基于Radiomics的中心型肺癌定量治疗评估与预后研究

Central Lung Cancer (CLC) often occurs in the vicinity of the mediastinum and hilus of lung. Generally, it can be treated by surgery of doing pneumonectomy, which patients cannot always bear. In recent years, CLC therapy mainly depends on radiotherapy, hybrid with chemotherapy. Multi-modal imaging technologies, including X-ray, CT and PET, play a significant role on lung disease diagnosis, therapy and prognosis assessment. In this project, we first study the feature extraction methodology from multi-modal images, which presents the tumor heterogeneity. Next, quantitative image biomarkers are then selected from the extracted features by utilizing Support Vector Machine (SVM). Afterwards, we established the quantitative therapy and prognosis assessment model based on artificial intelligence discriminator. Furthermore, the Bayesian Belief Network (BBN) is used to integrate quantitative image biomarkers and other types of biomarkers, in order to improve the accuracy of prognosis assessment. During the whole period of this project, we will collaborate with our partner to establish a dataset containing multi-model images of lung and medical follow-up records. The achievements of this project will be significantly important to CLC quantitive therapy and prognosis assessment in both theoretical research and industrial products.

中心型肺癌（CLC）常发生于临近纵膈和肺门的部位，手术通常只能做全肺切除，患者往往难以承受。近年来，CLC的治疗主要以放疗、化疗相结合为主。X线、CT和PET等多模态影像对于肺部病变的诊断、治疗和预后评估都起着重要的作用。本项目首先研究了从多模态影像中提取新的反映肿瘤异质性的特征方法，进而利用支持向量机（SVM）选择定量影像标记物；然后，我们建立了一种基于人工智能判别器的定量治疗与预后评估模型，利用贝叶斯信任网络（BBN）来融合定量影像标记物与其他类型标记物，来提高预后评估的准确性。在此过程中，我们与合作单位共同建立包含肺部多模态图像与随访记录的数据集。此项目的研究成果将对中心型肺癌的定量治疗和预后评估有着非常重要的理论研究意义和实际应用价值。

本项目围绕中心型肺癌临床诊断和预后风险评估中的实际问题，研究肺部多模态影像的配准和分割方法，从多模态影像中提取能够反映肿瘤异质性的影像组学特征方法，并进一步使用传统的机器学习方法和深度学习方法对肺鳞癌、肺腺癌、肺小细胞癌进行组织学亚型分类。本项目还研究了基于增强CT的肺部的图像薄层重建算法，评估中心型肺癌的血管侵犯情况。此外，本项目还收集了来自三所不同三甲医院的肺癌骨转移数据集、中心型肺癌的多模态影像数据集、中心型肺癌放化疗前后的PET/CT数据集。本项目的研究成果和收集的数据，对临床上通过影像学方法初步判定中心型肺癌是否侵犯血管、对中心型肺癌的定量治疗和预后评估、以及其他癌症和肺部疾病的影像组学诊断方法都有一定的理论研究意义和实际应用价值。