医学影像处理技术辅助小肠疾病诊断

The quantitative assessment of small bowel motility plays a pivotal role in the diagnosis of small intestinal disease. Two types of medical image processing methods are proposed to assess the small bowel motility. One type is the small bowel segments oriented method and the other is the small bowel area oriented. The small bowel segments oriented method includes AILBF algorithm and BLFM algorithm. The research about AILBF includes not only the automatic contour initialization but also an improvement method for the Level Set method with local binary fitting model. AILBF algorithm aims to boost the efficiency of traditional Level Set method, which is known as an accurate contour detection method. The research about AILBF includes orientation adaptive Fast Marching method and moving benchmark line, which aims at addressing the leakage problem owing to the presence of low contrast areas. BLFM algorithm aims to improve the accuracy of Fast Marching method, which is characterized as a fast propagation method. The small bowel area oriented method includes the DNMI algorithm that utilize the negative mutual information to detect the difference of small bowel areas between adjacent MR images, which break through the limit of traditional small bowel segments oriented method. Our research can boost the efficiency of diagnosis of vital diseases such as small intestinal tumor and Gastrointestinal bleeding, which is of great importance in the treatment of patients.

小肠运动功能分析结果是对胃、肠疾病进行诊断的重要依据。提出两类医学影像处理方法分析小肠运动功能：一类是面向微观小肠片段的分割方法，另一类是面向宏观小肠区域的相异度度量方法。第一类方法包括AILBF算法和BLFM算法。AILBF算法的研究包括：轮廓自动初始化与改进局部二值拟合模型，目的是提高精确分割图像Level Set方法的效率。BLFM算法的研究包括：边缘方向自适应Fast Marching方法与移动基准，以解决低对比度图像中边界溢出的问题，并且提高快速分割图像Fast Marching方法的准确率。第二类方法是DNMI算法，其研究负互信息度量相邻磁共振图像小肠区域相异性的问题,突破了传统面向小肠片段运动功能分析的局限。两类方法结合有助于胃、肠疾病，特别是危及生命的胃、肠出血的迅速诊断，对救治病人有重要意义。

评估小肠运动性在小肠疾病的诊断中十分重要。项目组完成了两类医学影像处理方法分析小肠运动功能：一类是面向微观小肠片段的评估方法，另一类是面向宏观小肠区域的相异度度量方法。. 面向微观小肠片段的评估方法，我们基于深度神经网络完成了一种半自动的方法FCN-Baseline和一种全自动的方法FCN-LSTM：半自动的方法FCN-Baseline摒弃了人工选择特征，能够自动提取特征；全自动方法FCN-LSTM摒弃了手动绘制基准线。两种方法都可以在小肠图像上自动标记剖面直径，测量直径长度变化，评估收缩频率，从而实现对小肠运动性的评估。. 面向宏观小肠区域的相异度度量方法面向小肠区域的算法可以度量指定小肠区域之间的差异，对其运动特性进行分析，突破了传统片段分析技术的局限性。. 以上两种评估方法得到的小肠运动频率在合理范围内，接近手工测量方法得到的结果。项目研究成果可以有效的辅助放射科医生做决策，应用于危及生命的小肠肿瘤、胃、肠出血以及Crohn病的迅速诊断，对救治病人有重要意义。