基于影像定量分析和特征可视化的虚拟内窥镜关键技术研究

Virtual endoscopy (VE) has many advantages over the existing screening options and shows potential as a valuable noninvasive diagnosis tool. There are several challenges remaining for VE to become a diagnosis and/or screening option, such as the concern on associated radiation, low sensitivty on the detection of small polyps and flat lesions, large variation on image intepretation and intepretation efficiency, and limited morphological information of inner surface used for diagnosis. These drawbacks should be addressed before it could become a valuable screening modality. In this study, based on the establishemnt of new VE framework, we aims to develop common key technologies for CT-based virtual colonoscopy, MRI-based virtual cystoscopy and ehanced CT-based virtual intravascular endoscopy. With the establishment of patients' image database, fully-automated image segmentation mitigating the partial volume effect will be developed to extract inner and outer boundaries of hollow organs. Based on the extracted organ wall, geometrical and textural features that demosntrate significant defference between cancerous tissue and normal wall tissue could be extracted and analyzed to detect early sign of abnormalities by either three-dimensional endoscopic views (human observer) or computer-aided detection (computer observer). In addition, Further improvement on computer-aided diagnosis will be expected for the detection of bladder cancer and its invasion, for polyps types, and for component and risk ananlysis of vulnerable plaques based on extracted morphological and texture information. and classified for accurate identification of suspicious patches in the mucosa layer. With the proposed new VE framework and key technologies, the performance of VE systems will be greatly improved, accompanying with simplified procedure for patient preparation and reduced interaction time, for non-invasive detection and screening of high-risk population.

作为前景可观的管腔内部肿瘤/病变检查方法，基于影像的虚拟内窥镜技术（VE）近年来得到快速发展并已在临床得到应用，但要成为临床的有效诊断手段，还面临着辐射量较大、小息肉及平坦型病变的敏感度低、医生对图像的解释诊断结果差异大/效率低、用于诊断的信息不足等问题和挑战。本项目通过建立VE系统的新框架，将可视化形态学信息与影像数据分析相结合，选取VE极富前景的临床应用领域，以虚拟结肠镜、虚拟膀胱镜、虚拟血管内窥镜为研究对象，基于CT和MRI影像数据，对虚拟内窥镜的共性关键技术，包括基于成像模型的数据优化、管腔内/外壁分割方法、基于形态及纹理特征的计算机辅助检测和诊断技术、基于特征的可视化技术等，进行深入研究，以提高VE系统对小息肉及平坦型病变的检测能力，实现肿瘤浸润深度、息肉/肿瘤分型、斑块成分的计算机辅助分析和诊断，从而极大提高VE系统的检测性能，促进VE成为腔内肿瘤及心血管疾病的有效诊断手段。

作为前景可观的管腔内部病变检查方法，基于影像的虚拟内窥镜技术（VE）近年来得到快速发展并已在临床得到应用，但要成为有效的临床筛查手段，VE还面临着辐射量较大、小息肉及平坦型病变的敏感度低、医生对图像的解释诊断结果差异大/效率低、用于诊断的信息不足等问题和挑战。本项目通过建立新型VE系统框架，将可视化形态学信息与影像数据分析相结合，选取VE 极富前景的临床应用领域，以虚拟结肠镜、虚拟膀胱镜、虚拟血管内窥镜为研究对象，利用相应管腔器官的CT/MRI 影像数据，对虚拟内窥镜的共性关键技术开展研究，以提高VE 系统的检测性能。项目在构建规范的虚拟内窥镜影像数据库的基础上，针对 CT成像存在的辐射剂量与图像质量间的矛盾，以及MRI成像存在的运动及噪声等伪迹，提出了系列基于成像模型的新型重建框架、数据优化算法及伪影校正方法，实现了基于低剂量CT成像的虚拟内窥镜，可使辐射剂量降低70%，并有效提高了胸腹部MRI成像质量。为提高系统对小息肉/平坦病变的检测能力，项目将整个腔壁作为研究对象，提出了多种膀胱、结肠与血管内外壁的准确分割算法，并创新提出可用于不同成像模态的管腔内外壁通用分割框架；利用获得的完整腔壁及表征病灶变化的影像特征，提出了基于腔壁特征差异的腔内病灶自动检测方案，提高系统对小息肉/平坦型病变、膀胱赘生物和血管斑块的检测性能。在此基础上，项目进一步将新型虚拟内窥镜(VE)系统框架与影像组学分析相结合，构建了息肉良恶性、肿瘤分级、肌层浸润性及分期等的预测模型。基于构建的影像数据库，对提出的检测和诊断模型的初步评价结果表明，所有预测模型的AUC值均高于0.85，极大提升了VE系统的无创检测与诊断性能。在此基础上，项目组与多家三甲医院合作，提出基于新型VE系统的结直肠癌、膀胱癌筛查流程，并开展了临床对照及筛查实验研究。这些研究成果将极大促进VE成为腔内肿瘤及病变的无创筛查及预后管理手段。