基于轮廓波变换的模型纹理特征提取方法和分类识别在胃癌术前分期中的研究

Gastric cancer is one of the most common gastrointestinal tumors in China. A good prognosis for patients with this disease requires choosing the correct therapy, and making the right therapeutic choice requires accurate preoperative staging. The recent development of multi-detector row CT (MDCT) scanner has allowed imaging with a thinner section collimation, translating into increased quality on transverse computed tomography scans and multipalanar reconstruction, contributing to the improved accuracy of TNM staging. However there are still some controversial problems. Serousal infiltration and lymph node status are of particular interest in the pretherapeutic staging of tumors, especially to establish different therapeutic strategies. The criteria for positivity are variable, which has not been well established. The 950 gastric cancer cases that were operated after either MDCT or dual energy spectral CT scans were enrolled in the retrospective study. Compared with the histological findings, three combination methods of contourlet transform with hidden Markov tree (HMT), gray-level co-occurrence matrix (GLCM), and local binary patterns (LBP) are used to integrate the medical imaging analysis method for the establishment of the diagnostic criteria for serosal invasion and lymph node metastasis on both MDCT and dual energy spectral CT imaging. In the prospective study, MDCT and dual energy spectral CT are performed on 300 patients with gastric cancer, in order to verify the accuracy of the diagnostic criteria on multiple data sets. It can not only provide new opportunities for detailed preoperative evaluation of both the tumor infiltration but also lymph nodes determination of gastric cancer, expecting for the future to guide clinical selection of appropriate treatment monitoring the reaction of treatment.

胃癌是发病率较高的常见消化道恶性肿瘤，其预后和治疗方案选择与准确的术前分期有着密切的联系。目前多层螺旋CT（MDCT）已经成为胃癌术前分期的主要检测手段之一。本课题组前期已对950例胃癌进行了MDCT及宝石CT与病理对照分析，发现肿瘤浆膜浸润和淋巴结转移是胃癌分期准确性的关键点，但国际上仍缺乏准确界定浆膜浸润和淋巴结转移的标准。为此，本研究拟将该组病例配合轮廓波与隐马尔科夫树、灰度共生矩阵和局部二值模式等三种经典纹理提取方法融合的医学影像学分析方法，确立胃癌浆膜浸润及淋巴结转移的图像纹理特征，整合出新的算法，并与其他影像指标进行分类识别分析，通过多个数据集、术后病理数据集确立MDCT和宝石CT胃癌浆膜浸润及淋巴结转移的诊断标准，并前瞻性地在大组胃癌病例中对该标准进行验证，以期获得胃癌术前CT分期的改良新标准，为今后指导临床选择适当的无创性胃癌分期方法及治疗效果评估提供一个更全面可靠的指标。

胃癌为常见的消化道肿瘤，MSCT目前已经确定为术前分期的主要手段，但是还存在较多不足。其中浆膜是否浸润一直是最大的难点之一。随着双能CT的出现，为这个领域带来了新的关注点，因此本研究优化并建立统一的胃癌双能CT扫描规范，同时构建胃癌能谱CT特征参数数据库并分析双能CT在胃癌术前分期中的应用价值。从另一个角度结合机器学习算法与双能CT特征参数，得出能辅助诊断胃癌浆膜浸润与否的客观标准。.第一部分比较胃癌患病灶在虚拟平扫和常规平扫图像，结果发现在常规和虚拟平扫图像中胃癌的最大厚度和肿大淋巴结的最大直径均无显著意义(P>0．05)。省去常规平扫所降低的放射剂量是21．40％±4．44％。虚拟平扫在胃癌术前分期扫描中具有代替常规平扫的潜在应用价值，并能因此降低辐射剂量。.第二部分收集经胃镜证实的128例胃癌患者，常规组采用常规扫描方式以及小剂量团注测试扫描方法进行术前扫描。两组结果分别与术后病理对照比较两组分期准确性。结果显示常规组T分期准确率为66.7%，N分期准确率为76.7%。TB组T分期准确率为63.3%，N分期准确率为70.0%，两组比较均无统计学差异。test bolus扫描方式在胃癌术前TNM分期中不仅具有较高准确率，而且可准确显示胃周血管信息。DSDECT具有放射低剂量优势，二者结合在胃癌术前分期中具有较大应用价值。.第三部分通过IWD-Citation-KNN算法提取胃癌图像示例特征，并将算法结果与其他基准算法相比较，结果发现基于加权距离的IWD-Citation-KNN算法准确率为76.92%，敏感性（TPR）为87.5%，特异性（1-FPR）为60.00%。同其他基准算法相比较，基于加权距离的IWD-Citation-KNN 算法在特异性和敏感性间取得了较好的折衷结果。回顾性分析行宝石能谱CT 双期GSI 增强检查的胃癌患者98 例。SVM-RFE 算法的准确率为87.5％-94.4％。SVM-RFE 的输出结果为门脉期脂肪（钙）、门脉期尿酸（钙）、动脉期钙（碘）、门脉期水（钙）、门脉期碘（水）。肿瘤大小和门脉期脂肪（钙）、门脉期尿酸（钙）、动脉期钙（碘）、门脉期水（钙）及门脉期碘（水）特征值可用于辅助判定胃癌是否浸润浆膜层。