摘要
膝关节退行性病变(knee osteoarthritis,KOA)是一种由关节软骨纤维化等引发的进展性膝关节疾病.病情发展大致可分为轻度与重度两个阶段,若能对其进行有效追踪,则可根据严重程度及时采取相应的防控措施,这对于提高患者生存质量有着重要临床意义.本文将这一过程称之为KOA分级诊断.相比传统的KOA诊断方法(CT,MRI等),骨振信号(Vibroarthrographic signal,VAG)有着无创无害,成本低廉,可便捷使用等优点,是近年来临床中正在探索的一种全新的KOA检查手段.然而,目前关于VAG信号的理论研究尚不充分,临床可提供的指导信息十分有限.基于此,本文以VAG信号为主要数据源,同时融入患者的生理信息,开展关于KOA分级诊断的辅助诊断方法研究.首先,在卷积神经网络框架下,构建了用于分析VAG信号的网络模块VAG-CNN-Block;其次,在前馈神经网络框架下,构建了用于分析生理信息的网络模块PI-FNN-Block;进而,结合VAG-CNN-Block和PI-FNN-Block,采用注意力机制设计了一种深度集成网络模型MBE-Net,并据此提出了KOA的分级诊断方法,用以实现正常受试者,轻度,重度KOA患者的自动识别.最后,采用西安市某两所医院的临床数据对所提方法进行验证.数值实验表明所提方法的准确率,灵敏度,特异度分别可达87.5%,87.2%与93.6%.
Knee osteoarthritis(KOA)is a progressive knee disease caused by cartilage fibrosis.Its development can be divided into two stages:mild and severe.If it can be tracked effectively,the appropriate prevention measures can be taken in time,which is important for improving the quality of life of patients and reducing the disability rate in clinic.This process is called KOA grading diagnosis in our work.Compared with the traditional diagnostic methods(e.g.,CT,MRI,etc.),vibroarthrographic(VAG)signal is a new diagnostic measurement of KOA being explored in clinic in recent years,which is noninvasive,harmless,low cost and convenient-to-use.However,the pathological information contained in VAG signal is not been understood enough,causing the less guidance for clinical practice.Based on this,this paper focuses on studying the automatic graded diagnosis method of KOA based on VAG signals and patient′s physiological information.Firstly,the network module VAG-CNN-block for analyzing VAG signals is constructed under the framework of convolutional neural network.Secondly,the network module PI-FNN-block for analyzing physiological information is constructed under the framework of feedforward neural network.Furthermore,a deep ensemble network model MBE-Net is designed using attention mechanism by combining with VAG-CNN-block and PI-FNN-block.Then,the MBE-Net model is trained to realize automatic detection of normal subjects,mild KOA patients and severe KOA patients.Finally,the clinical data of two hospitals in Xi′an are used to verify the proposed method.The numerical experimental results show that the accuracy,sensitivity and specificity of the proposed method can reach 87.5%,87.2%and 93.6%,respectively.
作者
宋江玲
郑田田
张瑞
Song Jiangling;Zheng Tiantian;Zhang Rui(School of Mathematics,Northwest University,Xi'an 710127,China)
出处
《纯粹数学与应用数学》
2022年第3期309-321,共13页
Pure and Applied Mathematics
基金
国家自然科学基金(12071369,62006189)
陕西省自然科学基金(2021JQ-430)
陕西省重点研发计划(2019ZDLSF02-09-02,2017ZDXM-Y-095)
陕西省创新才推进划(2018TD-016)。
关键词
膝关节退行性病变
分级诊断
骨振信号
深度集成网络模型
knee osteoarthritis
graded diagnosis
vibroarthrographic(VAG)signal
deep ensemble network model
