流形方法对边缘病变肺野分割的研究

为实现对高密度病变影响边缘肺野的正确分割,提出了利用相邻肺野形状的相似流形来对其进行分割的方法。首先,对胸腔CT影像中肺野形状的相似流形的形成以及用主成分分析(principal component analysis,PCA)约减肺野来构造光滑流形进行了研究。其次,通过构造流形所表达的肺野关系,采用流形插值来重构被影响肺野的形状。采用基于Harris特征的变形配准方法来降低重构区域的误差。最后,用该重构肺野形状掩模来对原胸腔CT影像进行分割得到真实的肺野区域。实验结果表明,采用样条流形插值重构肺野形状来完成对高密度影响肺野的分割是一种可行有效的方法。

Research on segmenting lung fields with borders of high density pathology diseases using the manifold method

The research focuses on the acquiring of the lung fields whose margins are affected by the high density pathology in thoracic CT scans by using the similarity characters of the shapes of the adjacent lung fields. Firstly, to invest the similarity manifold of the shapes of the 2D lung fields in a lung and the method of constructing the manifold with the PCA dimension reduction for them; Secondly, based on the relationship among the lung fields represented by the points on the manifold, the affected lung fields can be reconstructed by the manifold interpolating. Thirdly, the reconstructed shape is registered with the affected one using the found Harris features to decrease the area error. Finally, segment the CT scans with the reconstructed shape. The experiment results show that the proposed method is an effective method, and the evaluating data (e.g. accurate, sensitive and specificity) demonstrates that it can acquire the good segmentation result for the affected lung fields.