毕业生知识库

现代外科学经历不断地发展，已经逐渐步入精准外科时代。对于精准外科而言，手术过程中追求完整的肿瘤切除、最少的创伤以及最短的手术时间。虽然计算机断层扫描成像、磁共振成像等影像技术可以辅助医生进行术前精准诊断和术后疗效评估，但是这些成像技术难以在术中推广使用。超声成像可以用于术中肿瘤检测，但其分辨率有限。手术过程中医生依然凭借视诊、触诊等主观经验切除肿瘤，这容易导致肿瘤残余或者正常组织过度切除。因此，面向精准外科手术的需求，目前亟需研发有效的术中成像技术，辅助医生在术中定位肿瘤、判断周围正常组织结构、识别肿瘤边界。

近红外荧光成像以光作为信号载体，可以在术中动态可视化肿瘤组织，为实现精准外科手术带来了机遇。其中，成像波长在700-900 nm的近红外一区荧光成像目前已经应用于肿瘤检测、药物评价、肿瘤机制研究，并得到国际学者的广泛关注。但是近红外一区荧光成像受环境光和组织自发荧光的干扰，术中肿瘤检测的精度有待进一步提升。相比而言，成像波长在1000-1700 nm的近红外二区（second near-infrared window, NIR-II）荧光成像受组织自发荧光以及组织的吸收散射影响小，具有更深的组织穿透深度、更高的成像灵敏度和更高的成像信背比，有望实现术中更精准的肿瘤检测。但是，目前针对NIR-II荧光成像的研究大多集中在动物层面，而且以二维平面成像为主，难以获得肿瘤的三维信息，此外NIR-II荧光成像的临床应用还处于初步探索阶段。因此，本文针对目前近红外荧光手术导航存在的问题，基于神经网络的激励机制，从肿瘤在体三维成像、肿瘤及瘤周组织多目标成像、肿瘤边界实时检测三个方面开展了NIR-II荧光成像方法的研究，探索了NIR-II荧光成像引导肿瘤切除的新策略。具体来说，本文的主要研究工作如下：

1、针对肿瘤在体精准、快速三维成像的问题，提出了基于激励机制神经网络的NIR-II荧光断层成像方法。首先研究了多种生物组织对NIR-II荧光光子的散射、吸收特性，构建了NIR-II荧光光子传输的蒙特卡洛仿真数据集。然后提出了基于激励机制的神经网络，直接拟合NIR-II荧光光子在不同生物组织中的复杂传输过程，避免了传统数理模型的建模误差以及复杂的迭代优化求解过程。网络中通过引入激励机制，使网络能够自主关注与光源相关的节点，提高了重建结果的形态恢复度。网络损失函数中通过加入重建光源与真实光源的重心距离损失项，提高了重建结果的定位精度。仿真实验和小动物活体实验显示，相较已有的神经网络荧光断层成像方法，肿瘤的定位误差减少35%，形态恢复度提升42%，相较传统数理模型断层成像方法，重建速度由分钟级提高到毫秒级。

2、在肿瘤三维精准成像的基础上，针对肿瘤及瘤周组织多目标成像的需求，提出了基于激励机制U-Net网络的NIR-II多谱段荧光成像方法。首先针对术中NIR-II多谱段荧光图像的特性，构建了基于激励机制的U-Net网络，研究了肿瘤与肿瘤周围动静脉的术中融合成像方法，实现了术中多谱段荧光图像的快速融合。然后构建了NIR-II多谱段荧光手术导航系统，开展了术中3级和4级脑胶质瘤及血管多目标成像和荧光成像引导的肿瘤切除。临床试验结果表明，NIR-II多谱段荧光成像方法可以在术中识别肿瘤，同时成像肿瘤的供血动脉和引流静脉。在肿瘤切除过程中，通过肿瘤及肿瘤周围动静脉融合成像的引导，实现了对入组患者肿瘤供血动脉的精准阻断。在精准切除肿瘤的基础上，患者术中平均出血量相较同期对照组降低近40%。

3、在上述基于激励机制的NIR-II荧光成像方法研究基础上，针对肿瘤边界在体实时检测的需求，提出了基于OTSU算法的NIR-II荧光肿瘤边界实时检测方法。首先分析了NIR-II在体荧光图像的特性，提出了基于OTSU算法的NIR-II荧光肿瘤边界实时检测方法，用于在体实时检测肿瘤边界。在此基础上构建了NIR-II实时荧光手术导航系统，并开展了囊性肾肿瘤术中边界实时检测与荧光成像引导的肿瘤切除，实现了对入组患者肿瘤的完整切除，避免了切除过程中肿瘤破裂，同时最大程度地保留了正常肾实质。患者肿瘤完整切除率达100%，术后随访最长17个月没有复发转移。

总之，本文以新型的NIR-II荧光成像为技术手段，提出了基于激励机制的NIR-II荧光成像方法，开展了肿瘤在体三维成像、肿瘤及瘤周组织多目标成像、肿瘤边界实时检测方法研究，并构建了NIR-II荧光手术导航系统。最后通过医工交叉合作，将提出的NIR-II荧光成像方法应用于临床肿瘤切除手术。初步的研究结果表明，提出的NIR-II荧光成像方法可以在术中有效地辅助医生实施肿瘤切除，为患者带来了临床获益。

After continuous development, modern surgery has gradually entered the era of precision surgery. For precision surgery, radical resection of the tumor, minimal trauma, and the shortest duration of operation are pursued during the surgical process. Although imaging technologies such as computed tomography and magnetic resonance imaging can assist surgeons in preoperative precision diagnosis and postoperative outcomes evaluation, these imaging techniques are difficult to popularize and use in surgery. Ultrasound imaging can be used for intraoperative tumor detection, but its resolution is limited. Surgeons still rely on subjective experiences such as inspection and palpation to remove tumors, which often leads to tumors being missed or excessive resection of normal tissues. Therefore, in order to meet the needs of precision surgery, it is urgent to develop effective intraoperative imaging technology to assist surgeons in locating tumors, judging peritumoral normal tissue structures, and identifying tumor boundaries during surgery.

Using light as a signal carrier, near-infrared fluorescence imaging can dynamically visualize tumors during surgery, which provides opportunities for the realization of precise surgery. Therein, fluorescence imaging in the first near-infrared window (NIR-I) with an imaging wavelength of 700-900 nm has been applied in the research of tumor detection, drug evaluation, and tumor mechanism, and has attracted extensive attention from international scholars. However, the NIR-I fluorescence imaging is interfered by ambient light and tissue autofluorescence, which leaves scope for further improvement in the precision of intraoperative tumor detection. In contrast, fluorescence imaging in the second near-infrared window (NIR-II) with an imaging wavelength of 1000-1700 nm is less affected by tissue autofluorescence, absorption, and scattering, which can achieve deeper tissue penetration, higher imaging sensitivity, and higher signal-to-background ratio, making it possible to achieve more precise detection of tumor during surgery. However, most of the current research focused on NIR-II fluorescence imaging is at the animal level with two-dimensional planar imaging, making it difficult to obtain three-dimensional (3D) information of tumors. Besides, the clinical application of NIR-II fluorescence imaging is still in the preliminary exploration stage. Therefore, aiming at the existing problems of near-infrared fluorescence surgical navigation, and based on the excitation mechanism of neural network, this dissertation carried out research on NIR-II fluorescence imaging method, and a new strategy for NIR-II fluorescence imaging-guided tumor resection was explored from three aspects: in vivo 3D imaging of tumors, multi-target imaging of tumors and peritumoral tissues, and real-time detection of tumor boundaries. Specifically, the main research work of this dissertation is described as follows:

1. A NIR-II fluorescence molecular tomography method based on a neural network with excitation mechanism was proposed to address the problem of precise and rapid in vivo 3D imaging of tumors. Firstly, the scattering and absorption properties of NIR-II fluorescence photons in various biological tissues were studied, and a Monte Carlo simulation data set about NIR-II fluorescence photons propagation was constructed. Then, a neural network based on the excitation mechanism was proposed to directly fit the complex propagation process of NIR-II fluorescence photons in diverse biological tissues, which avoided the modeling errors of traditional mathematical models and the complex iterative optimization solution. By introducing the excitation mechanism into the network, the network autonomously focused on the nodes related to the light source, which improved the morphological recovery of the reconstruction results. The location accuracy of the reconstruction results was improved by adding the loss item of the distance between the reconstructed light source and the real light source in the network loss function. In vivo experiments on small animals showed that compared with existing fluorescence molecular tomography methods based on neural networks, the localization error of tumors was reduced by 35%, and the morphological recovery was improved by 42%. Simulation experiments showed that compared with fluorescence molecular tomography methods of traditional mathematical models, the reconstruction speed was increased from the minute level to the millisecond level.

2. Based on the precise 3D imaging of tumors, a U-Net network based on the excitation mechanism for NIR-II multispectral fluorescence imaging was proposed to meet the needs of multi-target imaging of tumors and peritumoral tissues. Firstly, according to the characteristics of intraoperative NIR-II multispectral fluorescence images, a U-Net network based on the excitation mechanism was constructed, the intraoperative fusion imaging method of tumors and surrounding arteries and veins was investigated, and the rapid fusion of intraoperative multispectral fluorescence images was realized. Then, a NIR-II multispectral fluorescence surgical navigation system was constructed. Intraoperative multiple-target imaging of grade 3 and grade 4 glioma and vessels was conducted, and fluorescence imaging-guided tumor resection was carried out. The results of clinical trial showed that the NIR-II multispectral fluorescence imaging method could identify tumors during surgery and detect the feeding arteries and drainage veins of the tumor at the same time. In the process of tumor resection, guided by the fusion imaging of the tumor and surrounding arteries and veins, precise occlusion of the tumor-feeding arteries of the enrolled patients was achieved. On the basis of precise tumor resection, the average intraoperative blood loss of patients was reduced by nearly 40% compared with the control group during the trial.

3. With the research of the above NIR-II fluorescence imaging method based on the excitation mechanism, a real-time NIR-II fluorescence tumor boundary detection method based on the OTSU algorithm was proposed to meet the needs of in vivo real-time detection of tumor boundaries. Firstly, the characteristics of in vivo NIR-II fluorescence images were analyzed, and a real-time NIR-II fluorescence tumor boundary detection method based on the OTSU algorithm was proposed for in vivo real-time detection of tumor boundaries. On this basis, a NIR-II real-time fluorescence surgical navigation system was constructed, and the real-time detection of tumor boundaries and fluorescence imaging-guided tumor resection were carried out. Complete tumor resection of the enrolled patients was achieved, tumor rupture was avoided during resection, and maximal preservation of normal renal parenchyma was achieved at the same time. The complete resection rate of tumors in the patients was 100%, and no recurrence or metastasis was observed for up to 17 months follow-up after surgery.

In general, the novel NIR-II fluorescence imaging was utilized as the technical means in this dissertation, the excitation-based NIR-II fluorescence imaging method was proposed, the research on in vivo 3D imaging of tumor, the multi-target imaging of tumor and peritumoral tissue, the real-time detection of tumor boundaries was carried out, and the NIR-II fluorescence surgical navigation system was constructed. Finally, through medical-engineering cooperation, the proposed NIR-II fluorescence imaging method was applied to clinical tumor resection. Preliminary results showed that the proposed NIR-II fluorescence imaging method can effectively assist surgeons in tumor resection during surgery, which brought clinical benefits to patients.

After continuous development, modern surgery has gradually entered the era of precision surgery. For precision surgery, radical resection of the tumor, minimal trauma, and the shortest duration of operation are pursued during the surgical process. Although imaging technologies such as computed tomography and magnetic resonance imaging can assist surgeons in preoperative precision diagnosis and postoperative outcomes evaluation, these imaging techniques are difficult to popularize and use in surgery. Ultrasound imaging can be used for intraoperative tumor detection, but its resolution is limited. Surgeons still rely on subjective experiences such as inspection and palpation to remove tumors, which often leads to tumors being missed or excessive resection of normal tissues. Therefore, in order to meet the needs of precision surgery, it is urgent to develop effective intraoperative imaging technology to assist surgeons in locating tumors, judging peritumoral normal tissue structures, and identifying tumor boundaries during surgery.

Using light as a signal carrier, near-infrared fluorescence imaging can dynamically visualize tumors during surgery, which provides opportunities for the realization of precise surgery. Therein, fluorescence imaging in the first near-infrared window (NIR-I) with an imaging wavelength of 700-900 nm has been applied in the research of tumor detection, drug evaluation, and tumor mechanism, and has attracted extensive attention from international scholars. However, the NIR-I fluorescence imaging is interfered by ambient light and tissue autofluorescence, which leaves scope for further improvement in the precision of intraoperative tumor detection. In contrast, fluorescence imaging in the second near-infrared window (NIR-II) with an imaging wavelength of 1000-1700 nm is less affected by tissue autofluorescence, absorption, and scattering, which can achieve deeper tissue penetration, higher imaging sensitivity, and higher signal-to-background ratio, making it possible to achieve more precise detection of tumor during surgery. However, most of the current research focused on NIR-II fluorescence imaging is at the animal level with two-dimensional planar imaging, making it difficult to obtain three-dimensional (3D) information of tumors. Besides, the clinical application of NIR-II fluorescence imaging is still in the preliminary exploration stage. Therefore, aiming at the existing problems of near-infrared fluorescence surgical navigation, and based on the excitation mechanism of neural network, this dissertation carried out research on NIR-II fluorescence imaging method, and a new strategy for NIR-II fluorescence imaging-guided tumor resection was explored from three aspects: in vivo 3D imaging of tumors, multi-target imaging of tumors and peritumoral tissues, and real-time detection of tumor boundaries. Specifically, the main research work of this dissertation is described as follows:

1. A NIR-II fluorescence molecular tomography method based on a neural network with excitation mechanism was proposed to address the problem of precise and rapid in vivo 3D imaging of tumors. Firstly, the scattering and absorption properties of NIR-II fluorescence photons in various biological tissues were studied, and a Monte Carlo simulation data set about NIR-II fluorescence photons propagation was constructed. Then, a neural network based on the excitation mechanism was proposed to directly fit the complex propagation process of NIR-II fluorescence photons in diverse biological tissues, which avoided the modeling errors of traditional mathematical models and the complex iterative optimization solution. By introducing the excitation mechanism into the network, the network autonomously focused on the nodes related to the light source, which improved the morphological recovery of the reconstruction results. The location accuracy of the reconstruction results was improved by adding the loss item of the distance between the reconstructed light source and the real light source in the network loss function. In vivo experiments on small animals showed that compared with existing fluorescence molecular tomography methods based on neural networks, the localization error of tumors was reduced by 35%, and the morphological recovery was improved by 42%. Simulation experiments showed that compared with fluorescence molecular tomography methods of traditional mathematical models, the reconstruction speed was increased from the minute level to the millisecond level.

2. Based on the precise 3D imaging of tumors, a U-Net network based on the excitation mechanism for NIR-II multispectral fluorescence imaging was proposed to meet the needs of multi-target imaging of tumors and peritumoral tissues. Firstly, according to the characteristics of intraoperative NIR-II multispectral fluorescence images, a U-Net network based on the excitation mechanism was constructed, the intraoperative fusion imaging method of tumors and surrounding arteries and veins was investigated, and the rapid fusion of intraoperative multispectral fluorescence images was realized. Then, a NIR-II multispectral fluorescence surgical navigation system was constructed. Intraoperative multiple-target imaging of grade 3 and grade 4 glioma and vessels was conducted, and fluorescence imaging-guided tumor resection was carried out. The results of clinical trial showed that the NIR-II multispectral fluorescence imaging method could identify tumors during surgery and detect the feeding arteries and drainage veins of the tumor at the same time. In the process of tumor resection, guided by the fusion imaging of the tumor and surrounding arteries and veins, precise occlusion of the tumor-feeding arteries of the enrolled patients was achieved. On the basis of precise tumor resection, the average intraoperative blood loss of patients was reduced by nearly 40% compared with the control group during the trial.

3. With the research of the above NIR-II fluorescence imaging method based on the excitation mechanism, a real-time NIR-II fluorescence tumor boundary detection method based on the OTSU algorithm was proposed to meet the needs of in vivo real-time detection of tumor boundaries. Firstly, the characteristics of in vivo NIR-II fluorescence images were analyzed, and a real-time NIR-II fluorescence tumor boundary detection method based on the OTSU algorithm was proposed for in vivo real-time detection of tumor boundaries. On this basis, a NIR-II real-time fluorescence surgical navigation system was constructed, and the real-time detection of tumor boundaries and fluorescence imaging-guided tumor resection were carried out. Complete tumor resection of the enrolled patients was achieved, tumor rupture was avoided during resection, and maximal preservation of normal renal parenchyma was achieved at the same time. The complete resection rate of tumors in the patients was 100%, and no recurrence or metastasis was observed for up to 17 months follow-up after surgery.

In general, the novel NIR-II fluorescence imaging was utilized as the technical means in this dissertation, the excitation-based NIR-II fluorescence imaging method was proposed, the research on in vivo 3D imaging of tumor, the multi-target imaging of tumor and peritumoral tissue, the real-time detection of tumor boundaries was carried out, and the NIR-II fluorescence surgical navigation system was constructed. Finally, through medical-engineering cooperation, the proposed NIR-II fluorescence imaging method was applied to clinical tumor resection. Preliminary results showed that the proposed NIR-II fluorescence imaging method can effectively assist surgeons in tumor resection during surgery, which brought clinical benefits to patients.