基于太赫兹波的颅脑创伤检测机理与方法研究

Traumatic brain injury (TBI) is the most common global disease with high mortality and disability. The detections of degree and edge of TBI is a significant clinical demand. Currently, the bottlenecks of the physical imaging methods for TBI include low sensitivity, high missed diagnosis rate, indistinct edge of lesions, and weak generalization ability. It is urgent to develop a novel label-free method for precise detection of TBI. Terahertz (THz) wave has the unique physical characteristics of water sensitivity, fingerprint spectrum, and safety in biomedical detection. In this project, we plan to obtain the specific frequency and content for TBI detection, and model the interaction between THz wave and TBI tissues by crossing modeling and nonlinear programming model. Both of them are analyzed to make the detection mechanism clear. An in vivo THz imaging system is designed and the reconstruction method for THz image based on morphology reconstruction is proposed to recognize the area and edge of TBI. A feature extraction method of biological THz images based on transmittance distribution features in spatial domain and statistical distribution features in normalized gray histogram is proposed to realize the automatic recognition of the TBI degree with machine learning. This research will accelerate the development of new detection methods for precise neurosurgery.

颅脑创伤是当前致死致残率最高的常见全球性疾病之一，颅脑创伤程度和边界的检测是神经外科领域的重大临床需求。当前，颅脑创伤成像方法存在灵敏度低、漏诊率高、病灶边界识别不准、病灶自动诊断识别方法泛化能力弱等问题，亟需发展一种新型无标记精确检测方法。太赫兹（THz）波在生物医学检测中具有水敏感性、指纹谱特性和安全性等特点，本项目拟通过解析THz波实现颅脑创伤检测的特异性检测频段和特异性组分，采用交叉融合建模技术和非线性规划模型，建立THz波与颅脑创伤组织相互作用的理论模型，明确THz波实现颅脑创伤检测的机理；搭建颅脑创伤在体THz成像系统，提出基于形态学重构的THz图像重建方法实现颅脑创伤灶区域及边界的精确识别；提出基于空间域强度分布特征和灰度直方图统计学分布特征联合的THz生物图像特征提取方法，结合机器学习技术实现颅脑创伤程度的高准确率自动识别。本项目的研究将促进精准神经外科检测新方法的发展。

项目以精准神经外科领域颅脑创伤程度和边界精确识别的重大临床需求为典型牵引，率先将太赫兹技术引入颅脑创伤检测的研究领域，建立了基于太赫兹波的精准神经外科检测新方法，并开展了初步的在体动物实验。主要成果包括（1）验证了太赫兹实现颅脑创伤检测的特异性主要由细胞密度和组织水含量决定，确定了特异性检测频段主要分布于1.2-2THz，阐明了太赫兹波实现颅脑创伤检测的机理，建立了颅脑创伤组织及其组分的太赫兹特征波谱数据库，波谱范围为0.1-2.0THz，分辨率为40MHz；（2）建立了太赫兹波与颅脑创伤组织相互作用的非线性规划模型，实现了生物组织在太赫兹波段介电参数和吸收系数的精准预测，误差小于6.6cm^-1；（3）研发了高灵敏、高信噪比的垂直反射式太赫兹生物成像系统，实现颅脑创伤的在体太赫兹成像，确定了成像的最佳临床检测条件，建立了颅脑创伤在体太赫兹图像数据库，建立了基于图像形态学重构的颅脑创伤太赫兹图像创伤灶区域及其边界高精度识别算法，分割误差小于0.5mm；（4）提出了基于空间域强度分布特征和灰度直方图统计学分布特征联合的太赫兹生物物理学特征提取方法，基于机器学习技术实现了颅脑创伤高准确率定级识别（假性、轻度、中度和重度），识别准确率达到91.25%。