基于射频信号的超声影像组学计算机辅助诊断肝纤维化进程的预测模型研究

Liver fibrosis and cirrhosis are common pathological results of chronic liver diseases. Liver fibrosis and cirrhosis in early stage are reversible histopathologically. It plays a key role in personalized treatment and response monitoring to assess fibrosis progression accurately and noninvasively in clinical setting. The standard pathological examination of hepatic tissue is an invasive procedure, what’s more, the common histological scoring system describes the morphological features based on architectural changes of hepatic lobules, rather than the amount of collagen which leads to liver fibrosis basically. In our preliminary research, we evaluated the feasibility and clinical value of elastography and contrast-enhanced ultrasound with quantitative analysis in the evaluation of liver fibrosis, however, it was far from sufficient to meet clinical needs in the accurate diagnosis of liver fibrosis progression. On the basis of recent research progress of computer aided diagnosis and pathology full quantitative examination of degree of fibrosis and the spatial structure of qFibrosis system, we plan to collect the original radio frequency signal of multimodal ultrasound images from hepatic fibrosis rat models and chronic hepatitis patients firstly, extract high-throughput image features from the multimode image information and establish the feature group. The quantitative analysis of qFibrosis based on precise measurements of liver collagen and fibrosis structure was set as the gold standard. Then, through deep learning, array signal analysis, sparse representation of signal processing method, and finally, a prediction model would be established as regards the matter of the correlation between radio frequency signal of radiomics and liver fibrosis progression with collagen fibrosis process and spatial structure. This will provide vital evidence for non-invasive accurate diagnosis of hepatic fibrosis progression and reduce clinical dependence on liver biopsy.

肝纤维化/肝硬化是慢性肝病的常见病理结局，是一个可逆的病理过程，准确而无创评估肝纤维化进程是个性化治疗及疗效监测的关键。肝组织活检创伤大，常用的病理诊断仅主观描述肝小叶的结构紊乱，而未测量真正反映纤维化进程的胶原含量及其空间构架，不能精确评估纤维化进展和逆转。课题组前期研究阐述了多种超声新技术评估肝纤维化的价值，但尚不能标准化应用和推广。结合近年来计算机辅助诊断和病理学全定量式精细检测胶原含量及空间结构学qFibrosis系统的研究进展，本项目拟通过获得大鼠肝纤维化模型和慢性肝病患者的标准化多模态超声原始射频信号，利用多模影像信息的互补性，建立超声影像组学高通量特征集；以全定量检测肝内胶原的qFibrosis为金标准，通过深度学习、阵列信号分析、稀疏表示等信号处理方法，建立超声影像组学高通量特征与胶原含量及空间构架关联的预测模型，为无创精准评估肝纤维化进程、减少对肝组织活检的需求提供依据。

肝纤维化是各种慢性肝病的共同病理过程，准确诊断和及时干预能有效逆转或延缓其发展成为肝硬化或肝癌，无创评估肝纤维化进程是个性化治疗和疗效监测的关键。肝脏超声射频信号蕴含了肝组织散射、衰减、声速和相位等原始声学回波信息，较超声图像能提供更丰富的组织信息。项目通过采集肝纤维化动物模型和慢性肝病患者的标准化多模态超声原始射频信号，利用多模影像信息的互补性，以肝纤维化病理形态学和空间结构学的qFibrosis全量化指标为金标准，从射频信号复杂且庞大的信息中挖掘反映肝纤维化相关的高通量特征，创建分类模型；利用分类模型的输出信息计算每个信号样本的预测值，并结合采样点的位置信息，将预测值叠加于超声图像上，直观而可视化地展现肝纤维化的程度及分布情况。.针对慢性乙肝患者，项目评估了迁移学习多模态影像组学在预测肝纤维化进程和肝脏储备功能方面的临床价值。项目收集肝脏多模态超声标准化影像，运用迁移学习的方法对肝脏灰阶模态、弹性模态、剪切波弹性测值及多模态联合影像进行人工智能特征提取及分析，提升纤维化分期诊断的训练速度和准确率。结果显示，在训练集中，灰阶和弹性联合模态迁移学习诊断S4、≥S3和≥S2的准确性分别为99.2%、99.2%和99.4%；训练集和验证集均显示迁移学习预测模型对肝纤维化分级的诊断价值高于非迁移学习模型和血清学指标。对于肝肿瘤患者，因不同程度肝纤维化时有效肝细胞数量减少，继而导致肝储备功能下降，发生术后并发症。项目通过与术后肝功能衰竭发生率对比研究显示，术后肝衰者术前肝脏弹性测值显著高于无并发症组；经单因素和多因素回归分析，筛选并创建术后肝衰的预测模型（PM）：PM=-15.451+0.095×LSM+11.7×INR+0.012×LN，预测价值AUC为0.82，敏感度和特异度分别为87%和70%；外部验证组的结果显示预测模型准确性高于MELD评分系统（0.72）及Child-Pugh分级（0.54）。多模态超声影像信息丰富，研究结果显示了其预测肝纤维化进程和肝癌术后肝脏储备功能的应用前景。