Three-Dimensional Quantitative Analysis of Chronic Obstructive Pulmonary Lesions on CT Images

Abstract

In this paper, the three-dimensional segmentation of pulmonary honeycomb lesions and the function of image biomarkers are integrated into a quantitative analysis module of the lung, and integrated into the PACS image diagnosis workstation for the quantitative analysis of doctors. It is difficult to segment honeycomb lesions in lung CT images because of its diffuse and blurred edges, but it has good texture characteristics. Based on this feature, this paper converts honeycomb lesions into a texture classification problem. Firstly, the lung parenchyma is extracted by automatic threshold segmentation and region growing method, and then the lung parenchyma is divided into several small regions according to the texture by watershed method. Then, according to the texture characteristics of each small area, the trained support vector machine is used to classify. Finally, the correlation between gray level and spatial position of slice data is used to correct the classification results, so as to reduce false positive areas. In order to expand the study of imaging biomarkers of chronic lung diseases to more extensive major diseases, a sample database was established for a wide range of multiple lesions. Through feature extraction and feature analysis of multiple lesions in database, potential feature differences can be excavated, which lays a solid foundation for further study of image biomarkers of multiple lesions.