磁共振多参数成像分析判别模型对肝纤维化分期的实验研究

Medical Imaging in quantitative study of staging hepatic fibrosis remains challenging in medical research. Currently，Liver biopsy is being used as gold standard for diagnosis of hepatic fibrosis clinically at present. However, its invasiveness and limitations restrict the clinical application. MR-based multi-modality techniques reflect different features of liver fibrosis, including MR elastograghy, T1ρweighted imaging, susceptibility-weighted imaging, IVIM and DCE-MRI. Pair-wise similarity is calculated by extracting imaging characteristic parameters. A support vector machine is trained using the labeled samples with pathological results as gold standards. Then the features of a new sample are computed and embedded into the trained space. The classification result is compared with pathological examination and the performance of the proposed multi-variate discriminant model for recognizing fibrosis staging is evaluated. This study aims to establish a non-invasive and precise machine learning model to identify staging of liver fibrosis and to promote the application of imaging in staging identification. As a complementary method to liver biopsy, MR imaging provides a new way for non-invasive investigation on chronic hepatic diseases and assessment of treatment effects.

肝纤维化分期影像学定量研究是医学研究领域的一个难题。目前，肝穿刺活检是临床用于诊断肝纤维化的“金标准”，因其创伤性和自身的局限性限制了临床应用。本研究利用磁共振多参数成像（弹性成像（MRE）、T1ρ成像、磁敏感成像（SWI）、体素不相干运动成像（IVIM）及动态增强成像，从不同侧面反映肝纤维化的疾病进程。抽取各序列影像学特征参数,计算其相似度指标，基于病理分类，使用支持向量机建立分析模型。引入新个体，嵌入分析模型，验证MR多参数分析模型诊断肝纤维化分期的准确性。本研究旨在建立一种无创、精准的肝纤维化分期分析判决模型，提高肝纤维化的影像学分期，作为肝穿刺活检的一种补充方法，为无创伤性的慢性肝病纵向研究、疗效评估提供新方法和新思路。

背景：慢性肝病引起肝纤维化是一个世界性公共健康问题，最终导致肝硬化、门静脉高压、肝性脑病、肝肾功能衰竭、肝癌等严重并发症。肝纤维化的治疗和监测取决于纤维化分期，肝纤维化分期至关重要。.主要研究内容：基于MR的各种检查新技术利用肝纤维化的物理性质改变（机械性能、T1ρ延长、铁沉积、水分子扩散运动、微循环和肝细胞功能），采用磁共振弹性成像，磁共振T1ρ成像及磁敏感加权成像、扩散加权成像、灌注成像进行定量研究，通过各自特定的成像原理反映肝纤维化的病理改变 。利用肝纤维化动物模型进行采用MR多参数成像提取图像的定量参数作为特征值，建立数据训练判别模型，等综合诊断能提高肝纤维化早期诊断的准确性，对肝纤维化的筛查、诊疗、监测及药物临床试验等具有重要的科学意义和应用前景。.重要结果和关键数据：在训练组，在鉴别早期与正常对照组，建立的预测模型明显优于T1值和Ktran，与肝实质弹性硬度值、肝/肌肉比值相似；在鉴别早期与晚期肝纤维化，预测模型曲线下面积明显优于T1ρ (z = 3.97, P < 0.001), SIR (z = 3.05, P = 0.002), and Ktrans (z = 4.08, P < 0.001), 但与肝实质弹性硬度值无明显差异；测试组与训练组诊断准确度和截断值一致。.科学意义：磁敏感成像、动态增强扫描及弹性成像在肝纤维化分期起到重要诊断价值，多参数MRI预测模型的建立进一步增强诊断准确性。