节点文献

生物颗粒流式成像关键技术的研究

Study of Key Techniques of Imaging Flow Cytometry for Bio-Particle Detection and Recongnition

【作者】 朱险峰

【导师】 王明时;

【作者基本信息】 天津大学 , 生物医学工程, 2010, 博士

【摘要】 目前临床使用的血细胞、尿沉渣等生物颗粒检测仪器的工作原理是基于库尔特或光散射原理,不能检测颗粒的形态。理想的生物颗粒测量技术应该能逐一、大量、快速地观测生物颗粒的形态,辅以其他参数,从而实现颗粒的识别、分类、计数和物理参数、生物化学成分的测量。论文构想的流式生物颗粒成像技术,将生物颗粒从微型喷嘴以鞘流形式喷出,使用高速数码相机对生物颗粒逐一、快速拍摄,辅以图像识别技术,对拍摄到的照片上的颗粒进行形态识别,从而实现颗粒分类和计数的目的。理论上,该技术具有准确度高、速度快、可追溯性、自学习性等优点。作者设计并搭建了流式成像实验装置。装置包括产生鞘流系统、鞘流速度测量与速度调控系统、颗粒到达检测与闪光灯和相机触发系统、以及显微成像系统。鞘流速度测量系统采用二次光散射方法,检测同一颗粒顺序通过两束相距已知的平行激光所产生的散射光的时间间隔,从而获得颗粒的飞行速度;鞘流速度的调控是用鞘流速度测量值反馈控制鞘液池内气压来实现的。实验表明,使用本方法鞘流速度可以稳定在5%以内;显微成像系统主要包括一只闪光灯、一套具有10mm观测距离、20μm景深的显微镜系统和一台快门时间高达16μs的CCD相机。运用该装置进行了血样、尿样生物样本的拍摄实验,成功地拍摄到血液白细胞、尿液结晶和管型等生物颗粒。颗粒图像的外缘、纹路清晰可辨。论文对白细胞图像识别技术进行了初步探讨。利用小波变换结合逐点噪声方差法对白细胞图像进行去噪处理,并在HSV颜色空间利用二元最大类间方差法(OSTU2)进行分割,采用了满意特征方法进行特征提取,使用工神经网络和支持向量机(SVM)对提取到的满意特征向量分类。对使用光学显微镜拍摄到的246幅嗜中性、单核、和淋巴白细胞的初步实验表明,中性粒细胞识别率可达84.5%,淋巴细胞正确识别率可达78%。论文总结了目前已有的全部液面探测技术,运用发明问题解决理论(TRIZ),分析了目前最流行的电容液面探测技术的技术成熟度,获得其已经处于技术成熟期下降阶段的结论,并预测了液面探测技术的发展方向。论文提出的基于光反射式的液面探测技术,使激光光源和位置传感器分别位于被测样本的左右斜上方,激光器发出的激光经样本液面反射,照射到位置传感器上,光斑在传感器上的位置与液面高度呈线性关系。实验证明,该方法的测量误差可以达0.27mm。论文提出的图像识别液面探测技术采用数码相机拍摄样本试管内壁,测量照片上液面与试管壁交界距试管口的距离,实现液面高度测量。实验证明,该方法测量液面高度呈良好线性,不受样本导电性影响、且能够探测出瓶盖和液面气泡的。论文提出了对慢速相机拍摄快速运动物体时的运动拍摄补偿技术。位于摄影区内并与物体运动方向成45°角的反射镜,探测装置探测运动物体进入摄影区的时刻并驱动反射镜随运动物体移动一定距离。CCD相机相向于反光镜运动方向拍摄反光镜内运动物体的像。模拟实验表明,使用该补偿技术可以提高10倍以上的拍摄速度。

【Abstract】 The present clinical bio-particle(such as blood cell and urine sediment) measurement techniques are not base on morphology of the particles, but the electric resistance or light scattering. The ideal technique or technology should observe the shape and pattern of numerous bio-particles one by one swiftly, with the help of other parameters, to recognize, classify, count the number, and measure the physical parameters and biochemical contents. This dissertation conceived an Imaging Flow Cytometry (I-FCM) technique. The diluted bio-sample with its bio-particles contained are sprayed out from a nozzle, then the bio-particles are photographed by a digital camera one by one. The photos are developed with image processing software, allowing the morphology of bio-particles to be studied.Theoretically, this technique has the advantages of high accuracy, high speed, traceability, and self-learning ability.For testing I-FCM, a prototype is designed and built. The flow velocity measuring system works on double light-scattering. When a particle passes through two light beams where the distance is known, two scattering light pulses are detected by a photodiode. The regulation of sheath flow velocity is approached by feedback from the flow velocity and adjusting the air pressure in the sheath liquid reservoir. Experiment shows, the sheath flow velocity can be stabilized within 5%. The photographic system includes a strobe light, a microscope which has 10mm observation distance and 20μm depth of focus and a fast CCD camera whose fastest shutter is 16μs. White blood cells (WBC), urine cast and crystal are captured successfully by this prototype. The shapes and patterns of the bio-particles were recognizable.The digital image process for WBC is studied. Wavelet transformation combined with the point wise variance method is applied to de-noise the WBC pictures, and OSTU2 in the HSV color space for separating images, a satisfactory feature selection method for extracting image features, support vector machine(SVM) for vector classification are applied. The primary experiment through 246 WBC images indicates that, the precision is 84.5% and 78% for neutrophile and lymphocyte respectively.This dissertation reviewed all the existing liquid level detection(LLD) techniques. The technical maturity of the most popular capacitive LLD is calculated with TRIZ; the conclusion is that the capacitive LLD is at its evolution stage of between maturity and decline. The evolution direction of LLD is made as well.A novel LLD technique based on light reflection is designed. A laser generator and a position sensitive detector (PSD) are placed at the upper-right and upper-left of a sample tube respectively. The laser, emitting from the laser generator, illuminates the surface of liquid sample, then reflects from the surface to the PSD. The position of the light spot on the PSD has a linear relationship with liquid level. Experiment shows that the measurement accuracy could reach to 0.27mm.Another level detection based on image process is discussed. It takes photo of the inner wall of sample tube, and measures the distance from the tube’s top to the boundary of the liquid surface on the tube’s inner wall. It can measure the level non-contactably, and is not influenced by the conductivity of detected liquid. Also it can detect if the tube cap is off and if there are bubbles on the liquid surface.A compensation technique for photographing a fast moving object with a slower camera is discussed. A mirror is placed beside the surface of the shooting area at a 45 degree angle toward the direction of the moving object. A mirror moves along with the object at the same speed. CCD camera focuses on the mirror, takes a photo of the object’s image in the mirror. A simulation shows that with the help of this compensation technique, the same camera can take clear photos of a 10 times faster moving object.

  • 【网络出版投稿人】 天津大学
  • 【网络出版年期】2011年 07期
节点文献中: 

本文链接的文献网络图示:

本文的引文网络