源信息挖掘与安全重点实验室,桂林541004 [2]南宁职业技术学院人工智能学院,南宁530008 [3]广西师范大学计算机科学与工程学院,桂林541004 [4]河池学院大数据与计算机学院,宜州546300 [5]桂林市人民医院脊柱骨病科,桂林541002 [6]柳州市人民医院脊柱外科,柳州545006

出　　处：《基因组学与应用生物学》2024年第4期708-718,共11页Genomics and Applied Biology

基　　金：广西多源信息挖掘与安全重点实验室开放基金项目(MIMS21-02);广西自然科学基金项目(2022GXNSFAA035625);广西高校中青年教师科研基础能力提升项目(2022KY1019);桂林市科学研究与技术开发计划项目(20210227-2);广西壮族自治区卫生健康委员会科研项目(Z20210639)共同资助。

摘　　要：脊柱侧弯是影响人类健康的疾病之一,Cobb角的准确计算是临床上确定脊柱侧弯分型和制定诊疗方案的关键。针对人工测量Cobb角存在耗时长、不够准确、效率低下等问题,本文设计了一种基于改进U-Net的脊柱侧弯Cobb角自动测量方法。由经验丰富的脊柱外科医生使用LabelMe工具对200例脊柱侧弯患者的X线片数据集进行标注。采用ResNet50作为主干网络改进基本的语义分割模型U-Net,并与另外2个语义分割模型DeeplabV3和PSPNet在脊柱侧弯X线片数据集上分别进行训练。实验结果表明,改进的U-Net模型的平均交并比(mean intersection over union,MIOU)值达到了94.72%,分别比PSPNet和DeeplabV3模型的MIOU值提升了5.36%和2.30%。最后,基于改进的U-Net模型设计了脊柱侧弯Cobb角的自动测量算法,并开发了可视化的自动测量软件。经过实际测试,发现在常规的电脑上输入一张患者的X线片,只需6.3 s即可自动计算Cobb角大小,其速度远快于医生手动测量,显著提高了医生的工作效率,表明本文设计的脊柱侧弯Cobb角自动测量方法是有效的。Scoliosis is one of the diseases that affect human health.The accurate calculation of Cobb angle is the key to determine the classification of scoliosis and make the diagnosis and treatment plan.In order to solve the problems of time-consuming,inaccuracy and low efficiency in the manual measurement of Cobb angle,an automatic measurement method of scoliosis Cobb angle based on improved U-Net was designed in this paper.The X-ray data sets of 200 scoliosis patients were first annotated by an experienced spine surgeon using the LabelMe tool,and then the basic semantic segmentation model U-Net was improved using ResNet50 as the backbone network.The three semantic segmentation models that improved U-Net,DeeplabV3 and PSPNet,were designed and trained on the scoliosis X-ray data sets.The experimental results showed that the improved U-Net model achieved 94.72%in the mean intersection over union(MIOU)index,which was better than PSPNet and DeeplabV3 models by 5.36%and 2.30%respectively.Finally,an automatic measurement algorithm of the Cobb angle of scoliosis based on the improved U-Net model was designed,and a visual automatic measurement software was developed.In the actual test,when inputing a patient′s X-ray to a common computer,the software could automatically calculate the Cobb angle size only 6.3 s,which was much faster than doctors manual measurement,so this method can significantly improve the efficiency of the doctors.The test results showed that the designed automatic measurement method of the Cobb angle of scoliosis is effective.

关 键 词：脊柱侧弯 X线图像 COBB角 U-Net 交并比 

分 类 号：R687.3[医药卫生—骨科学] TP391.41[医药卫生—外科学]