基于多参数磁共振深度表示学习及迁移学习的非酒精性脂肪肝病炎症、纤维化程度无创性评估研究

非酒精性脂肪肝病（NAFLD）诊疗过程存在两大关键问题：如何实现非酒精性脂肪性肝炎（NASH）的炎症及NAFLD肝纤维化的精准评估。目前，肝穿刺活检为金标准，但因有创性、可重复性差等缺点应用受限，亟需无创精准诊断方法。多种磁共振技术可反映NASH炎症、肝纤维化的病理变化，但我们前期研究发现其均有限度。近年来迅速发展的深度学习和迁移学习，为智能医疗发展提供了广泛的技术支持。我们假定基于多参数MRI智能诊断新技术能揭示反映NASH炎症、NAFLD纤维化程度的影像特性和诊断模式。本研究主要包括：1.基于深度学习的大鼠NAFLD多参数影像学特征提取和精准辅助诊断模型研究；2. 基于迁移学习建立适用于人类NAFLD病例的无创诊断模型研究；3. 人类NASH炎症分级、NAFLD肝纤维化分期无创性诊断模型验证及优选。本项目的实施将可能创建NAFLD关键问题无创诊断新方法，为提高NAFLD诊治提供基础。

There are two key issues in the diagnosis and treatment of non-alcoholic fatty liver disease (NAFLD): how to achieve the accurate assessment of the inflammation of non-alcoholic steatohepatitis (NASH) and the liver fibrosis of NAFLD. Currently, the liver biopsy is the gold standard, but due to the limitation of invasiveness and low reproducibility, there is a need for a non-invasive and accurate diagnosis method. Various magnetic resonance techniques can reflect the pathological changes of NASH inflammation and liver fibrosis, but we found that there are limitations in our previous studies. In recent years, the rapid development of deep learning and transfer learning has provided a wide range of technical support for the development of intelligent medical care. We hypothesis that a new technique based on multi-parameter MRI intelligent diagnosis can reveal the imaging characteristics and diagnostic patterns that reflect the degree of NASH inflammation and NAFLD fibrosis. This study mainly includes: 1. Study on NAFLD multi-parameter imaging feature extraction and precision aided diagnosis model of rat based on depth learning; 2. Establishment of non-invasive diagnostic model for human NAFLD cases based on transfer learning; 3. Human NASH inflammatory classification, NAFLD liver fibrosis stage non-invasive diagnosis model verification and optimization. The implementation of this project will create a new method for non-invasive diagnosis of the key problems of NAFLD and provide a basis for improving the diagnosis and treatment of NAFLD.