癌症是人类最难以攻克的疾病之一，一直以来就未有根治的方法。为了提高癌症患者总体的生存率，通过肿瘤标志物的早期检测以及预后判断获得了专家的认可，而循环肿瘤细胞便是关键的肿瘤标志物之一。对循环肿瘤细胞的研究分为富集和检测两个部分，传统富集方式是通过免疫纳米磁纳米颗粒结合循环肿瘤细胞表面的上皮细胞黏附因子（EpCAM，epithelial cell adhesion molecule），并通过磁场对结合磁颗粒的循环肿瘤细胞进行富集，然而不是所有的循环肿瘤细胞都含有EpCAM，并且当肿瘤细胞进入血液中时，有时会伴随着EpCAM的丢失，因此通过免疫纳米磁颗粒结合上皮细胞黏附因子的富集方式存在漏检的问题。传统的检测方式是基于免疫荧光染色使循环肿瘤细胞发光，并通过检测不同的荧光来对其进行计数。通过免疫荧光染色的检测方法只能对循环肿瘤细胞进行计数，包含的信息量极少；同时，对循环肿瘤细胞计数后细胞失活，无法继续进行下游检测；基于免疫荧光染色的方法也无法对循环肿瘤细胞进行分子分型。为了改善传统富集和检测方式的缺点，本文开发了基于3D打印微流控芯片的循环肿瘤细胞操控方法以及基于拉曼光谱技术的肿瘤细胞检测方法。

本文的研究包括：（1）设计可测试不同3D打印机的三维测试模型，通过对比不同打印机制备测试模型的细节，得到最优制备微流控芯片的商业化3D打印机；（2）提出了大尺寸微流控通道内的惯性聚焦方法，通过3D打印大尺寸微流控芯片，实现肿瘤细胞的惯性聚焦；（3）提出了拉曼光谱分峰算法，完成不同肿瘤细胞亚型之间、不同肿瘤细胞外囊泡之间的拉曼光谱有效分类；（4）提出了支撑材料去除量化方法，研究了最优工程化处理微流控芯片支撑材料的方法，为3D打印微流控芯片量产打基础。具体的工作包括：

第一，研究了四种常见的3D打印方式（喷墨固化、立体光刻、数字光投影和熔融堆积）以及其典型的商业化3D打印机的原理和性能。设计了两种测试模型：3D打印机测试模型和微流控通道测试模型。通过两个测试模型的打印件之间的对比，得出代表喷墨固化的ProJet 3600 HD的性能最好，因此接下来的微流控芯片由ProJet 3600 HD打印制备。

第二，通过理论、仿真和实验实现了细胞在3D打印大尺度微流控芯片内的惯性聚焦。设计了一种3D打印的蜿蜒微流控通道，从理论上分析了小球在蜿蜒通道中的受力，根据小球的受力情况得出在通道截面尺寸变大时，需要改变的两个物理量（蜿蜒通道曲率半径和流速）。通过仿真，得到小球实现三维惯性聚焦时的最优曲率半径和最优流速（r = 5.9 mm，v = 0.1 m/s）。打印了最优的曲率半径的微流控芯片并进行荧光小球实验，发现不同流速的实验结果与仿真结果在一定程度上具有一致性。最后使用两种肿瘤细胞（4T1和MCF-7）进行了实验，其聚焦半峰宽分别为23.06 μm和36.45 μm，很好的实现了肿瘤细胞在微流控通道内的惯性聚焦。

第三，提出分峰算法，通过分峰算法提取拉曼光谱中的特征，并通过机器学习对不同细胞和细胞外囊泡的拉曼光谱进行有效分类。（1）研究了不同乳腺癌细胞亚型拉曼光谱分类，首先测量了四种乳腺癌细胞亚型（导管A型、导管B型、非导管HER-2阳性与三阴性）与一种正常乳腺细胞的拉曼光谱；通过三步预处理，对细胞拉曼光谱进行去基线、降噪和平滑；接着通过极值寻峰和分峰算法得到每个光谱的13个峰的峰位、峰强、半峰宽与峰面积特征。最后使用支持向量机对不同的乳腺癌细胞亚型与正常乳腺细胞进行分类以及五种细胞之间进行分类，得到两种乳腺癌细胞之间的分类正确率最高为96.80 %，五种细胞分类的正确率为72.35 %。（2）测量了肿瘤细胞和正常细胞的外囊泡拉曼光谱。使用聚乙二醇方法分离细胞培养液中的胞外囊泡，然后合成了一种生物相容性较好的拉曼增强材料并表征了其增强因子，接着测量了四种不同的细胞外囊泡的增强拉曼光谱，并用PCA-SVM算法实现了不同胞外囊泡的分类，分类正确率达到85 %以上。

第四，为了使3D打印微流控芯片可以量产，提出了其支撑材料工程化的后处理方法。不同的3D打印方式中都存在支撑材料，对于ProJet 3600 HD打印的微流控芯片存在外部支撑材料和内部支撑材料两种。对于两种支撑材料分别提出了量化方法。对于外部支撑材料，根据质量的减少使用质量损失率来评价外部支撑材料的去除率，通过研究得出菜籽油热浴为最优去除方法，其质量损失高达89.9 %；对于内部支撑材料，根据支撑材料的不透光性质提出了透光率来评价内部支撑材料的去除效率，通过研究不同去除剂、不同温度和不同去除时间对内部支撑材料的影响，得出使用70 ℃食用油去除10 min的最优工程和去除条件。

本文为了解决传统循环肿瘤细胞富集方式的缺点，研究了不同商业化3D打印机制备微流控芯片的优缺点，并实现了细胞在3D打印大尺度微流控芯片内的惯性聚焦。为了解决传统循环肿瘤细胞检测方式的缺点，提出分峰算法，完成了不同乳腺癌细胞亚型、癌细胞和正常细胞外囊泡拉曼光谱的有效分类。为了解决3D打印微流控芯片的量产问题，提出量化方法，得出最优工程化支撑材料去除条件。综上，完成了使用3D打印微流控芯片和拉曼光谱检测循环肿瘤细胞的基础性工作，为实现使用3D打印微流控芯片的循环肿瘤细胞一体化检测平台做铺垫。

Cancer is one of the most difficult diseases for humans to overcome. To improve the overall survival rate of cancer patients, early detection and prognosis of tumor markers including circulating tumor cells (CTCs)have been recognized by experts. The research on circulating tumor cells is divided into two parts: enrichment and detection. The traditional enrichment method is to combine EpCAM (Epithelial cell adhesion molecule) on the surface of circulating tumor cells with immune nanomagnetic nanoparticles. But not all circulating tumor cells contain EpCAM, and sometimes the EpCAM could be loss when tumor cells enter the blood As a traditional method, immunofluorescence staining is used to count the CTCs based on cell fluorescence  However, this method can cause damage on viability of CTCs, limiting the downstream analysis. The method based on immunofluorescence staining is also unable to identify the subtype of circulating tumor cells. To improve the shortcomings of traditional enrichment and detection methods, this paper developed a CTC manipulation method based on 3D printed microfluidic chip and a tumor cell detection method based on Raman spectroscopy.

The research in this thesis includes: (1) by comparison the fabrication of the test models, a commercial 3D printer that optimally prepares microfluidic chips; (2) 3D printing of large-sized microfluidic chips to achieve inertial focusing of tumor cells; (3) proposed a novel Raman spectrum splitting algorithm, so that different tumor cell subtypes and different tumor extracellular vesicles can be effectively classified; (4) studied the optimal engineering method of removing support materials of microfluidic chips, for chip mass production. Specific work was described as following.

Firstly, the principle and performance of four common 3D printing methods (inkjet printing, stereolithography, digital light projection, and fusion deposition modeling) were studied. For the four 3D printers, two test models, 3D printer test model and a microfluidic test model were designed. The 3D printer test model is mainly used to test the printing performance of the printer, and the microfluidic channel test model is mainly used to test the performance of the four printers on different microfluidic channels. By comparing the printed test models, the inkjet printer ProJet 3600 HD has the best performance Therefore, the following serpentine microfluidic channels  are printed by ProJet 3600 HD.

Secondly, a 3D printed serpentine microfluidic channel was designed to achieve inertial focusing of tumor cells within the channel.  The forces acted on the particles in the serpentine channel were theoretically analyzed. When the channel section size becomes larger, two physical quantities (flow velocities and curvatures of the serpentine channel) need to be changed. Then, the focus state of the small particles at the outlet of the microfluidic channel when the curvature radius and the flow velocity are changed is simulated by COMSOL software, and the focus state is divided into six categories. The microfluidic chip with the optimal curvature radius (r = 5.9 mm) was printed, and the experimental results of different flow rates were compared with the simulation results. Finally, the optimal focusing conditions (r = 5.9 mm, v = 0.1 m / s) were used to perform experiments using tumor cells (4T1 and MCF-7), and the tumor cells in the microfluidic channel were well inertial focused.

The Raman spectra of different tumor cells are measured and classified by machine learning algorithms. (1) The Raman spectrum classification of different breast cancer cell subtypes was studied. First, four breast cancer cell subtypes (Luminal A, Luminal B, non-luminal HER-2 positive and triple negative) and a normal breast cell were measured. The Raman spectrum of the cell can be classified through three steps of preprocessing. The peak position, peak intensity, half-peak width, and peak area of the 13 peaks of each spectrum can be obtained by peak searching and peak splitting algorithm. Finally, the support vector machine was used to classify different breast cancer cell subtypes and normal breast cells and to classify the five cells. The highest accuracy between two subtype cell classifications was 96.8 %. The accuracy of the five cell classifications was 72.35 %. (2) The Raman spectra of tumor extracellular vesicles and normal extracellular vesicles were measured.Extracellular vesicles were isolated from the cell culture medium using the polyethylene glycol method. A biocompatible Raman enhancement material was synthesized, and its enhancement factor was characterized. Four different extracellular cells were measured. The enhanced Raman spectrum of vesicles and the classification of different extracellular vesicles using the PCA-SVM algorithm achieved a classification accuracy rate of more than 85 %.

It was found in the research that if ProJet 3600 HD is used for mass production of microfluidic chips, the engineering method of microfluidic chip post-processing needs to be studied. For microfluidic chips printed by ProJet 3600 HD, there are two types of external support materials and internal support materials. Quantitative methods are proposed for the two supporting materials. For external support materials, this article proposes a mass loss rate based on the reduction in mass to evaluate the removal rate of external support materials. For internal support materials, this article proposes a light transmittance based on the opaque nature of the support materials to evaluate the removal of internal support materials. effectiveness. According to the quantification method of external support material, four removal methods of external support material were studied, and the optimal removal method was obtained. According to the internal support material quantification method, the effects of different removers, different temperatures, and different removal times on the support materials were studied, and the optimal engineering and removal conditions for removing 10 minutes of edible oil at 70 °C were obtained.

In order to improve the shortcomings of traditional circulating tumor cell enrichment methods, the advantages and disadvantages of microfluidic chips prepared by different commercial 3D printers were studied, and the inertial focusing of cells in 3D printing large-scale microfluidic chips was achieved. To improve the shortcomings of traditional circulating tumor cell detection methods, a peak splitting algorithm was proposed to complete the effective classification of Raman spectra of different breast cancer cell subtypes, cancer cells and normal extracellular vesicles. To solve the problem of mass production of 3D printed microfluidic chips, a quantitative method was proposed to obtain the optimal engineering support material removal conditions. In summary, the basic work of using 3D printed microfluidic chips and Raman spectroscopy to detect circulating tumor cells has been completed, laying the foundation for an integrated detection platform for circulating tumor cells using 3D printed microfluidic chips.