颅脑硬膜血肿近红外光谱无创检测的最佳S-D分布研究

发布时间：2017-12-28 02:13

本文关键词：颅脑硬膜血肿近红外光谱无创检测的最佳S-D分布研究　出处：《天津工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于近红外光谱(Near Infrared Spectroscopy,NIRS)的脑血肿无创检测是生物医学领域中脑科学研究的重要内容。自然灾害等复杂环境下颅脑创伤伤情快速诊断是降低颅脑创伤死亡率的关键,近红外光谱实现脑血肿快速诊断,是解决此问题的重点方向之一,目前已有一定研究进展,而有关近红外光谱检测颅脑硬膜血肿的最佳光源-检测器(Source-Detector,S-D)分布的研究还未见发表,合理准确的选择S-D分布可能对颅脑硬膜血肿检测结果的准确度的提升、检测模型复杂度降低、脑功能的基础研究等提供更加丰富关键的参考信息。本文从基于近红外光密度差异检测脑血肿的角度出发,以获得无创颅脑硬膜血肿检测中最佳S-D分布为研究目标,在近红外光检测颅脑硬膜血肿基础理论、近红外光在脑组织中传输模型、检测模型仿真算法的设计与验证实验等方面进行了深入的研究。主要内容包括:基于Monte Carlo(MC)模型记录2.5cm、3.0cm、3.5cm、4.0cm、4.5cm位置在不同血肿出现情况下光密度变化,利用偏最小二乘法(Partial Least Square method,PLS)建立基于近红外光密度差异的颅脑硬膜血肿检测模型,在检测颅脑血肿的同时,可对颅脑硬膜血肿程度进行预测;提出近红外光谱血肿检测的等效信噪比概念,以此为最佳S-D分布的指导标准,进行最佳S-D分布仿真研究;引入808nm近红外光作为参考光,提供先验信息;利用变量投影重要性分析(Variable Importance in the Projection,VIP)对检测器个数及分布进行了优化,降低模型复杂程度;配制光学仿体对最佳S-D分布结果进行实验验证。本文所建立颅脑硬膜血肿程度预测模型相关度为88.94%,平均误差为0.0838cm-1,对样本数为16的预测集进行预测,预测集相关度为86.65%,平均误差为0.1047cm-1;根据等效信噪比和检测设备的信噪比,获得检测目标深度信息的最佳S-D范围;利用808nm近红外光作为参考光,建立针对不同头皮颅骨厚度最佳S-D位置的预测模型,训练集模型相关度为99.1%,平均误差为0.0291cm,对样本数为16的预测集进行预测,预测集预测相关度为98.8%,平均误差为0.0323cm;利用VIP分析从2.0-5.0cm范围内30个检测器中筛选出最佳的4个检测器(2.1cm、2.4cm、3.4cm、4.2cm),利用筛选出检测器对血肿程度进行预测,其精度优于30个检测器同时建模方案以及平均选择检测器建模方案;配置4种浓度的光学仿体进行实验,验证了本研究有关最佳S-D分布的结果,同时表明,对于不同头皮颅骨厚度的颅脑硬膜血肿患者,单组S-D检测时会有较大误差,验证了本研究所提出多通道检测方案的必要性。
[Abstract]:Noninvasive detection of hematoma based on Near Infrared Spectroscopy (NIRS) is an important part of the research of brain science in the field of biomedicine. Natural disasters and complex environment for rapid diagnosis of craniocerebral injury is the key to reduce the mortality of craniocerebral trauma, rapid diagnosis of cerebral hematoma achieve near infrared spectroscopy is one of the most important directions to solve this problem, there are some research progress, and the optimum source of near infrared spectroscopy to detect intracranial dural hematoma (Source-Detector, S-D) distribution detector the study has not been published, may choose S-D distribution with reasonable accuracy in craniocerebral epidural hematoma the accuracy of detection results, enhance the detection model of reduced complexity, brain function and basic research to provide more abundant key reference information. This article from the near infrared light density difference detection based on the angle of the hematoma, for noninvasive detection of brain hematoma best S-D distribution as the research object, the investigation in the near infrared detection head theory, epidural hematoma in the brain tissue near infrared light transmission model, detection model simulation algorithm design and experimental verification etc.. The main contents include: Based on Monte Carlo (MC) model, 3.0cm, 3.5cm, 2.5cm 4.0cm, 4.5cm position in different changes in optical density of hematoma cases, using partial least squares (Partial Least Square method, PLS) the establishment of near infrared light density difference of cerebral hematoma detection based on the model in the detection of intracranial hematoma at the same time that can predict the degree of brain hematoma; the equivalent SNR concept of near infrared spectrum detection based on optimal hematoma, S-D distribution guidance standard, simulation study on the distribution of S-D; the introduction of 808nm near infrared light as reference light, provide a priori information; variable importance in projection using Importance (Variable in the Projection, VIP) on the number and distribution of detector was optimized to reduce model complexity; preparation of optical phantom of the optimal S-D distribution of the results of an experiment. This brain epidural hematoma degree prediction model of correlation degree is 88.94%, the average error is 0.0838cm-1, the number of samples for prediction set of 16 forecast, prediction correlation degree is 86.65%, the average error is 0.1047cm-1; according to the equivalent SNR and detection equipment SNR, detection target depth information of the optimal range of S-D 808nm; using near infrared light as reference light, for different models of scalp skull thickness best S-D position, the training set model is 99.1%, the average error is 0.0291cm, the number of samples in the prediction set of 16 forecast, set predicted degree is 98.8%, the average error is 0.0323cm; using VIP analysis from the range of 2.0-5.0cm 30 detector selected 4 optimal detectors (2.1cm, 2.4cm, 3.4cm, 4.2cm), were selected by detector to predict hematoma degree, its precision is better than 30 detector At the same time modeling scheme and average detector selection modeling scheme; optical phantoms and 4 concentration experiments were conducted to verify the research about the optimal distribution of S-D results also show that for different thickness of the scalp skull brain hematoma patients, single group S-D test will have a greater error, verify the necessity of the proposed multi channel detection scheme.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R651.15;O434.33

【参考文献】