面向肿瘤转移早期检测的代谢参数荧光分子深度网络成像方法研究

Tumor metastasis is a major cause of the treatment failure and death for patients with malignant tumors. The early detection and diagnosis of tumor metastasis has important clinical and social value. As the accurate detection of tissue pharmacokinetics is urgently needed for the detection and diagnosis of tumor metastasis, in this project, pharmacokinetic parameters imaging method with high accuracy and resolution will be studied based on deep generative theory. Direct pharmacokinetic parameters imaging based on fluorescence molecular tomography with conditional variational auto-encoder will be studied and constructed, in order to provide reconstructed parametric images with high resolution and accuracy. Accurate simulation technology of pharmacokinetics in phantoms will be studied, in order to construct phantom data generator for dynamic fluorescence imaging and to meet the demand of massive data in deep network training and validation. The early detection of tumor metastasis will be carried out based on tumor bearing mice model, in order to validate the feasibility and valuate the accuracy of the proposed direct pharmacokinetic parameters imaging method in vivo. Finally, direct pharmacokinetic parameter imaging method with high accuracy and resolution can be established based on fluorescence molecular tomography with deep generative model, which will provide an in vivo, dynamic, and noninvasive imaging technology and tool for the early detection and diagnosis of tumor metastasis.

肿瘤转移是导致肿瘤治疗失败与患者死亡的主要原因，对肿瘤转移进行早期有效检测与诊断具有重大的临床和社会价值。针对肿瘤转移检测与诊断对组织药代动力学特性精确检测的迫切需求，本项目围绕生物体内代谢参数高精度、高分辨成像的问题，基于深度生成模型理论，研究建立基于条件变分自编码器的荧光分子代谢参数直接成像方法，实现代谢参数分布的高分辨、高准确度反演；基于数字鼠图谱与真实荷瘤小鼠药代动力学特性，研究荷瘤鼠药代动力学特性的动态荧光仿体精确模拟技术，构建动态荧光仿体数据发生器，解决深度网络训练和验证对海量数据的需求；开展基于荷瘤鼠模型的活体肿瘤转移早期检测实验，验证和评估荧光分子代谢参数直接成像方法在活体应用中的有效性和准确性；最终建立基于深度生成模型的面向生物体内代谢参数高精度、高分辨反演的荧光分子代谢参数直接成像方法，为肿瘤转移早期检测和诊断提供在体、动态、无创影像学技术与手段。

肿瘤及肿瘤转移的早期有效检测与诊断具有重大的临床和社会价值。本项目研究中，1）建立了一种基于深度神经网络的高精度、高分辨荧光分子直接反演成像算法，并采用一系列策略对算法的性能（包括空间分辨率、重建速度、成像灵敏度等）进行进一步提升；2）针对海量数据集构建方法开展研究，解决深度神经网络训练和验证过程中对海量数据的需求；3）借助已有的荧光分子代谢参数直接反演成像方法，基于荷瘤鼠模型对肿瘤早期在体成像检测性能进行测试和验证。研究结果表明，该方法可以获得高精度、高分辨、定量化的代谢参数图像，为在体开展肿瘤早期检测、器官功能评估、药物研发等生物学、医学研究提供有力的影像学工具和技术手段。研究成果目前已发表学术论文共12篇，均为SCI收录。