红外光谱成像结合化学计量学对关节软骨的分类识别

发布时间：2018-04-17 04:19

本文选题：关节软骨 + 骨关节炎　；参考：《南京航空航天大学》2017年硕士论文

【摘要】：关节软骨覆盖于骨表面,是骨关节的重要组成部分之一,主要作用是在关节活动中承受力学负荷,缓冲震动以及减少摩擦。关节软骨基质的主要成分为胶原蛋白和蛋白多糖。年龄、肥胖、外伤等因素会造成关节软骨的变性甚至损伤,进一步发展可能会导致骨关节炎的发生。由于在骨关节炎早期,关节软骨仅发生组分含量和结构的变化,并不出现形态学上的改变。这使得目前常用的临床诊断技术无法有效地识别早期骨关节炎。本文尝试采用FTIRI技术结合不同的化学计量学识别算法对正常和病变关节软骨进行分类研究,寻找最优的判别模型,期望为早期骨关节炎的准确诊断开辟新的途径。其中,傅里叶变换红光谱成像技术(FTIRI)可以同时获得被测样品的红外光谱信息及其形貌特征,具备丰富的组分种类和含量信息。化学计量学方法可以有效地提取光谱中的与相关化学组分对应的特征信息,常用于光谱的定量和定性分析。其在物质的定量分析和光谱分类识别等相关领域有着广泛的应用。本研究采集了关节软骨的正常样本、8周病变样本以及2年病变样本的光谱数据,利用主成分分析(PCA)算法、Fisher判别(FDA)算法、偏最小二乘判别(PLS-DA)算法以及支持向量机判别(SVM-DA)算法分别构建判别模型,实现对正常和病变光谱的分类识别。主要内容为:(1)基于光谱预处理方法,利用PLS-DA算法对正常和2年病变组光谱进行分类识别,预测准确率为96.92%。(2)利用PCA结合FDA算法分别对未经预处理的正常软骨光谱和8周病变光谱以及2年病变光谱进行分类识别。其中,正常vs 8周病变组的预测准确率为89.23%,正常vs 2年病变组的预测准确率为92.31%。(3)利用SVM-DA算法实现正常、8周病变和2年病变光谱的多类判别,整体预测准确率为90.33%。当利用SVM-DA实现正常vs 2年病变组光谱的二类判别时,其预测准确率为97.7%。比较3种模型的分类结果发现,上述3种模型均可以有效地实现对正常和病变光谱的分类识别。其中,SVM-DA算法具有最佳的分类效果且可以有效地实现关节软骨光谱的多类识别,有潜力发展成为一种新型的早期骨关节炎诊断方法,并为进一步的研究提供理论依据和数据支持。
[Abstract]:Articular cartilage, which covers the surface of bone, is one of the important components of bone joint. The main function of articular cartilage is to withstand load, cushion vibration and reduce friction in joint motion.The main components of articular cartilage matrix are collagen and proteoglycan.Age, obesity, trauma and other factors may cause degeneration or injury of articular cartilage, and further development may lead to osteoarthritis.In the early stage of osteoarthritis, the articular cartilage changes only in composition and structure, but not in morphology.This makes the current commonly used clinical diagnosis technology can not effectively identify early osteoarthritis.This paper attempts to study the classification of normal and diseased articular cartilage by using FTIRI technique and different chemometrics recognition algorithms to find the best discriminant model and to open up a new way for the accurate diagnosis of early osteoarthritis.Fourier transform red spectral imaging (FTIRI) can simultaneously obtain the infrared spectrum information and its morphological characteristics of the samples, and it has abundant information on the composition and content of the samples.The chemometrics method can effectively extract the characteristic information corresponding to the related chemical components in the spectrum, which is often used for quantitative and qualitative analysis of the spectrum.It is widely used in quantitative analysis of matter and spectral classification and recognition.In this study, we collected the spectral data of normal articular cartilage samples from 8 weeks and 2 years of pathological changes, and used principal component analysis (PCA) algorithm and Fisher discriminant FDA-algorithm.Partial least square discriminant (PLS-DA) algorithm and support vector machine discriminant (SVM-DA) algorithm are used to construct discriminant models respectively to realize the classification and recognition of normal and pathological spectrum.The main content is: (1) based on the spectral pretreatment method, PLS-DA algorithm is used to classify and recognize the spectrum of normal and 2-year lesion groups.The prediction accuracy is 96.92 / 2) the unpretreated normal cartilage spectrum, the 8-week lesion spectrum and the 2-year lesion spectrum are classified and identified by PCA and FDA algorithm, respectively.Among them, the prediction accuracy of normal vs 8-week lesion group was 89.23 and that of normal vs 2-year lesion group was 92.31 and 92.31.The SVM-DA algorithm was used to distinguish the spectrum of normal 8-week lesion and 2-year lesion, and the overall prediction accuracy was 90.33%.When SVM-DA was used to distinguish the spectrum of normal vs 2 year lesion group, the prediction accuracy was 97. 7%.By comparing the classification results of the three models, it is found that the above three models can effectively realize the classification and recognition of normal and pathological spectrum.SVM-DA algorithm has the best classification effect and can effectively realize multi-class recognition of articular cartilage spectrum. It has the potential to develop into a new diagnosis method of early osteoarthritis and provide theoretical basis and data support for further research.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN219;R68

【参考文献】