融合像素空间信息及加权模糊聚类的肺结节识别

Lung Nodule Recognition by Integrating Feature Weighted Fuzzy Clustering with Pixel Spatial Information

针对肺部计算机辅助诊断中孤立肺结节识别容易受噪声、气管、血管的干扰问题,提出一种融合空间信息及加权模糊聚类的肺结节识别算法.该方法利用融合像素空间信息及带特征权重的模糊C均值聚类算法实现感兴趣区域分割;利用特征选择算法计算感兴趣区域各特征权重,加权模糊C均值聚类算法分类感兴趣区域,识别肺结节.对比实验证明,该算法对感兴趣区域分割抗噪声性增强;感兴趣区域分类准确率提高;整体算法对肿瘤的检出率较高,漏诊率降低,为医生诊断早期肺癌病灶提供更加准确的客观依据.

In the computer-aided detection（CAD） for lung,the recognition of solitary pulmonary nodules may be interrupted by noise,trachea,bronchial or veins.A method is therefore proposed to recognize lung nodule by integrating the feature weighted fuzzy C-means clustering with pixel spatial information so as to segment the region of interest（ROI）.Every feature weight in ROI is calculated by the feature selection algorithm,and the weighted fuzzy C-means clustering algorithm is used to classify ROI,thus recognizing the lung nodules.Experimental results showed that the ROI segmentation algorithm is capable of denoising robustly and that the accuracy of ROI classification is improved.The integrated algorithm proposed has a high sensitivity to tumors with low undetected rate.It can provide helpful information to quickly identify suspicious focus in early stage of lung cancer.