【作者】 刘凯；

【导师】 蔡洪斌；

【作者基本信息】 电子科技大学 ， 电子信息（专业学位）， 2024， 硕士

【摘要】 脊柱侧弯作为一种常见的脊柱畸形疾病,在青少年中发病率呈上升趋势。该疾病会严重影响青少年的身心健康与日常生活。而早发现早治疗会大大降低疾病对患者的影响,减轻患者的痛苦,并降低治疗费用。在脊柱侧弯疾病的防控工作中,对青少年进行大规模筛查和对患者的诊断治疗过程里都存在大量传统的人工检测方式,效率与可靠性都有待提升。针对脊柱侧弯的防控中筛选与诊断两个环节的问题,本文进行了深入研究,并根据两个环节不同的需求设计了两种不同的方法。在诊断环节,针对Cobb角测量依靠手工方式,依赖经验及个体评判差异大的问题,提出了一种基于脊柱X光影像自动测量Cobb角的算法。相比于其他自动测量算法,本算法将对有向物体检测的深度学习模型与医学影像结合,并融合了针对角度特征提取的模块,使得模型对于角度的测量更加精准。实验证明该模型能够在实验数据集上取得优异的水平,SMAPE达到了5.82%,平均绝对角度误差为3.03度。在筛查环节,相较于传统深度学习筛查方式主要依赖图像数据,本文中提出了一种直接使用点云对脊柱侧弯进行筛查的改进模型。详细介绍了整个模型的设计及点云空洞的修复过程。实验证明模型具备不亚于传统基于图像的深度学习模型的预测能力,模型对脊柱侧弯的分类准确率达到了75.39%。对基于点云对脊柱侧弯进行筛查提供了新的研究思路。最后基于设计的方法跟实际业务需求,设计实现了一个轻量级的脊柱侧弯诊断与筛查系统,为实际业务带来便利性。更多还原

【Abstract】 As a common spinal deformity,scoliosis is on the rise in adolescent disease.The disease can seriously affect the physical and mental health and daily life of adolescents.Early detection and early treatment will greatly reduce the impact of the disease on patients,alleviate patients’ suffering,and reduce treatment costs.In the prevention and control of scoliosis,there are a large number of traditional manual testing methods in the process of screening and diagnosis.The efficiency and reliability need to be improved.This thesis conducts research on screening and diagnosis and designs two different methods according to the different needs of the two stages.In the diagnosis process,in order to solve the problem that Cobb angle measurement relies on manual methods,experience and individual judgment,an algorithm based on spinal X-ray images for automatic measurement of Cobb angle was proposed.Compared to other automatic measurement algorithms,this algorithm use combination of the deep learning model for oriented object detection and medical imaging,and the integration of the module for angle feature extraction,makes the model more accurate for angle measurement.Experiments show that the model can achieve excellent level on the experimental dataset,which SMAPE reached 5.82%,with an average absolute angle error of 3.03 degrees.In the screening process,compared with the traditional deep learning screening method,which mainly relies on image data,this thesis proposes a improved model that directly uses point clouds to screen scoliosis.The design of the whole model and the repair process of point cloud cavities are introduced in detail.Experiments show that the model has the prediction ability of the traditional image-based deep learning model.The classification accuracy of the model for scoliosis reached 75.39%.It provides a new research idea for the screening of scoliosis based on point clouds.Finally,based on the design method and actual business needs,a lightweight scoliosis screening and diagnosis system is designed and implemented,which brings convenience to the actual business.更多还原