基于放射影像组学和随机森林算法的肺结节良恶性分类

Classification of Benign and Malignant Pulmonary Nodules Based on Radiomics and Random Forests Algorithm

针对现有的肺结节良恶性分类算法存在分类准确率不高的问题,文中提出了一种基于放射影像组学和随机森林算法的肺结节良恶性分类算法.首先,提出了一种新的多尺度圆形滤波,用于对肺结节进行增强;其次,采用阈值法、形状指数和纹理特征自动获取种子点,并将种子点注入到随机游走算法中,以实现对肺结节的准确分割;然后,对分割的肺结节进行灰度、纹理、形状、小波和临床表征特征的提取;最后,采用随机森林构造肺结节良恶性的预测模型,并使用数据库LIDC对预测模型进行训练.实验结果表明,文中提出的算法对肺结节良恶性具有较高的分类性能,准确率、敏感性和特异性分别为94%、92%和94%.

To address the problem of low accuracy of the existing algorithms for classification of benign and malig- nant pulmonary nodules, a classification of benign and malignant pulmonary nodules based on radiomics and ran-dom frests algorithm is proposed in this paper. Firstly, a novel multi-scale dot enhancement filter is proposed for pulmonary nodule enhancement. Then, thresholding, shape index and texture featres are used to automatically acquire the seeds. The acquired seeds are injected into the random walker algorithm, thus accurate segmentation of pulmonary nodules is achieved. Secondly, the intensity, textron, shape, wavelet, and clinical featres are extracted from the segmented pulmonary nodules. Finally, random forest is employed to construct the predictive model for classifying benign and malignant pulmonary nodules. The LIDC database is used to train the predictive model. The experimental results demonstrate that the proposed algorithm has the high classification performance for classification of benign and malignant pulmonary nodules, and then accuracy, sensitivity and specificity are 94% , 92% and 94% , respectively.