基于多模态高分辨率超声和三维病理图像融合肝癌微血管侵犯精准诊断的方法研究

微血管侵犯是肝癌消融术后高局部肿瘤进展率的一个重要原因，基于影像诊断肝癌微血管侵犯是亟需解决的临床问题。目前研究多通过影像分析对肝癌微血管侵犯进行预测，缺乏定位定量的精准诊断。我们前期通过影像组学方法对灰阶超声图像进行分析，建立了肝癌微血管侵犯的预测模型，验证了超声图像分析评估微血管侵犯的可行性。在此基础上，本课题拟建立动物肝癌原位模型，采用多模态高分辨率超声对肝癌及癌旁组织进行三维成像，然后借助图像配准技术将其和大体病理切片检查获得的三维病理图像进行融合对位，再利用人工智能算法分析病理标记微血管侵犯区域的超声图像，对肝癌和癌旁组织进行亚像素级别的量化分析，分别构建肝癌和癌旁组织的影像组学标签，挖掘超声图像中高维定量的深度特征，最后通过小样本学习算法建立肝癌微血管侵犯可量化的精准诊断模型，为肝癌微血管侵犯个体化精准治疗奠定方法学基础，该项目具有重大的科学研究价值和广泛的临床应用前景。

Microvascular invasion(MVI) is an important factor of the high rate of local tumor progression after radiofrequency ablation of hepatocellular carcinoma (HCC). Imaging-based diagnosis of MVI is an urgent clinical problem to be solved. At present, most of the studies predict MVI through analysis on images, lacking precise diagnosis of MVI location and quantitative analysis. In previous research, we extracted the imaging features from gray-scale ultrasound images, established a predictive model of MVI and verified the feasibility of ultrasound radiomics to evaluate MVI. On this basis, we intend to establish an in-situ HCC model in rabbit, acuqiring the three-dimensional images of HCC and adjacent tissues by multimodal high resolution ultrasound imaging. Next, image registration technology was employed to fuse the three-dimensional ultrasound images with the three-dimensional pathology images obtained by pathological examination. After image fusion, radiomics algorithm were used to analyze the ultrasound images specifically in pathologically-confirmed microvascular invasion areas. Next, sub-pixel quantitative analysis of ultrasound images of HCC and adjacent tissues was used to construct radiomics scores and extract high-dimensional quantitative features respectively. Finally, we will establish a quantitative diagnosis model of MVI by few-shot Learning, which might provide a methodological foundation for individual precision therapy of MVI of HCC. This project is of great value and has broad clinical application prospect.