基于磁共振氢谱和弥散成像的诊断模型在颅内常见肿瘤的应用
发布时间：2018-08-07 10:41
【摘要】：目的： 在临床工作中对于胶质瘤、脑膜瘤和转移瘤等颅内常见脑肿瘤的鉴别诊断常常遇到困难，而不同肿瘤的临床治疗方法和预后又有很大差异。本项目拟开发一种颅内常见肿瘤的智能诊断软件，结合磁共振技术和人工智能技术两方面的优势，使临床常见颅内肿瘤的诊断正确率得到提高，使诊断程序更加简捷。 方法： 选自山东省医学影像学研究所2012年11月-2013年11月期间脑肿瘤病人60例，术前均行磁共振常规检查和磁共振波谱检查，包括胶质瘤25例(低级别胶质瘤10例，高级别胶质瘤15例)、脑膜瘤20例和转移瘤15例，部分转移瘤经临床证实，其余病例均经手术病理证实，另选20例正常志愿者。 在术前采用西门子公司SKYRA3.0T超导磁共振行MR常规检查(轴位TSE序列T2WI、T1WI和FLAIR)、1H-MRS、DWI检查，分别在肿瘤实质区、瘤周水肿区和正常对照区选取感兴趣区(ROI)，测定ROI区域的代谢物比值和ADC值，记录各ROI的NAA/Cr、Cho/Cr、NAA/Cho比值及ADC值。统计学分析使用SPSS13.0，计算三种肿瘤感兴趣区各代谢物比值及ADC值的平均值，用均数±标准差表示；采用双样本t检验，对比三种肿瘤实质区之间、三种肿瘤水肿区之间、不同级别胶质瘤之间各代谢物比值及ADC值有无差异，P值小于0.05为差异具有统计学意义。 对波谱数据集行遗传算法分析，选择与优化特征值，将提取出的20个显著特征值作为最优特征子集输入分类器；将代谢物比值和ADC值组成的典型特征值，直接作为特征值输入分类器。用于病人样本分类的分类器采用Fisher判别法和支持向量机(SVM)两种。然后根据每种单分类器的权重和四个结果的差异对分类结果进行评估，最后确定诊断结果。在实际医学诊断过程中，将新病例输入最优化的多分类器组进行分类，将分类结果作为人工智能诊断结果。 结果： 1.三种颅内肿瘤实质区之间NAA/Cr值差异有统计学意义(p0.05)，脑膜瘤实质区与胶质瘤、转移瘤实质区之间Cho/Cr值差异有统计学意义(p0.05)，脑膜瘤实质区与胶质瘤、转移瘤实质区之间的NAA/Cho值差异有统计学意义(p0.01)，脑膜瘤实质区与胶质瘤、转移瘤实质区之间ADC值差异有统计学意义(p0.01)，胶质瘤瘤周水肿区与脑膜瘤、转移瘤瘤周水肿区之间ADC值差异有统计学意义(p0.05)，见表3；高、低级别胶质瘤瘤周水肿区之间Cho/Cr、 NAA/Cho及ADC值差异有统计学意义(p0.05)，见表4；高级别胶质瘤与转移瘤之间瘤周水肿区NAA/Cho、Cho/Cr、NAA/Cr及ADC值均有统计学差异(p0.01)，见表5。 2.1H-MRS波谱数据经遗传算法(Genetic Algorithms，GA)特征提取，得到20个特征值，得到的经典特征值包括弥散加权成像后处理得到ADC值，波谱后处理测得的NAA、Cho、Cr、Lac、Lip浓度值以及NAA/Cr、Cho/Cr、NAA/Cho等浓度的比值。经交叉验证实验后，将提取的特征及经典特征值进入Fisher分类器、SVM分类器，得到分类结果。评估计算机诊断模型的诊断正确率，见表6。 结论： 1．在肿瘤实质区，脑膜瘤与其他两种肿瘤的代谢物比值及ADC值存在明显差别；高、低级别胶质瘤瘤周水肿区的代谢物比值及ADC值差异明显；瘤周的NAA/Cho及ADC值可用于鉴别高级别胶质瘤与转移瘤。利用肿瘤和瘤周水肿区的1H-MRS代谢物比值及ADC值，可以较好的对三种颅内肿瘤作出鉴别诊断。 2．以1H-MRS和DWI为基础建立的人工智能诊断模型通过数据挖掘，，特征向量提取，将不同肿瘤的代谢和弥散特征分辨出来，诊断正确率高，达到鉴别诊断的目的。在一定程度上起到了非侵入性活检的效果,对临床治疗方案的制定具有更强的指导意义，具有良好的临床应用价值。
[Abstract]:Objective:
In clinical work, the differential diagnosis of intracranial common brain tumors, such as glioma, meningioma and metastatic tumor, often encountered difficulties, and the clinical treatment methods and prognosis of different tumors are very different. This project intends to develop an intelligent diagnosis software for intracranial common tumors, combined with the two aspects of magnetic resonance technology and artificial intelligence technology. The diagnostic accuracy of clinical common intracranial tumors has been improved, making the diagnostic procedure simpler.
Method:
60 patients with brain tumors were selected from the Shandong medical imaging Institute in November 2012 -2013 November. Preoperative magnetic resonance imaging and magnetic resonance spectroscopy were performed, including 25 gliomas (10 cases of low grade gliomas, 15 high grade gliomas), 20 meningiomas and 15 metastatic tumors. Some metastatic tumors were clinically confirmed, and the rest of the cases were all confirmed. All the remaining cases were all confirmed cases were all cases were all cases were all confirmed cases were all cases were all cases were all confirmed cases were all cases all cases were confirmed by clinical After the operation and pathology, 20 normal volunteers were selected.
The SIEMENS SKYRA3.0T superconducting magnetic resonance (MR) routine examination (axial TSE sequence T2WI, T1WI and FLAIR), 1H-MRS, DWI examination were used before the operation to select the region of interest (ROI) in the tumor parenchyma, the peritumoral edema area and the normal control area, respectively, to determine the metabolite ratio and ADC value of the ROI region. Statistical analysis was used to calculate the average value of the ratio of metabolites and ADC values in three regions of interest with SPSS13.0, and the mean values of average number of ADC were calculated, and the ratio of metabolites and ADC of different grade gliomas were compared between the three tumor edema areas by double sample t test, and the value of P was less than 0.05. The difference was statistically significant.
By analyzing the genetic algorithm of spectral data set, selecting and optimizing the eigenvalues, 20 significant eigenvalues are extracted as the optimal subset to input the classifier, and the typical eigenvalues composed of the ratio of metabolites and ADC values are used directly as the eigenvalue input classifier. The classifier used in the classification of patient samples adopts the Fisher discriminant method and support. Two kinds of vector machines (SVM). Then the classification results are evaluated according to the weight of each single classifier and the difference between four results. Finally, the diagnosis results are determined. In the actual medical diagnosis process, the new multiple classifier group is classified into the optimized multiple classifier group, and the classification results are made as artificial intelligent diagnosis results.
Result:
1. the difference of NAA/Cr value between the three kinds of intracranial tumor parenchyma was statistically significant (P0.05). The difference of Cho/Cr between the parenchymal area of meningioma and glioma and the parenchymal area of metastatic tumor was statistically significant (P0.05). The difference of NAA/Cho between the parenchymal area of meningioma and glioma and the parenchymal area of the metastatic tumor was statistically significant (P0.01), and the parenchyma area of meningioma and glioma were and glia. The difference of ADC value between tumor and metastatic tumor was statistically significant (P0.01). The difference of ADC between the edema area of glioma and meningioma and the edema area of metastatic tumor was statistically significant (P0.05), see Table 3; high, Cho/Cr, NAA /Cho and ADC between the edema regions of low grade glioma, and NAA /Cho and ADC value (P0.05), see Table 4; high grade The values of NAA/Cho, Cho/Cr, NAA/Cr and ADC in the peritumoral edema area between glioma and metastatic tumor were statistically different (P0.01), see table 5..
The 2.1H-MRS spectral data are extracted by genetic algorithm (Genetic Algorithms, GA), and 20 eigenvalues are obtained. The classical eigenvalues obtained are ADC value after diffusion weighted imaging, NAA, Cho, Cr, Lac, Lip concentration measured after spectral processing, and the ratio of concentration to NAA/Cr, Cho/Cr, and etc. after cross validation. After cross validation experiments, the extract will be extracted. Classical eigenvalues are incorporated into Fisher classifier and SVM classifier to obtain classification results. The diagnostic accuracy of the computer diagnostic model is evaluated in Table 6.
Conclusion:
1. in the tumor parenchyma, the metabolite ratio and ADC value of the meningioma and other two kinds of tumor were significantly different. The metabolite ratio and the ADC value of the high, low grade glioma edema area were significantly different. The NAA/Cho and ADC values of the tumor peritumor could be used to identify the high grade glioma and metastatic tumor. The 1H-MRS metabolites of the tumor and peritumoral edema area were used. The ratio and ADC value can be used for differential diagnosis of three kinds of intracranial tumors.
2. the artificial intelligent diagnostic model based on 1H-MRS and DWI is established by data mining and feature vector extraction to distinguish the metabolic and dispersion characteristics of different tumors. The diagnostic accuracy is high and the purpose of differential diagnosis is achieved. To a certain extent, the results of non invasive biopsies have been achieved, and the formulation of clinical treatment schemes is stronger. It is of guiding significance and has good clinical application value.
【学位授予单位】：泰山医学院
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R739.41;R445.2
本文编号：2169770
[Abstract]:Objective:
In clinical work, the differential diagnosis of intracranial common brain tumors, such as glioma, meningioma and metastatic tumor, often encountered difficulties, and the clinical treatment methods and prognosis of different tumors are very different. This project intends to develop an intelligent diagnosis software for intracranial common tumors, combined with the two aspects of magnetic resonance technology and artificial intelligence technology. The diagnostic accuracy of clinical common intracranial tumors has been improved, making the diagnostic procedure simpler.
Method:
60 patients with brain tumors were selected from the Shandong medical imaging Institute in November 2012 -2013 November. Preoperative magnetic resonance imaging and magnetic resonance spectroscopy were performed, including 25 gliomas (10 cases of low grade gliomas, 15 high grade gliomas), 20 meningiomas and 15 metastatic tumors. Some metastatic tumors were clinically confirmed, and the rest of the cases were all confirmed. All the remaining cases were all confirmed cases were all cases were all cases were all confirmed cases were all cases were all cases were all confirmed cases were all cases all cases were confirmed by clinical After the operation and pathology, 20 normal volunteers were selected.
The SIEMENS SKYRA3.0T superconducting magnetic resonance (MR) routine examination (axial TSE sequence T2WI, T1WI and FLAIR), 1H-MRS, DWI examination were used before the operation to select the region of interest (ROI) in the tumor parenchyma, the peritumoral edema area and the normal control area, respectively, to determine the metabolite ratio and ADC value of the ROI region. Statistical analysis was used to calculate the average value of the ratio of metabolites and ADC values in three regions of interest with SPSS13.0, and the mean values of average number of ADC were calculated, and the ratio of metabolites and ADC of different grade gliomas were compared between the three tumor edema areas by double sample t test, and the value of P was less than 0.05. The difference was statistically significant.
By analyzing the genetic algorithm of spectral data set, selecting and optimizing the eigenvalues, 20 significant eigenvalues are extracted as the optimal subset to input the classifier, and the typical eigenvalues composed of the ratio of metabolites and ADC values are used directly as the eigenvalue input classifier. The classifier used in the classification of patient samples adopts the Fisher discriminant method and support. Two kinds of vector machines (SVM). Then the classification results are evaluated according to the weight of each single classifier and the difference between four results. Finally, the diagnosis results are determined. In the actual medical diagnosis process, the new multiple classifier group is classified into the optimized multiple classifier group, and the classification results are made as artificial intelligent diagnosis results.
Result:
1. the difference of NAA/Cr value between the three kinds of intracranial tumor parenchyma was statistically significant (P0.05). The difference of Cho/Cr between the parenchymal area of meningioma and glioma and the parenchymal area of metastatic tumor was statistically significant (P0.05). The difference of NAA/Cho between the parenchymal area of meningioma and glioma and the parenchymal area of the metastatic tumor was statistically significant (P0.01), and the parenchyma area of meningioma and glioma were and glia. The difference of ADC value between tumor and metastatic tumor was statistically significant (P0.01). The difference of ADC between the edema area of glioma and meningioma and the edema area of metastatic tumor was statistically significant (P0.05), see Table 3; high, Cho/Cr, NAA /Cho and ADC between the edema regions of low grade glioma, and NAA /Cho and ADC value (P0.05), see Table 4; high grade The values of NAA/Cho, Cho/Cr, NAA/Cr and ADC in the peritumoral edema area between glioma and metastatic tumor were statistically different (P0.01), see table 5..
The 2.1H-MRS spectral data are extracted by genetic algorithm (Genetic Algorithms, GA), and 20 eigenvalues are obtained. The classical eigenvalues obtained are ADC value after diffusion weighted imaging, NAA, Cho, Cr, Lac, Lip concentration measured after spectral processing, and the ratio of concentration to NAA/Cr, Cho/Cr, and etc. after cross validation. After cross validation experiments, the extract will be extracted. Classical eigenvalues are incorporated into Fisher classifier and SVM classifier to obtain classification results. The diagnostic accuracy of the computer diagnostic model is evaluated in Table 6.
Conclusion:
1. in the tumor parenchyma, the metabolite ratio and ADC value of the meningioma and other two kinds of tumor were significantly different. The metabolite ratio and the ADC value of the high, low grade glioma edema area were significantly different. The NAA/Cho and ADC values of the tumor peritumor could be used to identify the high grade glioma and metastatic tumor. The 1H-MRS metabolites of the tumor and peritumoral edema area were used. The ratio and ADC value can be used for differential diagnosis of three kinds of intracranial tumors.
2. the artificial intelligent diagnostic model based on 1H-MRS and DWI is established by data mining and feature vector extraction to distinguish the metabolic and dispersion characteristics of different tumors. The diagnostic accuracy is high and the purpose of differential diagnosis is achieved. To a certain extent, the results of non invasive biopsies have been achieved, and the formulation of clinical treatment schemes is stronger. It is of guiding significance and has good clinical application value.
【学位授予单位】：泰山医学院
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R739.41;R445.2
【参考文献】
相关期刊论文 前1条
1 周宏伟;孔博玉;兰文婧;刘洋;谷艳英;;氢质子磁共振波谱在颅内常见肿瘤诊断中的应用价值[J];吉林大学学报(医学版);2010年02期
本文编号：2169770
本文链接：https://www.wllwen.com/yixuelunwen/fangshe/2169770.html
