基于胶囊网络的青少年特发性脊柱侧弯智能分型研究  

Research on Intelligent Classification of Adolescent Idiopathic Scoliosis Based on Capsule Network

作　　者：段文玉 何越 杜钦红 杜钰堃 西永明[2] 杨环 DUAN Wen-yu;HE Yue;DU Qin-hong;DU Yu-Kun;XI Yong-Ming;YANG Huan(College of Computer Science and Technology,The Affiliated Hospital of Qingdao University,Qingdao University,Qingdao 266071,China;Department of Spine Surgery,Laoshan Hospital,The Affiliated Hospital of Qingdao University,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,青岛266071 [2]青岛大学附属医院崂山院区脊柱外科,青岛266071

出　　处：《青岛大学学报（自然科学版）》2023年第2期43-49,共7页Journal of Qingdao University(Natural Science Edition)

基　　金：山东省泰山学者项目(批准号:ts20190985)资助。

摘　　要：针对青少年特发性脊柱侧弯疾病的影像诊断问题,基于胶囊网络设计一种智能辅助医生诊断的脊柱侧弯分型算法,直接通过脊柱X光影像给出脊柱侧弯分型结果。该算法在胶囊网络使用共享注意力模块关注脊柱区域多尺度的特征信息,利用胶囊解码结构产生一致性的胶囊向量,使用动态路由算法共同激活脊柱畸变区域的胶囊向量完成脊柱侧弯分型的预测。实验结果表明,本算法的准确率、F1分数、G-mean等指标均优于Vgg、ResNet、EfficientNetV2等卷积神经网络分类方法。Aiming at the problem of imaging diagnosis of idiopathic scoliosis(AIS),an intelligent scoliosis classification algorithm based on capsule network was designed to assist doctors in the diagnosis of scoliosis.The scoliosis classification results are directly given from the X-ray images of the spine.A shared attention modale was used in capsule networks to focus on multiscale features of the spine region.Then,consistent capsule vectors were generated using the capsule decoding structure.Finally,a dynamic routing algorithm was used to activate the capsule vectors of the spinal lesion areas to predict scoliosis classification.The experimental results show that the algorithm is superior to the convolution neural network classification methods such as Vgg,ResNet,EfficientNetV2 in accuracy,F1-score and G-mean.

关 键 词：青少年特发性脊柱侧弯 胶囊网络 注意力模块 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]