Automated Classification of Pulmonary Nodules for Lung Adenocarcinomas Risk Evaluation: An Effective CT Analysis by Clustering Density Distribution Algorithm

作者：Vanbang Le; Yang, Dawei; Zhu, Yu^*; Zheng, Bingbing; Bai, Chunxue; Quocthinh Nguyen; Shi, Hongcheng

来源：Journal of Medical Imaging and Health Informatics, 2017, 7(8): 1753-1758.

DOI：10.1166/jmihi.2017.2259

• Lung Nodule Classification
• Density Distribution Feature
• Risk Evaluation
• K-Means

To improve the lung nodule classification efficiency, we proposed an effective density distribution feature extraction algorithm and combined with pattern recognition models to evaluate the classification system. Our proposed feature extraction algorithm first randomly collected a large number of blocks from lung tumor images and determined the distance matrix by calculating the relationships among the image blocks. Then, we used the K-means clustering methods to classify the current image blocks and obtained 10 cluster centers. After that, we calculated the distribution density features by mapping CT value of nodule image pixels with the 10 cluster centers and extracted a 10-dimensional feature vector. Finally, the extracted feature vectors were divided into training and testing set to identify lung adenocarcinomas risk levels by Random Forest classification model. The AUC and the best accuracy achieved by the proposed method are 0.9144 +/- 0.0411 (p = 0.0002) and 89.20% respectively for ZSDB dataset. We also evaluated the classification framework in LIDC-IDRI dataset, the AUC is 0.8234 +/- 0.0703 (p = 0.0009) and the best accuracy is 82.92%. The proposed method outperforms the most recent techniques, and the experimental results show the great robustness of the proposed method for different lung CT image datasets.

• 出版日期2017-12
• 单位华东理工大学; 复旦大学