结合PET/CT图像特征的肺癌病变区域的活动轮廓分割模型研究

With the rapid progress of precision medical technology, identification and delineation of lung cancerous lesion of PET/CT images show an important role in tumor precise medical plans. However, different principle of medical imaging and irregular anatomical structure have given greater challenges for PET/CT image segmentation technology. This project mainly studies the key technologies of PET/CT lung cancerous lesion segmentation based on active contour model (ACM) and includes the following aspects: (1) Based on the feature from low level to high level, the PET/CT feature set of lung cancerous lesion is extracted and described. The relevant information between two different modalities of PET/CT lesions is highlighted. (2) Based on the lesion from simple region to complex region, a novel active contour segmentation model by the constructed feature of lung cancerous lesion is designed. The proposed model can provide the model design and theoretical analysis for PET/CT image segmentation. (3) Further, the mechanism of deep learning is used to optimize the proposed segmentation model. This project builds an ACM-Net based on shared representation learning of PET/CT images. From the above three aspects, this project aims to establish the common feature representation of PET/CT images, and realize the identification and delineation of lung cancerous lesion automatically. The project can provide an auxiliary role for clinicians to make a precise medical plan.

随着精准医疗技术的快速进步，PET/CT图像中肺部病灶区域辨别及勾画已在精准医疗计划制定中显现出重要作用，然而医学图像成像原理差异和不规则性给予了分割技术更大的挑战。本项目利用活动轮廓模型进行PET/CT图像肺癌病变区域的分割，主要研究以下内容：（1）基于特征从底层到高层的策略，提取肺癌病变目标的PET/CT图像特征集，实现病灶区域的特征描述，突出PET/CT两种不同模态病灶区域间的相关联信息。（2）通过上述构造的肺部病灶区域特征，基于病灶区域从简单到复杂的策略，设计基于PET/CT图像肺癌病变区域的活动轮廓模型，为癌变区域分割提供模型设计与理论分析的方法。（3）在此基础上，利用深度学习的理论机制对所设计模型进行优化，构建基于PET/CT图像共享特征学习的活动轮廓模型学习网。本项目通过PET/CT图像的共同特征表达，实现肺部病变区域的自动辨别勾画，为临床医师制定精准治疗计划提供辅助作用。

随着精准医疗技术的快速进步，PET/CT图像中肺部病灶区域辨别及勾画已在精准医疗计划制定中显现出重要作用，然而医学图像成像原理差异和不规则性给予了分割技术更大的挑战。本项目利用活动轮廓模型进行PET/CT图像肺癌病变区域的分割，主要研究以下内容：（1）基于特征从底层到高层的策略，提取肺癌病变目标的PET/CT图像特征集，实现病灶区域的特征描述，突出PET/CT两种不同模态病灶区域间的相关联信息。（2）通过上述构造的肺部病灶区域特征，基于病灶区域从简单到复杂的策略，设计基于PET/CT图像肺癌病变区域的活动轮廓模型，为癌变区域分割提供模型设计与理论分析的方法。（3）在此基础上，利用深度学习的理论机制对所设计模型进行优化，构建基于PET/CT图像共享特征学习的活动轮廓模型学习网。本项目通过PET/CT图像的共同特征表达，实现肺部病变区域的自动辨别勾画，为临床医师制定精准治疗计划提供辅助作用。