【作者】 王楠；

【导师】 朱守平； 陈雪利；

【作者基本信息】 西安电子科技大学 ， 生物信息科学与技术， 2022， 博士

【摘要】 光学投影断层成像技术(Optical Projection Tomography,OPT)通过采集样本多个角度的投影图像,结合重建算法获取样本的三维结构像(透射式OPT)或荧光图像(激发式OPT)。该技术可实现毫米级样本微米级的空间分辨率,具有动态成像、无辐射、成本低、使用方便等优势,为胚胎、组织、器官等研究提供了一种有效三维影像学手段。然而,激发式OPT需要样本进行荧光标记,这会引起光毒性以及光漂白等问题;透射式OPT可以提供样本微米级或亚微米级分辨率的结构影像,却无法获取样本特异性分子化学组分信息。拉曼光谱成像技术可以提供样本分子组分的指纹信息,尤其是基于自发拉曼效应的宽场成像方式是实现大视场、高分辨率免标记三维显微成像的较好选择;但是该技术无法测定样本微结构以及成分的具体空间分布信息。鉴于此,本文将这两种光学分子成像技术进行模态融合,开展双模态光学-拉曼投影断层成像技术与方法研究,实现三维样本显微结构与分子组分融合影像的获取。本文针对双模态光学-拉曼投影断层成像技术及方法进行探讨研究,主要内容如下:1.探讨了双模态光学-拉曼投影断层成像技术的可行性。首先,搭建了一套光学投影断层成像系统,并对系统性能进行了全面测试;其次,搭建了一套宽场拉曼光谱成像系统,通过对不同样本的信号检测,验证了系统的可行性,为进一步搭建双模态光学-拉曼投影断层成像系统奠定了基础。考虑样本散射对信号产生和探测的影响,利用贝塞尔光的长聚焦和自我修复特性,搭建了基于贝塞尔光的拉曼光谱成像系统,通过对二甲基亚砜以及对乙酰氨基酚的光谱信号测试,验证了系统的可行性;通过对不同浓度对乙酰氨基酚及其在散射介质中的拉曼光谱测试,证实了所搭建系统在弱散射样本中的拉曼光谱检测潜力,为弱散射样本的光学-拉曼投影成像提供新思路。2.围绕三维样本显微结构和分子组分融合信息的双模态获取,研发了双模态光学-拉曼投影断层成像系统。该系统采用“双路激发、共路探测”架构,即光学投影成像和拉曼投影成像支路分别采用强度可调白光光源和固体激光器产生的均匀平面光束作为激发光源,收集端采用同一个CCD相机作为投影图像采集探测器。首先,本文对双模态系统的性能进行了测试,包括空间分辨率和检测灵敏度,结果表明该系统具备5.12-36.42μm的跨尺度空间分辨能力以及优于70 m M的脂质分子检测能力;其次,通过单微球和多微球实验验证了该双模态系统及其配备算法的可行性和有效性;最后,通过斑马鱼、拟南芥以及果蝇等模式生物样本的双模态成像与三维图像重建,证明了在生物样本方面的成像能力和应用潜力。3.针对传统解析类投影断层重建算法所需投影数据量大、采集时间长等问题,本文构建了基于迭代算法和深度学习网络的稀疏重建方法框架,重建所需数据量减少至传统算法的1/20。首先,建立了基于结合像素顶点驱动模型-全变分正则化的联合代数迭代算法和基于φ-net深度学习网络的稀疏重建方法,并利用仿真数据对两个重建方法的性能进行了对比和评估;其次,基于两套方法框架对上一章获取的生物样本的双模态光学-拉曼投影图像进行重建,验证稀疏重建方法的应用能力。结果表明,当投影数量减少到30个时,迭代算法可以重建出可接受的图像,重建所需数据量减少为传统算法的1/6;而基于φ-net深度学习网络在投影数量减少为9时,重建图像依然具有较高的准确性,重建所需数据量减少至传统算法的1/20。4.考虑传统断层成像常规样本固定方式不利于活体成像,设计了一种摒弃成像腔的新型样本固定方式,由此会带来数据采集投影角度缺失问题,本文研究了在有限角度数据采集条件下的双模态光学-拉曼投影断层成像的精度和速度问题。首先,建立了基于结合像素顶点驱动模型-全变分正则化的代数迭代算法和两阶段深度学习网络的有限角度投影重建方法,并利用仿真数据对两个重建方法的性能进行了对比和评估,确定了满足最佳成像结果的最小采集角度;其次,在确定的最小采集角度范围内,进一步结合稀疏重建策略减少数据采集时间,利用上述有限角度投影重建算法对实测数据进行重建分析,确定有限角度内的最少投影数量;最后,探索了用于有限角度数据采集的新型样本固定方式,并在光学投影断层成像子系统上进行了验证;通过采集实验数据并进行有限角度重建,结合结构相似度、均方根误差等评价指标,证明了该新型固定方式和有限角度重建方法的应用能力。更多还原

【Abstract】 Optical projection tomography(OPT)technology can obtain a three-dimensional(3D)structural image(transmission OPT)or fluorescence image(emission OPT)of a sample by collecting the projection images from multiple angles and combining them with a reconstruction algorithm.It enables micro-level spatial resolution at millimeter-scale samples,and it has the advantages of dynamic imaging,no radiation,low cost,and ease to use,providing an effective 3D imaging method for embryo,tissue,and organ research.However,the emission OPT requires fluorescence labeling of the sample,which will induce the issues such as phototoxicity and photobleaching.The transmission OPT can provide structural images of the sample at micro-or sub-micron resolution.However,this technique cannot obtain the specific molecular composition information of samples.Raman spectroscopic imaging technology can provide fingerprint information of molecular chemical bonds of samples.Spontaneous Raman scattering-based wide-field imaging is a better choice for realizing a large 3D field of view and high-resolution imaging in a labelfree manner.However,this technique cannot determine the microstructure of samples and specific spatial distribution information of chemical components.Given this,we integrated these two optical molecular imaging technologies to develop a dual-modality optical-Raman projection tomography technique and method to acquire 3D sample microstructural and molecular components fusion images.In this dissertation,we investigated the technology and applications of dual-modality optical-Raman projection tomography imaging.The main contributions of this dissertation can be summarized as follows:1.We explored the feasibility of the dual-modality optical-Raman projection tomography technique.Firstly,we built an optical projection tomography system and verified the system performance comprehensively.Secondly,a wide-field Raman imaging system was built,and the feasibility of the system was verified by detecting the Raman signal of different samples,which laid a foundation for the further construction of the dual-modality optical-Raman projection tomography system.Further,considering the effect of sample scattering on signal generation and detection,we built a Bessel beam-based Raman spectroscopic imaging system by using the long focusing and self-healing properties of the Bessel beam.The feasibility of the system was verified by investigating the Raman spectral signals of dimethyl sulfoxide and acetaminophen.We then explored the potential of the system in detecting the signals at weakly scattering samples by testing the Raman spectra of different concentrations of acetaminophen in the scattering medium,which provided a new idea for optical-Raman projection imaging of the weakly scattering samples.2.A dual-modality optical-Raman projection tomography system was developed to obtain the fusion information of 3D sample microstructure and molecular components.The system adopts the architecture of “dual excitation,common-path detection”,that is,the optical projection imaging and Raman projection imaging used an intensity-tunable surface light source and a solid-state laser as the excitation sources to generate uniform plane beam,respectively,while sharing the same CCD camera to obtain the projection images of samples.Firstly,the proposed system,including spatial resolution and sensitivity,was detected,and these results showed that the system had a cross-scale spatial resolution of 5.12-36.42 μm and a lipid molecule detection capability better than 70 m M.Second,the feasibility and effectiveness of the dual-modality system and its reconstruction algorithm were investigated by single and multi-microsphere experiments.Finally,we validated the imaging capability and potential of the system in biological samples by dual-modality imaging and 3D reconstruction of model biological samples such as zebrafish,Arabidopsis,and Drosophila.3.To address the problems of large amount of projection data and long acquisition time required by traditional analytical algorithm-based projection tomography imaging,this dissertation constructed a framework of iterative algorithm and deep learning network-based sparse-view reconstruction scheme,which can reduce the amount of data required for reconstruction to 1/20 of the traditional algorithm.We first established the pixel vertexdriven model-based total variation regularization simultaneous algebra reconstruction technique and φ-net deep learning network-based sparse-view reconstruction scheme.The performance of these two reconstruction methods was then compared and evaluated by using simulation data.Second,the dual-modality optical-Raman projection images of biological samples acquired in the previous chapter were reconstructed based on these two reconstruction algorithms to verify the application capability of the spare-view reconstruction schemes.These results demonstrated that the iterative algorithm can reconstruct acceptable images when the number of projections was reduced to 30,and the data required for reconstruction was reduced to 1/6 of the traditional algorithm.When the number of projections was decreased to 9,the reconstructed images still have high accuracy,the data desired for reconstruction based on the φ-net-based deep learning algorithm was decreased to 1/20 of the conventional algorithm.4.Considering the conventional sample fixation method of traditional tomography is not conducive to in vivo imaging,a new sample fixation method is designed to abandon an imaging cavity,which will bring the problem of missing projection angle of data acquisition,this dissertation investigated the accuracy and speed of dual-modality optical-Raman projection tomography under limited-angle data acquisition.First,the pixel vertex-driven model-based total variation regularization algebra reconstruction technique and two-stage deep learning network-based limited-angle reconstruction schemes were established.The performance of these two algorithms was compared and evaluated using the simulation data to determine the minimum acquisition angle that satisfied the best reconstruction results.Second,to further reduce the data acquisition time,combined with the sparse-view reconstruction strategy,we used the above-mentioned limited-angle reconstruction schemes to reconstruct the experimental data,so as to determine the minimum number of projection data within the limited-angle.Finally,this dissertation continued to explore a novel sample fixation method for limited-angle data acquisition and validated the performance by acquiring different biological sample data from the dual-modality projection tomography sub-system.Combined with the evaluation indexes such as structural similarity and root mean square error,we demonstrated the application capability of this novel fixation method and limited-angle reconstruction algorithm.更多还原