基于支持向量机与图搜索的PET-CT图像中肺肿瘤分割算法研究

In recent years, cancer has been becoming more and more serious to public health, and lung tumors are the leading cause of cancer death. In worldwide, annual mortality from lung cancer is expected to exceed 13 million by 2030. Due to the integration of the advantages of the two imaging, PET-CT bimodality imaging has become one of the best choices of this stage lung cancer diagnosis, in order to non-invasively describe the physiological and morphological properties of lung tumors. However, lung tumors in PET images and CT images are often displayed differently, and the quality of PET-CT images are comparatively low; besides of individual differences, the anatomy, physiology and pathology of the human organs are extremely complex, which makes lung tumor segmentation in PET-CT become a major research focus and difficulty. In order to overcome the problems, this project will be concentrated on researches and development of lung tumor detection based on support vector machine (SVM), precise segmentation of lung tumor based on graph search,so as to accurately provide physiological metabolism and anatomical structure information of lung tumors and related structures. This project will provide basis and proofs for doctors to determine the objective staging of lung cancer, accurate diagnosis, rapid treatment plan and effectively evaluation of treatment, which will greatly reduce trial-and-error efforts and unreliability of lung cancer diagnosis and treatment.

近年来，癌症对人类健康的危害日益严重，肺癌则居于癌症死亡病例中的首位，到2030年，全球每年肺癌的死亡人数预计将超过1300万人。PET-CT双模态影像技术由于整合了两种成像技术的优势，为描述肺肿瘤的生理学与形态学特性提供了非侵入式手段，已经成为现阶段肺肿瘤诊断的最佳选择之一。然而，由于PET与CT图像中肺肿瘤显像通常有较大差异，PET-CT图像质量较低，人体器官的解剖、生理与病理的结构较复杂、个体差异较大等问题的存在，使得PET-CT图像中肺肿瘤的精确分割成为当前医学影像处理研究领域的一大研究热点与难点。为克服上述问题，本项目拟进行基于支持向量机的肺肿瘤检测、基于图搜索的肺肿瘤精确分割等算法研究，为客观地确定肺癌分期、准确地做出诊断、快速地制定治疗方案和有效地评估治疗后效果提供决策依据，这将大大降低肺癌诊断与治疗的盲目性和不可靠性，降低死亡率，提高生存率。

近年来，癌症对人类健康的危害日益严重，肺癌则居于癌症死亡病例中的首位，到2030年，全球每年肺癌的死亡人数预计将超过1300万人。PET-CT双模态影像技术由于整合了两种成像技术的优势，为描述肺肿瘤的生理学与形态学特性提供了非侵入式手段，已经成为现阶段肺肿瘤诊断的最佳选择之一。然而，由于PET与CT图像中肺肿瘤显像通常有较大差异，PET-CT图像质量较低，人体器官的解剖、生理与病理的结构较复杂、个体差异较大等问题的存在，使得PET-CT图像中肺肿瘤的精确分割成为当前医学影像处理研究领域的一大研究热点与难点。为克服上述问题，本项目研究PET-CT 多模态影像中的肺肿瘤精确分割算法，先后提出了基于随机游走和图割联合算法的PET-CT肺肿瘤分割、三维图割算法结合随机游走算法的PET-CT肺肿瘤分割方法、基于随机森林和BSS特征的PET-CT肺肿瘤分割、基于统计形状模型和可形变图割的CT图像脏器算法、基于统计形状模型与非均匀图搜索分割算法等五种方法，从而解决肺肿瘤个体差异、不同模态数据中形状有所差异等问题，从而降低肺肿瘤诊断的假阴性和假阳性率，并在国内外高水平学术期刊上发表SCI论文7 篇、EI检索论文4篇，获批国家发明专利授权3项，申请国家发明专利4 项。另外，研发计算机辅助肺癌诊断软件平台MPAV，对肺肿瘤区域的定量分析提供可靠的方法与工具，为医生客观地确定肺癌研究、准确地做出诊断、快速地制定治疗方案提供决策依据，有助于降低肺癌诊断与治疗的盲目性和不可靠性。